This compilation and all annotations are copyright © Jean Renard Ward, 1993, 2004, 2011, 2013, 2016.
I can be contacted at email@example.com.
History of Pen and Gesture Computing:
Annotated Bibliography in On-line Character Recognition, Pen Computing, Gesture User Interfaces and Tablet and Touch Computers
References from the approximate years 1981 to 1983.
Permission is granted to use this information in publication, including confidential reports, provided that accompanying text clearly makes reference to the URL for this page, along with the statement:
Source: Annotated Bibliography in On-line Character Recognition, Pen Computing, Gesture User Interfaces and Tablet and Touch Computers Copyright Jean Renard Ward
Most of these are in my personal collection, either as physical items or electronic files.
Copyright © Fri Jul 21 20:52:55 EDT 2017.
Mirrors of this page can be found at:
biblio80 (Prev page)
(Next page) biblio85
Product information announcement about Micropad handwriting-recognition terminal (see file) from Micropad, Inc.: Micropod is a local or remote input terminal which captures handwritten data at the time of writing, translating It into machine readable form.
Chain-code transformation, for matching similar curves.
Dictionary lookup for word-by-word handwriting recognition of Arabic.
Segmentation: some Arabic letters are discontinuous, so Arabic words are 1 to 7 strokes (not counting dots?).
Handwriting spelling dictionary look up by number of strokes, number of dots, number of intersections in Arabic.
Cited in FoleyJD82.
Input model: precursor to PHIGS and CGI, postcursor to SIGGRAPH CORE?
Critique of CORE graphics standard, need for composite devices. Compare with proposal by Pencept for handwriting recognition input to PHIGS standard. Model supports multiple (simulated) devices, two-handed input, contrasted with ping-pong input of one device at a time.
Refers to user-interface of simulating a virtual soft keyboard /function buttons on a tablet.
General user manual for Apollo workstation. Hierarchical file system, windows. No citation to Unix. Touchpad included on workstation keyboard.
Modem command set for Hayes-compatible Apple modem.
Operating system reference for Apple Lisa (predecessor to Apple MacIntosh): low-level programming, mouse, no tablet.
cited in RubinSM84: Ledeen recognizer?
Fourier coefficients of strokes as feature points.
For Roman alphabet, 29 single-stroke and 7 two-stroke character shapes.
For Hiragana alphabet, 10 single-stroke and 18 two-stroke character shapes, 13 3-stroke, 5 4-stroke.
Shows basic features of first Pencept product: single-stroke and multi-stroke forms, boxed input.
Refers to similarity/variability of loop and arc in script handwriting.
Script recognition using clock-wise vs counter-clockwise loops and arcs.
Refers to corner(cusp)/loop transition in script writing.
Cites Baum-Petree algorithm for decoding.
Font- and language-independent recognition by doing cryptanalysis on whatever categories and arbitrary recognition comes up with?
Realistic testing: artificial tasks vs natural tasks (for a priori grammar for speech).
Speech: acoustic/phonetic encoding vs communication theory model.
Video text editor to emulate different computer keyboard terminals, with separation of data and display modules: does not involve pointing/pen or character recognition input.
(Copy provided by Gary Odom).
DCR Dynamic Character Recognition devices allow several novel kinds of user interactions / interfaces. Says needs much higher resolution digitizer touch-pad than is available.
GTCO Digi Pad 5 Pressure Pen Tablet; comments on lack of smarts used in microprocessors for functionality of pointing devices: example is buffering of data in pointing device to avoid data over-run from host buffering-- does not comment on latency. Labanotation input for dance transcription (Benesh editor).
User interface usability testing of Xerox Star GUI, primarily on use of mouse buttons, 1/2/3-button mouse: separate button for drag/copy/select, multiple clicks (double click, etc.) Drawthrough gesture for selection: not quite press-and-hold.
Capacitive rotary touch input combined with light display: rotary control on panel without a mechanical opening for a rotor. Compare with virtual devices?
Test results on 12 of the 35 Devanagari alphabet handwritten characters (all work done by hand: did not have a real computer?).
Fuzzy set classifier for Devanagari (Indian) handwriting recognition.
Data collection: tested on the 70% "satisfactory" sample writing sheets: bias in sample? points out subjectiveness of training?
(pairwise comparison?) syntactic method only applied to handwritten characters which were confusible.
Low-pass digital filter for a tablet to pre-process handwritten character to eliminate wobble/digitization noise before recognition.
eye-tracking direction of gaze used to show more detail (semantic detail) where user is looking, less in peripheral vision/field-of-view. Corneal reflection, not eyeglass frame sensors.
Handwriting cursive script recognition using channel model of dictionary: allows for splitting, merging, and substitution (segment parsing errors). Cited in Bozinovic89 for context: probabilistic spelling corrector based on a distance metric for strings.
User guide for GRAFPAD, low-cost graphics tablet for ZX Spectrum early PCs. Resolution 256x320 (16x20 of boxes of 16x16 points), stylus with tip switch. Printed overlays for virtual keyboard (not on-screen), application menus. See also CPCWiki reference.
Review of commercial OCR products and applications (document sorting), including Kurzweil,
Report on handwriting styles taught and learned in the UK.
Points out that there is no single style of handwriting taught in copybooks in the UK, unlike some other countries.
Advisor, Tom Davis, comments on the diversity of writing styles in signature verification and the detection of forgeries.
adaptive/trainable recognition, suggests "closed loop" verification (feedback to user) is helpful.
Much DCR/CSR handwriting/gesture recognition research is commercial/proprietary, therefore not published. DCR (gesture/handwriting input) picking up as result of human factors in Man/Machine interface. Writer (author) independent recognition would be more desirable than present trained recognition. In development, necessary to hand-correct for real-world errors in input data collection.
NTIS citation index.
Quest Automation / Datapad product patent on handwriting recognition: Claims on alignment/registration of paper on digitizer tablet by marking on pre-printed alignment targets. Pressure-sensitive tablet with floating hand rest to keep hand from pressing on tablet.
See also Micropad.
Handwritten sketch/character input recognition: refers to IEEE handwriting data base collection 1.2.4: (24x36 binary OCR).
Like Greenberg77, throws out handwriting samples from IEEE Database 1.2.4 as "poorly written".
Context by matching grouped letters into words using Unix "spell". Separate "training dictionary" for each user: single-stroke characters, discrete characters. Segmentation by pen-lift only, therefore unistroke only: "t" "i" etc. must be made with one stroke.
Study of windowing (actually, panning) vs scrolling on displays: difference is direction: whether upward motion moves apparent window up, or apparent document up. Most subjects preferred windowing, but scroll users strongly preferred. Input was cursor arrow keys, not mouse or tablet: so not clear on effects of having a scroll bar or a flick/scroll gesture. Scrolling done with arrow keys.
trainable, short-hand single-stroke symbols for graphical shapes, graphics editing: Unistroke, terminated by pen lift at the end. Gestures are checkmark, downward stroke, and upwards stroke.
Gesture input: tap/circle gesture. Says velocity-sensitive gestures difficult. Digitizer stylus must be held vertical to get good results -- reference to stylus tilt problem, often miss-called parallax. Cited in Rhyne86.
Lexical (chunking and phrasing) aspects of compound gestures.
MENULAY/MAKEMENU system: User-interface editor, user can sketch user interface and then associate actions with user interface elements. Uses four-button puck on graphics tablet.
Video demonstration to accompany "Continuous Hand-Gesture Driven Output". Complexity of marking menus can be reduced by structuring the dialogs.
Video demonstration to accompany "Towards a Comprehensive User Interface Management Systems. System for defining user interfaces using sketching and marking menus.
Overview of IBM personal computer. IBM support both CP/M-86 and MSDOS. "A Closer Look at the IBM Personal Computer" (Gregg Williams). IBM Basic (GW/Basic).
Also articles on Cromix (Unix OS clone), Atari graphical smooth scrolling implementation. HIPAD tablet digitizer from Houston Instruments, with keyboard simulation.
Detailed hardware and software description of features of the IBM personal computer. MS/DOS 2.0 installable drivers, background tasking. Pencept PenPad text/graphics with handwriting recognition. IBM Kana/Japanese keyboard / IBM 5550 system keyboard. "Big Blue goes Japanese" Has pictures of the complexity of keyboards for Japanese and Chinese characters.
Article on Concurrent CP/M operating system.
Mobile computer with keyboard, small character display (8 lines), optional fully-attached printer -- computer with Wang portable computer? Intel 8088, auto power-off power management on 3 inch hard disk drive, small touch panel / touchpad above keyboard: cursor control in center, fixed function buttons at sides (e.g. View/Cancel/Open), pop-up menus. Touchpanel pointing input is relative, not absolute. Double/triple/quadruple taps for word/text selection.
Early Micropad-like British handwriting recognition product.
Force-feedback to user using pneumatic air sacks in glove to apply pressure for tactile feel.
Digitizer measuring X and Y sequentially in time. Mentions velocity correction for bowing of diagonal lines, corrections for non-orthogonality, course and fine measurements in two stages. X/Y Grid, puck with magnifying glass shown.
User interfaces discussed in terms of keyboard interactions. One section on Grafcon tablet: section on pointing/selection speed only compares joystick vs. mouse.
Early essay on user interface design for varied users: range to be dealt with is not just beginner/intermediate/experienced users, also range of learning styles (hands-on versus read the documentation, visual learners versus text learners).
User interfaces better if friendly: no unfriendly error messages. Also mentions discoverability in user interface design.
Combine character segmentation with classification in adaptive decision tree.
Optical resolution good enough for recognition may be too low to segment characters.
(like cursive/connected vs discrete characters segmentation): some touching characters in OCR can only be segmented by recognizing component characters.
Touchscreen controller for Casio wristwatch. Electronic file includes additional material on VDB-1000, Casio Databank with small keyboard on wristwatch. Very early Smart-Watch. User manual for Casio wristwatch: may be the model with touchscreen input and handwriting recognition, manual describes small keyboard input.
Chalkboard Power Pad: touch-sensitive membrane graphics tablet responding to more than one touch at a time, can image contact of entire hand. Resolution 120x120. Sold as educational toy. multi-touch capable?
Chalkboard Power Pad: pressure-sensitive membrane graphics tablet responding to more than one touch at a time. Resolution 120x120. Sold as educational toy.
Pressure-switch touch graphics tablet, 14,400 sensor points for 10x10 resolution on 12x12 inch surface. "Has no problem resolving simultaneous multiple inputs" (multi-touch), "software stretching" to interpolate (?) to higher resolution.
Date approximate: OCR product for reading serial numbers and other identifying numbers in manufacturing and shipping, such as product IDs on boxes.
Vendor survey in 1982 for digitizing tablets, video scanner digitizers.
Announcement of Buffered Micropad handwriting recognition input terminal, 512 characters, built-in one-line LED display.
Image Data Tablet System: Tablet and monitor with handprint (handwriting) recognition, mathematical calculation, graphics mode input, on-tablet keyboard icons. May be Pencept PenPad 200?
Early reference to digitizing tablets with IBM PC: rotational transformation, digitizer resolution much higher than display: typical resolution of tablet 2048 points (sic).
Nestor Graphics patent.
Nestor Graphics patent.
Curve follower to turn OCR data into DCR character data.
Nestor Graphics patent on pattern/handwriting recognition.
Cited in Arcelli85: about line-thinning and reconstruction.
Barry's group: functional attribute (cognitive) vs synthetic (generative).
Deal with embellishments separates from base pattern (in OCR).
Hew Crane: amendment to Crane79: patent 4,156,911 (?).
Human-reading for signature verification.
Writing pen with obstructed view: accelerometer (no touch screen). Mentions three-axis force tablet.
History of VM/370 virtual machine operating system, based on CP/CMS. Files relayed over virtual communications links between machines. All system commands implemented as separate files (executable files).
Capacitive key matrix keyboard, references touch-sensing circuit in IBM journal.
Auto-completion of typed text: if what is typed so far matches text in a list/dictionary, monitor shows rest of that text without advancing cursor, user can type over or hit a function key (right arrow) to accept. Compare with auto-completion in MAC LISP, word prediction in accessibility on-screen keyboard software?
Overview of forensic analysis of handwriting an signature for authentication. No two signature examples by same person identical, examiner must understand idiosyncracies. Three styles of handwriting: natural when capturing information, formal, and unnatural when intentionally modified. Affect pressure, writing speed, size. Document examination includes infra-red and ultraviolet examination of ink and paper.
Use of timestamps in key-exchange algorithms and signatures to prevent re-plays. (Compare with use of nonce).
Textbook on cryptography and security -- security primarily access controls (Bell-La Padula, etc.) Certificate and public/private key PKI distribution. Hamming encoding.
Frequency-domain analysis of handwriting movements in X and Y: Lissajous analysis.
Best description of AEG's segmentation/parsing algorithm for handwritten characters.
Blithely claims that rasterization of dynamic on-line character data would reduce this with OCR to the same problem (but the OCR segmentation problem is harder!).
Scan in documents with OCR, use on-line handwriting recognition user-interface to edit them.
Many limiting characteristics of digitizers (width of pencil line, etc.). List of what can be spec'ed for a digitizer performance (but not the trade-offs).
Differential linearity error is a digitizer's worst enemy; Clumping and stretching of digitizer co-ordinates for X-vs-X non-linearity; Digitizer accuracy vs digitizer stability; General review of digitizer characteristics and proper specifications.
Capacitive sensor for measuring pen/stylus acceleration for signature verification.
Cites gaps in human factors/engineering and cognitive psychology literature re text editing user-interface; human factors of using digitizing tablets for text editing.
Cites "taste for federal funding" as reason some human factors studies have skipped various interests: research selection bias.
Users frequently will not use mouse in text editing: (we say "3-handed monkey" effect): three-handed monkey.
Tablet: pointing with light pen fatigueing, and not accurate enough.
Digitizer tablet simulation of virtual input devices: turntable, multiple number wheels, three-axis trackball.
Dual-function pushbutton switch: one pair of layers is simple on/off to turn on power of (MOS) device, second pair of layers is force/pressure-sensitive transducer (FSR material).
Force/pressure-sensor using thin film of material with microscopic particles of conducting material (force-sensitive resistor). Does not depend on resilience/elastomeric/rubbery of material for a restoring force: uses contact resistance.
Claims better test results for machine recognition than for human reader (!).
Discusses pairwise discriminators to reduce memory requirements for OCR of typed text: Similar to Pencept?
Experimental system for creating structured documents (contents organized as directed graphs), with hypertext links. Desired platform is slate-format tablet computer with touchscreen, color display, (animated/active) pictures and text. Actions in a document are part of the document (compare: Flash, HTML).
Speculative article on use of tablets / flat-screen displays as electronic newspapers, continuous updating. Refers to tactile controls (touchscreen). Compare with Boston Globe / NYTimes electronic distribution on tablets 2010.
Not a whiteboard system: video conferencing arrangement with each user facing one monitor and one video camera per other user.
96% correct on handprinted OCR recognition: skeletonizing, "critical point" strokes with initial and final slope features for 18 types of stroke segments.
Admits to and spells out several weaknesses: 4 vs 9, patching OCR skeletonization breaks.
Handwritten OCR recognition: simple template matching works well if feature extraction is good.
Critical points for features and segmentation in OCR handwriting: points of high curvature, then Freeman chain codes.
Refers to Knoll Database of hand-printed numerals, 21x25 binary grid, IEEE Pattern recognition data base 1.2.2).
Precursor to FoleyJD84.
Three-point calibration, general graphics reference.
Comparison of tablet and mouse "locator" devices, absolute position versus relative motion. Section 5.3 "Technique for simulating logical devices": "locator device simulation" / virtual devices, simulated keyboard, character recognizers. Says Ledeen recognizer simplified version of Teitelman 1964, described in Newman 199.
from Tappert's bibliography.
NTIS index: neural network for handwritten numerals recognition, using a scanner: learning-with-a-teacher better than learning-without-a-teacher (training set makes a difference, but what?).
Very little treatment of UI issues or pointing devices: primarily deals with keyboard commands and display of monofont text.
GTCO Electronic grid digitizer, serial interface: pressure-sensitive pen/stylus, 16-button puck/cursor.
Not from Blesser group.
Gives many reasons for using forms as fundamental office automation model. (taken from Hekmatpour86). Use of forms eases transition from manual to office automation.
Cited in Suen86.
Suen86 cites this on what matrix resolution needed for OCR on hand-print.
Stem processing for data reduction in dictionary storage file storing word list file with prefix and suffix truncated so that only unique root element remains.
Cited in Litvin89.
Gesture recognition for a text editor? Handwritten text input?
Gesture-based user interface with handwriting recognition for editing text with electronic ink: first reference for prior art?
Early reference to visual parallax on electronic ink with integrated tablet/display.
Simulation of "perfect" speech recognition, showing that there is indeed a user-interface problem in addition to recognition problems.
Chinese recognition by recognizing component strokes/radicals.
Refers to "spelling" of Chinese/Kanji: order strokes/radicals are written in.
States that strokes are classified into 42 categories for segmentation, and that Chinese/Kanji characters are made up of only 72 basic symbol elements/alphabet.
Cites problem with sonic/acoustic digitizers: sensing point offset from writing point.
If Kanji/Chinese character is not recognized, user can add it to the prototype set for recognition on the fly.
Digitizer specifies +-.015" on 0.001 resolution, but also +-.01 accuracy.
Features for Chinese character recognition are Walsh coefficients, projected onto X and Y axes: character alignment (normalization), 99.5% accuracy on 3000 characters.
OCR on Chinese/Kanji characters.
Using a binary decision tree for pattern recognition of a large number of classes (Chinese characters OCR).
Performance: noisy characters only: clean characters do not occur in real life.
Transparent switch matrix for touchscreen, conformable to a display: grid of conductors separated by spacer dots, pressing on top flexible sheet/membrane closes switch of two conductors.
File contains reference for HP-150 touchscreen computer: MS/DOS personal computer (8088) with infra-red touchscreen on 9-inch video display. Also known as Touchscreen MAX. No OS support, applications had to be written specifically for touchscreen input.
Cited in Bozinvic89.
Bozinovic cites for human recognition by outline of word (as shown in WrightG52).
OCR for Chinese using features of stroke structure, complexity, direction, connective relation, and relative location for 95.4% of block-style handprinted data.
from Tappert's bibliography -- Chinese recognition.
Handprinted recognition: feature is break into piecewise segments, compare with decision tree: 32 handwriting terminals on one computer (cluster).
Overview/survey of tactile sensing for robotics: force-sensitive surfaces capable of reporting graded signals and parallel patterns of touching, contrast with simple touch as contact sensing only at one or more points.
Cited in FoleyJD84.
Early (DOS-age) touchscreen computer product.
High-resolution touch sensor, several hundred sensing points per square inch. Uses grid/matrix of anisotropic elastomeric material: crosspoint problem addressed by grounding all other lines than the ones being sensed, so that no current flows. Shear forces determined by force needed to move the probe across whatever it is touching.
Imaging touchpad sensor for robotics using small pressure/force sensors.
(see also Hollerbach78).
Unclear fancier curve fitting to handwriting models gives any more insight.
Maximum writing speed/velocity is 25 mm/sec.
Acceleration peaks in handwriting not just a tablet/pen artifact.
Variation in slant in handwriting (vertical) is about 10 degrees.
Cusp/loop: substitution of clockwise vs counter-clockwise motion in handwriting.
Variability of corner shapes for single writer.
Says reduction in concentration on handwriting needed to think and write at same time.
Multics operating system programming tutorial: special section on dynamic linking. All files are memory-mapped into address space, no separate I/O. Mentions direct file types stream/sequential: stream is ascii file? APIs for access control ACLs.
Multics operating system CLI command-line commands: set_acl; copy command preserves ACLs access control lists; send_mail over ArpaNet.
Multics file transfer facility: describes I/O Daemon process. On target system for file transfer, ring access must be at higher numbered access level (more restrictive), I/O daemon must have ACL access rights, user on target/foreign system must grant access. Read/write access can be restricted by setting zero length for segment. AIM (capability) permission must be less or equal.
Cited in Lipscomb91.
Cited in Faggin89.
from Tappert's bibliography.
User interface (fill in charts) for handwriting recognition input of engineering drawings (mechanical drafting, maps, NC control drawings).
Features are passage through 3x3 template grid, then chord lengths and directions (cusp, stroke, rotation, straight): cusps turn into small loops, vice versa is a source of error.
Signature recognition/verification using first twelve initial segments, segmenting by zero velocity in Y.
Notes that a frequency cut-off of 20Hz is o.k. for signature signal.
NTIS abstract: fix skeletonization/line-thinning breakdown at intersections by simulating writing motion in OCR.
Optical scanner which read digital co-ordinates from a map, using co-ordinates in magnetic ink, but also describes optical scanning.
from Tappert's bibliography.
Viterbi and binary n-gram for context in optical character recognition: most efficient implementation: Viterbi algorithm is based on probabilities of confusion of pairs of characters.
NTIS abstract: looks very similar to other 1982 paper by same authors.
Letter substitution errors on text recognition corrected by various means of context: combining bottom-up and top-down (syntactic and semantic) context works better than separately.
Context for spelling correction: goodness measure extended to probability that it is a corrupted form of another letter.
Context for spelling correction: letter probabilities: digrams/pairs and trigrams (bottom/up context) (may involve a false assumption about input language).
Context for spelling correction: dictionary lookup (lexicon for top/down context).
Many sources for spelling errors: typographical in original text, keying/writing errors, character recognition error.
Numonics article on digitizers and stepper-motor plotters.
Digitizers: no point being more accurate than the application needs. accuracy may be affected by pen angle/tilt. two-phase coarse/fine position determination using two characteristics (interpolation?).
Copy on file does not give name of author: since it mentions "elastic matching", most likely it is C. A. Tappert.
Use offset of character from its center of gravity as a substitute for the writing baseline.
Add horizontal positions of strokes to each other as an additional factor in recognition: computer with Functional Attributes of Shillman/Blesser (example is "A" and three-stroke "asterisk").
Also filed under Tappert and under Karnaugh.
For touchscreen over LCD, do not need separate touchscreen layer: by sensing capacitance to ground of LCD electrodes between display cycles. Cited in More93.
Product brief on AutoCad: light-pen, or Sun-Flex touch pen (touchscreen digitizer).
Discrimination of similar characters: boundary recognition. Uses different methods for characters of different number of strokes.
Early British handwriting recognition product: single line display, paper forms (shop invoices) fit into device, shown at Comdex 1982.
Note: Inforite used as product name by other companies.
Cites problem of needing very large training samples to get all variations.
Claims "only samples of low quality can improve dictionary" (training) (vs. ambiguous?).
Claims "recognition rate tells you nothing about how performance goes down with quality of characters".
Shows artificial forms used in Japanese JIS hand-print standard.
Not enough to read good characters, but how well does it do on bad ones?
Claims 98.5% recognition rate.
Use artificial variability instead of real handwritten data (!).
Goodale83. Application of fuzzy set theory to visual recognition / OCR: imperfect lighting, other imprecise measurements. Variation in connected regions, thinning / edge detection.
Human recognition features: visual groups of strokes in Chinese/Japanese characters consist of sub-groups of strokes, usually about four: visual group count corresponds to letter count in Western languages.
Describes finer control and finer parsing of access control in Multics with AIM Access Isolation Mechanism (nondiscretionary access control) in stricter set of rules for matches between segment and user attributes. AIM is to prevent user from releasing data to which user may have access via ACL: User's access is the more restricted of the two: clearance (of user) vs. classification level (of data).
Cited in Meisel85.
Describes GMU's Information Display Systems project (see Foley). Critique of CORE (and GKS) input device model, lack of extensibility.
Cited by Marlin Eller, Microsoft Pen Computing group. 2006 available at www.prioartdatabase.com / http://ip.com.
Interactive user-interface for handwriting recognition, sketching and sketching, using angle variation and stroke type (straight line, angled line, ellipse, circle, arc) for features.
Contains user interface for text entry to pretty up character spacing and alignment, fixing sketches (without recognition).
See also Tappert papers.
Correct cursive script for baseline drift.
Very large catalog of reversible and ambiguous text to show variability of human recognition and perception. Has informal but very interesting bibliography on variability of human recognition.
Optical mouse of surface having passive, position-related marks in a pattern of two colors / Sekendur? (Shading optical tablet).
Optical mouse using checkerboard square pattern on the mouse-pad / tablet: Sekendur?
For brush-type user-interface in graphics drawing.
Voice/speech/tablet/mouse/joystick input overview, 1983.
Fortran compilers beat the heck out of "C" compilers for code efficiency.
Distance metric for differences in linear strings: Levenshtein distance: mutations in chromosomes, UNIX "diff" files, string matching, minimal mutation distance etc.
Clustering /boundary definition using relative difference (dyadic), not absolute position (monadic).
Levenshtein distance easier to use than probabilistic/statistical estimate of how much change would be required.
Common sense: if your pattern recognition algorithm works better, it is better.
Describes boundary comparison using monadic variables/features (absolute value) vs dyadic (relative comparison only).
Different approaches for dyadic comparison: common sense, adapt monadic variables, and Levenshtein relative distance.
Recognize object outlines by Fourier analysis of 2D chain-code outline, lowest mode of fit is an ellipse. Mentions quantization / pixelation of images, show effects.
Pencept pendpad terminal product, handwriting recognition on tablet.
Mentions reduction of "dots" as a stroke type.
Makes disparaging remarks about "boxed" input for discrete recognition.
Discusses delayed strokes, "t"-crossings for discrete recognition.
Japanese patent showing segmentation, handwriting, sketch input, drawing extents.
Cited in Sklarew patents.
Cited in Kerrick88, Li89.
Quest Automation MicroPad, handwriting recognition product, small stand-alone terminal: pressure-sensitive tablet using flexible membrane under tension, resistive film.
Reference manual for Cedar object-based language, system, and GUI, successor to Smalltalk. Cedar kernel language is low-level subset, with direct access to primitive or OS types not permitted in Cedar. Strong type checking. No description of Cedar GUI.
Cited in Rhyne86.
Rhyne86 cites this that for verbal command names, very poor agreement on informal name subjects give for text editing commands (mnemonicity for gestures?).
"naturalness" in user-interface not helpful: for example, better if different names for options with similar semantics, but different syntax.
Human factors / user-interface: different methods of eliciting preferred command names get different results for same subject -- in particular, naive users make poor choices for command names.
Cited in Kankaanpaa87.
gesture/handwriting user interface on digitizing tablet (using puck/stylus). Text editing gestures for add (caret), delete (underline), replace (inverted pigtail), paragraph, change font, etc. Trash can, filer, other icons (compare with Wang Freestyle?).
Korean recognition: six types of Korean characters (first consonants, second consonants, vertical vowels, horizontal vowels) using syntactic recognition and 8-direction chain codes.
substrings of chain-codes.
Chain code string comparison in OCR for handwriting recognition?
Cited in Litvin82b.
See also Recognitive Sciences and Skylight Software (Yuri's consulting business names). Mentions retrace removal (page 7) User interface: points out difference between errors understandable to user, and errors not ("qualitative errors").
Cited in Goodale83.
Uses LED lights and digital cameras as three-dimensional 3D digitizer sensor to recognize hand gestures of ASL American Sign Language.
Assistive technology for the blind: compare with tactile/haptic Braille displays. Analysis of spatial bandwidth (resolution) limitations of human touch for tactile input. Cites to Optacon, TVSS vibrotactile display for the blind.
Reduce handwritten characters to a polygon / Freeman chain codes, then compare syntactic feature vector, then statistics on segment lengths.
Serif/hook removal, preprocessing/smoothing, retrace collapsing on on-line character recognition.
Handwritten samples "only" constrained to one of 69 writing styles/shapes: 99% and 97%: 20Kbytes memory, 500 Ms on 8086.
Asserts for on-line recognition that learning new shapes (adapting to different user's writing?) is more important than accuracy / recognition rate.
Resistive sheet tablet, spacer dots (compare Elographics). Hand rejection / resting hand does not press hard enough (compared to stylus point) to make contact between the sheets. (palm rejection).
Touchscreen over CRT display using total internal reflection of light generated by CRT, reflected where finger is in contact with screen. Like a lightpen, detection of position by knowing timing of CRT raster position, therefor unaffected by CRT raster drift.
Refers to problem of encountering a shape system not trained to for adaptive recognition.
OCR of handwriting recognition using polygonal approximation (chain codes), fuzzy labelling: thinning, tail-removal, fuzzy sets.
Cited in IchikawaS84. Human recognition of figures by "wire-frame" model.
Charge-ratio electrostatic configuration for particle detection: resolution not limited by dimensions of wedge anodes and different-width strip anodes. Different geometries: dual wedge anodes between strip anodes, wedges in opposite directions of taper; radial configuration; zig-zag pattern with no vias / through holes in substrate.
Survey of problems and technologies for processing Japanese Characters: standardized keyboards, speech recognition, handwriting recognition.
Humans recognize (recall? identify?) shapes better if the are recognizable as a familiar object.
Chain codes: something similar to BLRTs for describing images.
General paper on information dependencies and context.
Give a beep periodically (one second intervals) while a device is operating, or give an audible and tactile feedback each time a key is pressed / device is operated. For calculators, beeps on the keys?
What is user-friendly? friendly to a beginner may be bad for expert.
Multi-font OCR off-line recognition.
Early multi-touch input device, optical detection of fingers on frosted glass, but no rear-projection (separate display). Also recognize shapes and objects on surface. Cited by Buxton as historical note.
multi-touch/multi-hand input tablet (camera underneath, optical) with projection of keyboard or template images on top. Reconfigurable keyboard/touch service. Compare with Kaplow (reconfigurable keyboard) and Jeff Han (FTIR multi-touch) and Wellner DigitalDesk.
Text editing: primarily keyboard-based editing systems, long treatment of Xerox STAR with mouse pop-up menus, but no use of gesture (other than drag?), digitizer, stylus, or handwriting editing.
Text editing: primarily keyboard-based editing systems, long treatment of Xerox STAR with mouse pop-up menus, but no use of gesture (other than drag?), digitizer, stylus, or handwriting editing.
handwriting-terminal using a digitizer with two conductive sheets, sheets covered by a hand-rest; single-character discrete character recognition.
also showed a GUI application picking from a diagram of replacements parts, plus character recognition.
Press release included, data-entry use in New Scotland Yard: "gets write to the point"; Joe Crivello, National Sales Manager, Illinois.
Musical touch keyboard that detects X and Y position of fingers, key velocity (via position sensor), and force/pressure capacitive sensing. Z/Force detected with conductive rubber with greater capacitance as it is conformed by force to a curved bar capacitor plate, four-wire resistive sheet touchpad for X and Y.
Grammar and semantics for user-interfaces (commands). Example input devices are keyboard and display.
Generative variability in handwriting.
Indexing document automatically with OCR recognition by substituting likely-confused characters to same pseudo-character, similar to the Soundex system by Davidson for similar-sounding English names. Electronic ink / sketch indexing?
Claims 97.9% recognition rate on 120 samples (small sample size).
Refers to segmentation errors in recognizing two-dimensional flowchart symbols.
Sketch recognition - flowcharts.
Estimate of hardware requirements (memory, processor, display) for an electronic book including graphics, with search and indexing functions. Reference works (PDR, etc.), educational textbooks, coupled with speech synthesizers for the blind. Exemplary UI would be keyboard/buttons.
Cited in Kruskal83: Dynamic time warping for recognition.
NEC handwriting terminal product for personal PCs, low-cost.
NEC 2100 kanji and hiragana symbols ... product.
NEC handwriting terminal product, high-end version of personal PC product.
NTT 1900 kanji and hiragana symbols product "Aesop" on-line handwriting recognition, Nippon Telephone and Telegraph.
See also list in CIC folder on NTT.
(Date may be wrong) sketch/scribble/gesture input and editing system. Mechanical digitizer(!).
See also Suenaga80, same paper.
Handwriting mark recognition, OFF-LINE scanned OCR user interface for graphics editing.
Mark-up OFF-LINE scanned handwriting recognition user interface for on-line changes to scanned line drawings.
Has many Japanese citations for OCR of handwritten drawing and character recognition.
Testing: substitution error rates for OCR in practice two to three orders of magnitude lower than reported in academic literature.
Optical digitizer resolution of 0.004" sufficient for typed text.
Human adaptation reasons for some handwriting recognition systems' success.
Cursive writing not as useful as speech, or discrete writing.
Most optical scanners for OCR barely have resolution adequate for recognizing ideal characters, much less real ones.
Optical scanning digitizer characteristics: geometric, photo-metric, control.
Optical scanning digitizer characteristics: cite for tablet digitizer as comparison.
Testing: cites work by Chow on statistical relation of substitution vs reject error rate.
Kahan87 cites this as saying Duda72 binary Bayesian statistical classifier is widely used in OCR.
Optical digitizer characteristics: no vendor willing to be pinned down on performance.
Optical scanning digitizer characteristics: geometric linearity.
Optical scanning digitizer characteristics: stability/repeatability.
Optical scanning digitizer characteristics: cite for tablet digitizer as comparison.
Tongue-in-check list of how researchers fudge pattern recognition statistics/results to bias them in their favor.
Handwritten numerals and katakana recognition, coping with facsimile distortions and variability.
Hand-held scanner for character recognition.
Simulate real machines / controls for virtual devices using real-time computer graphics display and a touch screen digitizer: refers to switches and controls appearing on display in response to input from other switches or controls, such as a calculator changing modes and keys. Virtual keyboard, virtual mouse. Compare with Kaplow?
Generally credit with coining term RPC Remote Procedure Call. RPCs integrated into computer, or preprocessor for source file. Marshalling, exceptions, stub functions. Cites to other references for PKI authentication, authorization mechanisms. See also RFC 707, 1976.
User-interface application: special symbols and markings with handwriting recognition for application involving anesthesia record keeping.
User-interface: shows forms with combinations of writing, handwriting recognition, drawing, check-off menu areas, etc.
Matrix of capacitors with elastic/elastomeric dielectrics to measure force distribution. Multi-touch touchpad?
Change in writing styles as kids grow up.
Vendor of electromagnetic digitizer tablets: small vendor, mostly custom products.
Folder of OCR products circa 1981: Burroughs 1200 and Burroughs 1205 Series, AEG Formularleser PFL6160 POLYFONT, ECRM Concept 1 Pagereader; Burroughs PS100; DEST Corporation WorkLess Station; Datapro Research "All About Optical Readers" 1978; Datacopy CIR Software Character Image Recognition; Kurzweil KDEM 1200 Intelligent Character Recognition; Compuscan Alphaword Series 80.
Japanese handwriting recognition / text editing, boxed (large format) input on tablet integrated into desk surface.
NTT character recognition patent: Use of "feature points" in character recognition vs octants, etc.
NTT character recognition patent.
Patent on stroke-order independent recognition for Kanji/Chinese.
Feature points for Kanji/Kana are lengths of segments.
See 1980 paper of same title in Japanese.
Better than 99.8% correct recognition on Chinese/Kanji, hiragana, katakana, and alphanumerics/Romanji.
Three to six feature points on each stroke for on-line handwriting recognition: points just resampled to a minimum distance? end-points of (Chinese) straight strokes only?
Shape matching for handwriting recognition, followed by specific algorithms for ambiguous cases. Uses distance measurements and not stroke-order information.
Japanese handwriting recognition: features are inter-stroke distance pairs: claims 99.5% accuracy on 2057 Kanji.
Refers to AESOP user-interface for handwriting text/script editing. Measures accumulated distance along each stroke of an input with the strokes of an ideal sample -- compare with Tappert elastic matching?
Cited in Leung87.
Cellular automaton for character recognition (?).
Character and word segmentation in OCR documents using overlapping rectangles.
NTIS abstract of Japanese publication: Katakana Japanese handwriting recognition.
Cited in Arcelli85.
Arcelli85 cites as general reference on line thinning, and its applications.
Cited in Arcelli85.
Arcelli85 cites chapter 9 as general reference on line thinning, and its applications.
Cited in Lam86.
3-point calibration on digitizer tablets (not four points).
handwriting-input terminal product: VT-100 terminal emulation. Forms data entry, paper form placed on tablet, forms definition language. Eliminates re-typing/re-keying of data. Optical Character Recognition / OCR vs. Dynamic Character Recognition / DCR (on-line character recognition).
handwriting-input terminal product: DCR dynamic character recognition (a.k.a. on-line handwriting recognition). Mentions boxed input for stroke parsing/segmentation. Character Coordinate mode sends position information with each recognized character. Also Keyboard Compatible Mode sending backspace/delete.
handwriting-input terminal product: DCR dynamic character recognition (a.k.a. on-line handwriting recognition). Mentions boxed input for stroke parsing/segmentation. Tablet 11" high by 15" wide: standard paper with printed form on left, control area on right.
handwriting-input terminal product.
Handwriting terminal with digitizing tablet and handwriting recognition.
Handwriting recognition terminal from Pencept: VT-100 terminal emulation, with internal language for forms management. P. 3-7 keyboard emulation mode for input. p. 3-36 virtual tablets, virtual areas with independent programming, coordinate origin/rotation/scaling.
handwriting-input terminal product: Software control at the stroke of a pen. Only one hand needed for input, not two hands plus mouse (three-handed monkey). Penpatch software handwriting input in parallel with keyboard.
Resistive-film touchpanel, finger may either be source or sink of signal. Signal flowing through user's body to ground, or signals generated by ambient noise (e.g. 60-Hz hum). Pressure-sensitive touch panel, transparent touch panel using ITO or transparent polyester film. Triangular resistive sheet touchpanel with three sides.
Touch-screen digitizer that senses intensity of touch force or pressure.
Resistive-film touchpanel, finger may either be source or sink of signal. Signal flowing through user's body to ground, or signals generated by ambient noise (e.g. 60-Hz hum). Pressure-sensitive touch panel, transparent touch panel using ITO or transparent polyester film. Triangular resistive sheet touchpanel with three sides. Displays icons for touch locations (indicia).
Edge connectors to linearized distortions at edges of resistive film digitizer / touch-screen.
Microsoft mouse: article compares mouse to light pen and to joystick. Piano demonstration program.
Review of CERN user-interface (man-machine interface) developments, including capacitive touchscreen. Capacitance measurement by charge-time detector (did not work well), and by voltage divider (worked better). 16-button touchscreen with hierarchical menus. Cites to touchscreens based on cross wires, infrared light, acoustic waves, resistive, capacitive.
Tell users to change their handwriting: changing slant works, but if say to make taller, users also make wider, and vice versa. (Handwriting variability).
Uses tilt angle in signature verification.
Xerox 5700 printer with touch-screen interface showing simulated buttons and switches, touch-to-begin.
Cited in Boie 1985. Matrix of elastomeric resistive silicone rubber bars with crosspoints grid for measuring force at crosspoints. No anti-aliasing?
Micropad dynamic on-line character recognition product: Resistive film digitizer?
Quest Automation: Micropad signature verification product.
NTIS abstract: OCR on handwriting recognition of numerals: features are true endpoints, true group points, and true cross points.
Pencept Penpad M200 handwriting terminal, circa 1983, showing system box, digitizer, etc.
Micropad-like handwriting recognition product from England, using two resistive sheets and an air separator on the tablet.
Neural (net) model for supervised learning, later applied to handwriting recognition.
Patent on original RSA public-key encryption/decryption algorithm.
Electromagnetic digitizer with stylus loop as the transmitter, grid as the receiver.
Acoustic digitizer design: two ultrasonic sound sources on the stylus, permitting tilt / parallax correction. Digitizer accuracy (for comparison): pencil lines are 0.005", Ink is 0.01" wide, visual acuity is 0.0005" max.
Brief overview of interactive graphics CAD/CAM system for circuit design, user interface "portable" to different pointing devices e.g. mouse, tablet.
Line-segment approximation shows average quantization error a function of grid size, not angle resolution: gives formula for grid size vs matching accuracy.
Dynamic programming, time-skipping, dynamic time warping, for speech (author has also published on handwriting character recognition).
Same Ralph Sklarew of Grid computer: structures in space the form of a ring around a planet or other orbiting body: cites Larry Niven "Ringworld" as prior art.
Analysis (no user studies) of variability expected in Japanese handwriting recognition, based on what had been found in ASCII alphabetic handwriting and stylus dynamics.
Image processing: group lines/regions in scanned raster image / bitmap into quadtrees, where tree nodes are maximal blocks (e.g. minimal fragmentation of regions).
Single-touch pressure sensitive (contact area) capacitive touchscreen/tablet for musical instrument input. Matrix of X and Y conductors, measure capacitance change on conductors individually. Adjust two values at once using X/Y input of tablet, into SSSP digital synthesizer music program. Third value from "pressure" as amount of capacitance change with surface area of contact. Multi-touch input would be better. Cited in Lee 85.
Cleaning up graphical drawings using low-level symbol recognition: (see sketch editing, prettifying).
30 Hz sampling of handwriting on a tablet, DPCM/PCM coders to send handwriting at 200-300 bits/second, 100-500 bit buffer: tablet performance/handwriting signal extraction/frequency response.
NTIS: Supervised learning on statistical recognizer.
Put-That-There project: hand-pointing gestures and voice command input. 3D gestures using Polhemus digitizer.
Comparison of touchscreen digitizer / touch entry devices (TED) technologies: touch wire, cross wire, capacitive, conductive film (resistive), acoustic ranging, infrared (optical beam) and pressure sensitive. Cites to E. A. Johnson 1960's for invention of touch entry device.
Handwriting input overview paper by Bob Shillman.
Gives more recent address for Micropad Limited, Image Data Products Limited, handwriting commercial vendors.
Shows lots of collected base forms (with Suen).
Context helps on character recognition: specifically modified Viterbi algorithm, even on "unconventional" English text.
Context via Markov method, plus a dictionary, better than predictor-corrector method.
Cited in Sinha88: on spelling/dictionary context correction.
Review of extracting 3D information from 2D images: shading, binocular imaging: mostly machine vision. Also sketch recognition for hand-drawn (scanned) electronic circuit diagrams.
Refers to IBM direct-manipulation office desktop user interface "Pictureworld" (like Microsoft's "Bob"?), with file cabinets, mailboxes, notebooks, phone messages.
Direct manipulation: gesture/command symbols user interface. States that direct manipulation UIs are more like actual objects people already comprehend, therefore easier to use (e.g. VisiCalc spreadsheet and a paper spreadsheet). Defines direct manipulation as manipulating digital objects on a screen without the use of command-line commands. Has definition of direct manipulation, page 64.
On-line handwriting recognition using two-dimensional graph (chain codes) using slope/angle of lines, intersections, labyrinthology, for features.
Recognition performance 71% on Munsun's OCR handwriting data set, improved to 86% with Viterbi context algorithm.
Digitizer tablet with four-button puck, experiment in GUI design with floating (pop-up?) and dynamic menus.
Context: using low-level speech context, hypothesizes word for higher-level context analysis. Context to resolve co-articulation on continuous speech recognition. Cites experiments to determine human use of context in speech. On speech recognition, discusses possible biases and justifications for them in training and test data collection.
Design of Xerox Star GUI with mouse and keyboard: mouse uses trackball rather than SRI wheels. Development philosophy: throw away first implementation, it was always a prototype. Properties and application sheets for graphical objects. Move is by click-source/click-MOVE-button/click-target, not drag/move gesture. Keyboard windows / on-screen keyboard / virtuelle tastatur, similar to soft input panel SIP?
Desktop user interface patterned after physical office: forms/records have no physical counterpart. Data icons and function icons, e.g. icons for printers, file cabinets, folders, property sheets. Keyboard windows / on-screen keyboard display, similar to soft input panel SIP? press-and-hold? Earlier references back to 1977.
Touch-sensitive touchpads in music for expression: add touch surfaces to clavier keyboard keys to send finger-force during a note, or movement in in Y direction (X is pitch, need the discrete keys). Back of black and white keys merge to a flat surface for continuous pitch control in X direction. Linear touch sensor may be sub-divided into sections for multiple fingers on same pitch (multi-touch). Reducing motion travel of key allows faster playing: substitute with force sensor.
Cites multi-touch to Bill Buxton in 1983 for musical instrument input. Musicians rely on haptic feedback (feel of the physical instrument) in playing. Optical multi-touch input using "wine glass effect" of total internal reflection on a glass surface.
Hierarchical menu breadth / wide menus gives faster user actions than tall / narrow menus.
Paper study (as in paper-based prototype) for a interface for database use (office) combining touchscreen pointing, gesturing, touching, and spoken voice recognition commands. Anticipates using wall-sized display, in-air pointing by ultrasonic or capacitive sensor to point to icons. Refers to in-air pointing (hover, proximity) as Teletouch, "fingerprint" confirming cursor to indicate icon is being pointed to. Cites to Micropad handwriting recognition terminal, and Bolt "Put-that-there" gesture input.
Not sure about conference name.
Spelling dictionary string correction, using a posteriori computation after recognition is done.
Context from bottom-up (probability from previous letter sequence), channel (probability of "A" corrupting to "B"), and top-down (lexicon).
Spelling correction for substitution errors: same as Hull83a.
Cited by Marlin Eller, Microsoft Pen Computing group.
Line-thinning for handwritten data.
Skeletonization/thinning: side effects are spurious projection, necking, tail generation, noise holes, with heuristics to fix them.
feature selection critical in pattern recognition (see other references on choice of features).
Cited in Faggin89: for pattern recognition on human faces.
Cited in Faggin89: in connection with Wizard/Wisard adaptive image classifier.
Describes Suen's data base of 174 types of sample characters.
Describes printing, script, and manuscript styles and their legibility to humans. Says writing speed for printing increases with practice. Digitizer spec is for 3/8" proximity sensing range.
User's manual for Summagraphics Bit Pad One magnetostrictive tablet digitizer. Describes (accurately?) magnetostrictive technology. Has Schematics of control electronics.
Application note for Summagraphics data tablet, pattern recognition for handwritten characters at University of New Brunswick Computing Centre, handwritten characters: pre-processing and classification subroutines, with menu-picking areas (rectangles) on side of tablet.
BitPad-1: de facto serial interface standard for digitizing tablets, especially with stylus. Sales literature mentions data entry, cursor control (mouse/locator input), elimination of typing. File includes photo from Buxton Collection of four-button tablet puck.
Electromagnetic digitizer: describes increment mode, with minimum distance to record movement in stream mode. Compare with press-and-hold?
Viterbi algorithm plus a trellis of partial patterns, to do recognition of whole from context of parts.
Fuzzy sets applied to pattern recognition where classifications overlap.
Joystick, track-ball, digitizer tablet, with two-handed input.
Handwriting cursive recognition by point-by-point distance measurements against "ideal" cursive writing. Does not do well with loops vs cusps, which are forms that transition into each other in handwriting motion and variability. Refers to hook effects due to weak tablets as a big problem: compare with Ward patent and article on digitizer tables. Combines letters in connected script to match whole words (context/dictionary.).
Cited in Bozinvic89.
Bozinovic89 cites for human recognition by outline of word (as shown in WrightG52).
Business plan for CAD/CAM company. Product used two displays, one for displaying menus and using lightpen/touchscreen input, the other for the CAD display.
Workshop report on graphical input interaction user interfaces: including handwriting recognition GUI.
Tablet pen with cylindrical off-axis force/pressure transducer in the tip, to measure side-force on the stylus, for graphic arts rendering. (Inventor of KoalaPad).
Cited in Bokser88.
OCR hardware: recognition via template comparison.
Cited in Sklarew99 patent.
Remote chalkboard / whiteboard / telautograph, display remote cursor showing where the stylus/marker is in proximity (?) so that things can be "pointed to" remotely without marking. Digitizer tablet patent on automatic cursor, mark erasure (recognition strokes removed in electronic ink user-interface), GUI display of eraser in remote whiteboard system. Uses "telautograph" system as term for whiteboard system, but unidirectional whiteboard, only communicating one way. Use of a display/proximity cursor on whiteboard to help other user notice where things are being written, and to "point to" images.
Quoted in Hakmatpour86 (missing from this list?).
Deals with forms in office management, says flow of forms is important.
Mentions non-paper "forms" for voice.
Refers to operations on forms: automatic calculation, etc.
Capacitive fingerprint sensor: array of individual capacitors, finger ridges push plates closer together, alternatively sensing plates capacitance to flexible conductive sheet deformed by finger ridges. Later cited in touchpad design.
Skeletonization / line-thinning.
Cited in Hull83a.
spelling/context correction using structural information: invalid prefix ("in" for "im", joining of suffix (beginner vs beginner), etc. rather than just comparison with word list.
Speech: how contextual factors are controlled in test protocols affects recognition results. Speech recognition failures completely explained by a limited set of underlying phonemic (human recognition) features.
Combinatorial solution to stroke connection and stroke order variations. Break strokes into equal-length pieces, then match them as a feature.
From collection of internal documents on the history of AutoCad: June 21, 1983. cites special concern on marketing deal with Sun-Flex (touch-pen touchscreen digitizers), Touch-pen considering alternative vendor P-CAD.
opacity mask and image blending (similar to alpha blending) for anti-aliasing.
Cited in Yu90: on context for line thinning using chain codes.
Uses eye fixation to determine what features are cognitively important.
Repeats and continues Blesser et al's early work from Massachusetts Institute of Technology.
Dedicated dual processor chip used for speech recognition.
TABLET - Query language, more procedural than SQL. Human Factors (learnability) of a programming language. TABLET is more verbose, Cobol-like syntax.
Handwriting recognition using nonlinear adaptive procedure: thresholding of scanned OCR images.
Pre-processing to clean up OCR scanner images of carbon copy forms, bank checks, smudges, scenic backgrounds, etc.
Beringer89, electronic ink hardware?
Includes definition of touchscreen and touchpad which may be separate from display, review of touch-screen digitizer technologies. Mentions position accuracy errors at light touch or at end of a stroke as finger is lifted. Touchpad uses mylar sheets held apart by air pressure: micropad? Fall-out errors (inadvertent skips) on touchpad/tablet.
IBM 5550 Japanese (Katakana and Kanji) display terminal, 24x24 matrix characters. See also CIC Handwriter handwriting recognition tablet for 5550.
Allograph (variant) selection based on preceding context character.
Cited in Lipscomb91.
Template matching of varying signals by filtering/smoothing at different scales to reduce noise, filtering is adaptive and variable.
Early multi-touch sensor (cited in Lindemann): Pad sensor, 8x8 grid of pressure (force) sensitive sites, connected to strain gauges for parallel force. References for artificial force-sensing/touch-sensing robotic skin.
Chain codes of optically scanned / static character recognition OCR.
Introductory User's Guide to Xerox Star. printing, mailing, document editing, ford processing, etc. Point-and-click with mouse, no drag gestures.
Video demonstration of Xerox Star graphical user interface circa 1982. Posting comments opine no checkboxes or direct manipulation. Two-button mouse: separate point-and-click specification of selection and destination: special function keys on keyboard for Again, Find, (make) Same, Open, Delete, Copy, Move, (show) Properties. Additional for expand/contract, margins, font, undo, etc. Three steps for move: click-to-select, function key, click-on-destination. Single/double/triple click to select character, word, sentence, paragraph. Uses term virtual keyboard. See also 1998 retrospective demo.
Using multiple features rather than single feature, and weighting results, gives better performance on character recognition.
Outermost point recognition method for Romanji, numeral, Katakana.
Tactile/haptic display/feedback in hand-held radio pager in conjunction with alert tones.
Science Citation Index.
Well-writing: teaching users how to write Chinese, not recognize Chinese handwriting.
Encoding of handwriting motion/motor input based on model of handwriting motion.
Model of handwriting dynamics motion for fast handwriting (more dynamic).
Shneiderman83 cites for direct-manipulation desktop user interface. Desktop UI displayed on character-mode terminal. No stylus/digitizer input shown, uses joystick, arrow keys, keyboard.
National differences/variability in writing styles (Japanese, Chinese, Korean).
Claimed 97.8% accuracy by excluding 5% of data as "poorly written".
Stroke order, stroke connection variations in Chinese.
The 214 standard Chinese radicals "too many": some are rare.
Recognition diagram: signal filter, segment and direction, stroke.
Classification, alphabet/element recognition, composite ideograph output.
Contrast: features of primitive "stroke element" recognition vs chain code segments.
Transparent touchscreen with anisotropic conductor (transparent silicone rubber with stainless steel fibers to convey touches) and transparent elastic silicone rubber insulator (durable) or polyurethane.
System connects strokes of "normal" to make "running" forms (variability?), feature is angle sequence (chain codes?).
Claims 99.5% accuracy on handwriting recognition.
Character represented as branches in a reference pattern feature sequence. Detail "discrimination logics" for character which the regular method cannot handle. Kanji and Hiragana handwriting.
biblio80 (Prev page)
(Next page) biblio85
This compilation and all annotations are copyright © Jean Renard Ward, 1993, 2004, 2011, 2013, 2016.
I can be contacted at firstname.lastname@example.org.