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 1970 to 1975.
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 © Sun Feb 19 08:19:30 EST 2017.
Mirrors of this page can be found at:
biblio70 (Prev page)
(Next page) biblio80
Draft for ANSI74 standard handwriting styles, with subset character sets.
ANSI74 (ANSI X3.45) standard handwriting styles, with subset character sets.
Shows preferred, rapid writing style for block-printed capital letters and digits, with stroke direction: taught by the US Military also, constrained handwriting for human recognition.
There are specific feature detection cells in human vision, and that there must be something similar for hearing: in any case, a good recognition system will be modelled on these feature extraction mechanisms. E.g. Vision has specific cells for horizontal and vertical lines.
Measure a number of features in a (typed character) dataset, then pick the features for recognition based on pairs of correlations.
String pattern matching.
Constrained handwriting styles to make them readable by OCR.
"pressure" resistive sheet tablet, two sheets held apart by thin web of rubber or other elastomeric material with many small perforations. Material is such that pressure from wrist or hand is not sufficient to make contact between sheets, but stylus point is. Assigned to Telautograph.
Web language (directed graph grammar): graphical programming (visual programming) language on a tablet could be transformed into resistor, capacitor, or other components, transformed by heuristics into higher-level logical objects.
Mentions use of two-dimensional tablet GUI for a programming interface for a graphical editor: actual article discusses a block-oriented programming language, somewhat similar to that of Plamondon from about 1990. Cites GRAIL (flowcharting, text); Anderson, Blackwell and Bernstein (mathematical notation); Datastructures (AMBIT).
Capacitive tablet: proximity sensing to 8" to stylus, use tablet as 3D/three-dimensional sensor for Z height. Using multiple sensors, can get 3D sensing of model objects (doll figures).
Sketch recognition / world's first gesture recognition user-interface in a commercial product: describes Ledeen gesture recognizer?
XDS 9300 computer with ADAGE AGT/10 graphics computer, light-pen input, basic graphics operations (rotation/translation/scaling of drawings, etc.).
Deployment of IMP processors for ArpaNet: debugging of single IMPs in simulation works well, debugging in cluster not well. Code checksummed periodically for reliability. Multiple redundant processors easy to debug, system continues operation even with some wires pulled out, etc. Case-analysis of network surviving partial failures. Compare with high-availability systems?
Optacon touch sensor for the blind: small set of vibrating haptic transducers with hand-held scanner. Output to one or two fingers. Intended as accessibility aid for reading of printed material.
Transparent digitizer touchscreen using transparent switches on a display. Mentions need to mount closely to display to minimize visual parallax so that combination appears to user as a single device. Note: multi-touch -- "child touches the clock face" and a second image concurrently. Small pads for use with stylus for high resolution (contact density). Cites to Johnson 1969.
Chapter 2: Contains examples of boundary theory decision surface diagrams, but doesn't discuss it that way. Refers to human "teacher" for trainable recognition: e.g. perceptual/cognitive ambiguity about what is an "A" vs an "H".
Decision surfaces in multi-dimensional feature analysis (pair-wise comparison?). Statistics: similarity, distance, and compactness of decision surface/classification. Implies boundary theory vs category theory. Accidentally shows European variants of hand-written "H" character.
Two-axis / two-dimensional haptic force-feedback display, used to illustrate to students electromagnetic and gravitational fields in two dimensions.
Transparent capacitive touchscreen over CRT display. Copper on glass is "invisible". Approach of finger changes capacitance to ground. See also Stumpe.
Book review of 1974 reference: use of context, frequency of occurrence of binary words (two characters / character pairs?) in character recognition.
Book edition of doctoral thesis on combining subpatterns that are classified/recognized less reliably to get a better overall recognition performance.
Tactile/optical scanner for the blind. Cites to Optacon. Mechanical spiral scanner, display "pens" (similar to braille display) show dark/light and dark/light transitions, e.g. edges.
Review/opinion article on mathematical expression input and output using tablet. Cites to GRAIL, etc.
NTIS citation index.
Speech understanding system, describes acoustic-phonetic and lexical-mapping processes.
NTIS citation index.
Speech understanding system, describes acoustic-phonetic and lexical-mapping processes.
Refers to Blesser, et al and functional attributes (distinguishing features). Mentions use of pressure (actually, force) and timing (!) in handwriting recognition -- very gesture-like.
Barry's paper on perception in reading medical X-rays. Set error leading to recognition/perception bias: you found what you are looking for, so you do not recognize other, fairly obvious things. Plausible interpretation == grammar of seeing, what is a legitimate perceptual "sentence".
Examples of attributes/features: functional closure, perceptual closure, physical closure.
Published version of Shillman's Thesis work: pair-wise distinguishing characteristics (features) of hand-printed characters.
Character recognition based on ambiguously shaped characters or "difficult cases" of low quality.
Graphical context rules extracted via psychological experimentation.
Arguments for boundary-theory pair-wise comparison for recognition.
Contains chart of Shillman's ambiguously shaped characters.
Cited in Milloy78 for confusion/substitution errors in human recognition for lower-case characters.
Cited in Bozinovic89 for human recognition by outline of word (as shown in WrightG52).
low-resolution touchscreen using separate sensing pads individually wired to circuitry.
Fundamental knowledge of the brain would greatly help on pattern recognition by machine.
Difficulties in pattern recognition are generally combinatorial.
Most theoretical papers on pattern recognition are quite worthless.
Problems like handwritten character recognition are not even well-defined.
Handwritten "A"s are not mathematically defined: just what people agree on.
Qualitative and formal-language "theorems" do not help solve pattern recognition problems.
Pattern recognition theorems (perceptron/neural-net) working without quantitative combinatorial analysis.
Adaptive systems may not converge on optimum performance: stick on local maxima.
Most papers give no reasons for choice of recognition features.
Most recognition systems fail on choice of features.
Features chosen must separate patterns in recognition.
Humans do not use mathematical features, like string probabilities.
Gives fundamental reasons why polynomial mathematical features will not work.
Cybernetic recognition (measure deformation from paradigm) vs fuzzy sets vs perceptrons, etc.
Steepest descent, other gradient methods in statistical pattern recognition fail due to local minima.
Picture grammars (linguistic scene descriptions) useless if elements cannot be identified.
Creative imagination of researchers better than statistical recognition.
Examples of nature best design inspiration in recognition.
Many papers just an exercise in fashionable recognition techniques.
"mount Bayes is overpopulated" -- critique of statistical pattern recognition.
Feature extraction plays a role in visual cortex.
Cited in Kruskal83.
Handwriting recognition using acceleration chain codes as the feature (?).
3D three-dimensional digitizer for multiple points. Points are light-emitting diodes LEDs. The light-emitting diodes are easily distinguished from one another by flashing them on and off in sequence, under computer control. The position of a light source is measured by scanning the environment. Fast one-dimensional, geometric scans quickly determine planes in which a light source lies. The point of interest is at the intersection of these planes.
Cites to Lincoln Wand, Lawrence G. Robert, 1963, three-dimensional acoustic/sonic digitizer: this design is an optical digitizer in three dimensions.
From Kuklinski file.
Broad (1972) review of problems handwriting recognition must deal with: consistent performance (gross features only are used), etc.
Has many pages of samples of the handwriting collection they used.
handwriting recognition on for writing the ANSI standard characters: constraints.
Many examples of calligraphy, hand-lettering, embellishments on basic hand-written letter forms, including stroke directions, hooks, etc. Handwriting variability.
Capacitive keyboard, keys wired into matrix of rows and columns. Individual keys are read by addressing rows, then columns, and the state read into a shift register where the index in the shift register matches the row/column address of the key. Thus, all keys can be addressed and read independently for N-key / whole keyboard rollover. Compare with Kaplow multi-touch.
Early packet-switching-network paper describing TCP/IP (pre-UDP), including varying packet sizes, end-to-end error checking, tolerance of transmission failures.
Cited in MartinGL90 on analysis of how well neural nets work.
Cited in Fisher75.
Massachusetts Institute of Technology undergrad Thesis on line width and functional attributes.
Final report on project "Machine-Independent Software Programming" AMBIT. Visual programming language, touchscreen graphical editing with gestures, running on TX-2 computer. References to Ledeen recognizer, constraint languages. agd symbolic debugger. See additional materials in files.
Implementation of visual programming language AMBIT/G, represented as two-dimensional directed-graph diagrams for both data and rules.
Visual programming language AMBIT/G, requires definition of all data structures/representations before starting programming algorithms.
Cited in Herbst76.
signature verification stylus using tip force/pressure vs. time.
Optical touchscreen, light emitters are pulsed in sequence so that beam collimation/focus not required.
Brief talk on Leon Cooper on getting interested in neural networks and memory, after getting Nobel prize re superconductivity. Founder of Nestor handwriting recognition company.
Optical touchscreen/tablet: two highly collimated light beams (e.g. laser) swept across above surface from two corners on same side, reflected from reflective stylus, position determined by triangulation.
Cited in IchikawaS84.
Angle (vertical vs horizontal) perception in humans?
Context of adjacent lines in human perception of vertical and horizontal line length: human feature perception (many other references on this topic).
using keyboard, shaft encoder, other devices as virtual replacements for lightpen/tablet/mouse. finger as stylus, Lincoln Wand as 3D digitizer.
Cited in Wallace76 for "virtual devices".
OCR: Variability is two sources: type font, and scanner/ink effects. Most variability among characters describable in terms of stylistic consistency. Stated aim of pre-processing is to reduce variability recognizer must deal with. Adaptive recognition should analyze for stylistic consistency as a separate variable, and separate variability from recognition. Variability model is NOT same as picture grammar approach.
SRI/Crane's accelerometer-pen character recognition: Uses ULDR chain codes, similar to BLRT chain codes.
SRI/Crane's accelerometer-pen for signature verification/recognition e.
On-screen/virtual keyboard with labels displayed on monitor, transparent plastic keys over the display. Keyboard functions change with application. (On-screen menus).
Acoustic digitizer: horizontal raised lines across surface, piezo transducers or microphones pick up sound of ballpoint pen's vertical movement over lines while user writing signature. Biometric verification of signature based on Y velocity profile.
Claims 2% error rate (but 8.5% reject): feedback to users greatly improves handwriting recognition.
Handwriting recognition features are coding zones, sequence similar to chain codes, number and length of horizontal/vertical strokes.
Describes resistive-sheet/film digitizer tablet, one character big: user-interface for handwriting, one-character input.
Early LCD projector patent. Static information can be printed on the glass as well.
Early electromagnetic digitizer tablet patent, using Gray Code grid pattern encoding, using phase of induced voltage.
Finding/detecting lines in B/W pictures: example is edges of a machine cabinet, extracts parameters first, lines detected indirectly. Cited in Kahan87 regarding a statistical Bayesian classifier using binary features.
Herbst76 cites this on signature verification using force, velocity, pen lifts, correlations among them.
Signature verification using velocity and acceleration, using a graphic tablet.
Cited by Marlin Eller, Microsoft Pen Computing group.
Contextual pre-processor for script using letter length, word context, stroke parsing: best character alternatives given to word-context recognizer (likelihood ratio).
Due to scope of problem, one becomes mired in programming.
1.3% error when training and test data were identical.
Word-wise recognition of hand-written script.
Refers to certain degenerated, trivial, toy recognition problems and their pointlessness.
Need for low-quality data in training a system.
Goshtasby88 cites for binary letter pairs (digrams) for spelling context, following by spelling dictionary look-up for what's left.
Describes dividing a tablet surface into a simulated keyboard area, a sketch area, virtual tablets, etc. Relevant to Schumer patent.
Cited in Sklarew99 patent.
Mentions response bias in measuring human recognition errors.
Problems of matching recognition results of tests and predictions based on only one variable: human recognition is a correlation function of multiple variables.
Points out problems of weighting multiple variables/features.
Capacitive grid tablet digitizer, using phase-modulated waves for relative and absolute position. Alternative form uses capacitive traces on a rotary drum, stylus/probe couples to conductors on drum.
Cited in Agui79.
Cited in Tappert's patent list of 1986 on on-line/dynamic handwriting recognition.
Cited in Fox88: on annotation drawing/handwriting.
Optical digitizing tablet: stylus shines a light on a transparent glass plate, with a scanner underneath.
Pair-wise discrimination: guidelines for when to work harder on hard-to-separate classification pairs.
Cited in Fox92.
Translucent (transparent) touchscreen with rear-projection, using X/Y grid of conductors with physical contact of an X and a Y wire. Mentions multiple touches (multi-touch) as something to be detected and rejected. Did not find a transparent and flexible conductive material: Polacoat (commercial sheet of lenses) is conductive and transparent, but not flexible. Metallic-coated mylar for resistive/conductive film touchscreen.
Adjustment of perception of stimulus to be what is "reasonable": We accept hypotheses about the identity of individual letters and words only if they yield a sensible interpretation of longer sequences like phrases and sentences. (Context information in recognition).
Cited in Meisel: 85.
Referred to as "seminal paper" on separating semantic, syntactic, and lexical levels in user interfaces for interactive applications, and equivalent devices (e.g. pointing devices). Cited in Wallace76 for "virtual devices" (virtual tablet).
Cited in Sinha88, Hull83a.
Description of Viterbi algorithm, applications to text/character/OCR recognition using context.
Cited in Nouh84.
Long treatment of art on extracting line drawings / line structures (line traces) from images of line drawings, and also on semantic understanding of line structures. Cites to terrain recognition from topological map data.
Cited in Kerrick88.
Pressure/force sensitive tablet using compliant surface over cross X and Y matrix of ridges, electret capacitive pad in wells of ridges. Not multi-touch.
Cited in Baird84.
from Tappert's bibliography.
Program to repair distorted line drawings (not images) automatically. Pitch: Compare with later graphical constraints, constraint-based programming, graphical syntactic rules, Waltz 1970's on picture understanding.
Earlier, more detailed presentation of other Goldstein 74. Drawing program uses Turtle graphics, planner creates program.
Capacitive stylus combined with writing pen: metal housing. Not retractable, just an inking stylus for digitizing while writing on paper. Cartridge position can be adjusted to capacitive coupling to tablet.
Study of positioning time with various cursor, pen, keyboard devices.
User-interface: Hardware patent on digitizer erasing on display with a light pen, makes broad claims on erasure/editing of freehand writing or sketching of electronic ink.
"goodness" scale for subjective evaluation of speech quality: relates to cognitive psychology experiments to extract functional attributes.
Method for subjective evaluation of cognitive/psychophysical perception/recognition, where theory is not adequately developed for quantitative prediction.
A standard reference on compiler /programming language design.
Master's Thesis at USAF Inst. of Tech. Recognizer for human-sent Morse code, 95% accuracy, biggest problems is detecting the tone in the presents of noise (sampling).
Early patent on zone/region on-line character recognition.
Cited in Fisher75.
Cited in DosterW77.
Cited in Litvin82, Nouh84.
Transparent touchpanel using grid of conductors, may be overlaid on printed indicia display. Modular construction as component for different application systems.
Technical report on IMP/TIP interface message processors / terminal input processors for ARPANETY (pre-Internet). Data performance files transferred over network, instead of tape. Incremental/continuing changes to routing tables, routing based on dynamic connectivity/performance.
German/European handwriting style textbook.
Cited in Bokser88; Cited in Tappert88b: on segmentation combined with character recognition.
Optical/typed character recognition by comparing with a set of individual prototypes for each possible shape, and taking best match score. Comparison may be pixel-wise or slice-wise comparison.
Massachusetts Institute of Technology Thesis on generating simulated/artificial variability in handwriting.
Find samples at the boundary of where recognition problems occur.
Says problems in recognition algorithms should be found systematically.
Says need 10,000 training samples to get feature space error down to 1%.
Says Neisser's figure for human recognition too poor for practical device(!).
Says testing takes up most of the effort, but is treated too lightly.
Quotes "dominant factor factor in performance is care in printing".
Three methods of character recognition are matrix matching, Feature extraction, and curve tracing.
Context in sketch recognition (sketch editing?) on tablet. Example is syntactic rules for construction of a house: e.g. house/living-room/wall/window.
Cited in Purcell77.
Digitizer using surface acoustic wave: transparent touchscreen.
Simple OCR for handwriting recognition, using a single vertical line as a constraint, measuring closed/open loops and areas as features.
Multics operating system architecture and overview: CLI command language, input/output facilities, file system. Segmentation/paging, virtual memory; access control lists for selective sharing (ACL), access isolation mechanism (AIM), ring structure and levels of privilege. Introduction to Multics software development environment, dynamic (run-time) linking to subroutines. Unstructured, sequential, indexed files. Section VI on access control, modes, syntax of ACLs. AIM allows delegation of access privileges grantor has via ACL. Processes (and all objects) identified by name Project.User.Tag. See ContentGuard Google. Automated full and incremental backups (dumps) of files/objects. Access class associated with an object for its lifetime.
Character recognition features used by IBM OCR page reader. Measurements on data (feature extraction) and classification algorithm closely tied to problem at hand.
Cited in Bennett 91.
optical stylus digitizer tablet somewhat similar to Anoto: visually transparent surface using a coded grid placed over a display: appears to give proximity detection also.
Resistive sheet digitizer, two resistive sheets with AC voltage gradients in orthogonal directions. Single probe (not stylus) punctures both sheets and picks up both signals.
Resistive sheet digitizer, single sheet, voltage gradient applied alternately in vertical and horizontal direction, using second conductive sheet and intermediate layer of deformable insulation such as a gel, that permits the two sheets to be pressed into effective contact.
Resistive sheet digitizer, single sheet, voltage gradient applied alternately in vertical and horizontal direction, using second conductive sheet and small spacer bumps to hold two sheets apart: palm discrimination versus narrow stylus tip.
Refers to Munson's data collection / test sample data base for handwriting.
Essentially a telautograph: mechanical linkage with orthogonal potentiometers to determine position of writing stylus on a tablet. Writing can be recorded on a tape recorder (analog format) for storage.
Tilt correction patent for tablet digitizer.
Cited in Litvin82.
From a old list of Polhemus patents.
Inner-product comparison (pixel map) OCR character recognition against standard reference template patterns, RMS average of match scores compared to threshold.
Cited in MaedaK87.
OCR of characters (handwritten?) using sums of squares of values for similarity: improvement over patent 3,688,267.
First description of BIT BLT routines for Xerox Alto computer: Bit Block Transfer. Main operation is FillBitMap(...). includes Source code with date of April 20, 1976.
NLS text editing with mouse for text selection and keyboard: could show two (text) files at once, multiple users edit same text from separate terminals.
Light pen digitizer, subdividing display into blocks in the X and then in the Y direction.
U-V recognition: functional attributes Blesser's and Shillman's group at Massachusetts Institute of Technology.
Optical touchscreen designs. Light beams in X/Y grid over CRT display, finger interrupts and X and a Y beam. Internal reflections in transparent deformable layer (compressible silicone rubber), finger pressing on layer interrupts internal beams. Also sonic Rayleigh waves in surface of glass detected acoustically.
UNIX documentation on YACC; regular-expression pattern matching.
Neuronal activity, evoked potentials in frontal lobe (but not visual cortex), matches human perception/recognition of recognizing ambiguous character shapes.
"It is well known from neuro-physiological studies that the visual systems of cats and monkeys incorporate cluster detectors in several stages of hierarchically increasing complexity, beginning at the retina of the eye and extending to the highest levels of the cerebral cortex.".
Why adaptive recognition is not human perception/cognition: multiple levels.
Perception (low-level, immediate) vs cognitive (high-level, takes time).
Mentions cognitive/perceptual problem of reading alphabets.
TICS system for authoring tutorials for interactive education. Cites to Coursewriter. Examples include foreign-language instruction. Courses structured as web of nodes, connected by multiple possible branches. Compare with TutorText early 1960s.
Early paper by Alan Kay on Dynabook, does not show handwriting input: touchscreen (strain gauges at corners) allows on-screen virtual keyboard display, some references to drawing. Early personal computer with ARPANet/Internet.
Cited in FoleyJD82.
Re-label keyboard by projecting images on top. Compare with Kaplow.
Graphical context on rotated text perception with human readers: relation to dyslexia.
Cited in Suen79.
Cited in Blatt88: 97.1% character recognition rate.
Polhemus patent on generating rotating fields to digitize position and angular position at same time (3-D tablet plus rotation, theta, phi, rho), six-dimensional electromagnetic digitizer.
(3,868,564 is a motor patent from the USSR).
Cognitive experiments showing correlation between human rating of "goodness" of paired character shapes, and recognition time.
Graphical context affects perception of characters, not just semantic context.
Graphical context affects perception of characters, not just semantic context.
Variability/uncertainty a fundamental property of formal description.
Linguistics/perception/cognition as study of boundaries, not categories.
Justification of fundamental philosophy for our boundary-cognitive.
Analysis of functional attributes/features.
Feature analysis: not single property, but co-occurrence of large number of items.
Touch sensor, two electrodes (shaped to match finger tip): conduction by (high resistive) skin when electrodes touched. Alternative is to put elastomeric film above sensors, bottom layer conductive, finger presses film to the two electrodes. Not capacitive.
Y-V discrimination: functional attributes Barry's and Shillman's group at Massachusetts Institute of Technology.
Optical hand-held scanner using a single sensing element, straight-edge or other guides for moving the stylus across a page: compare to optical digitizers.
UNIX documentation on LEX: regular-expression pattern matching.
Cited in Purcell77.
Tree hierarchy for graphical data objects for fast rendering, based on fast search and access within a given sphere. See later work by Levine on Wang Freestyle.
Static/OCR handwriting/handprinted character recognition, if first level classifier has no clear winner (20 weighted features), uses factors of two highest-scored candidates. 20-segment constrained handwriting style, Gaussian point-to-line distance to segments. Limited punctuation: period/dot does not fit onto segments.
Study of human character recognition for handwritten letter pairs, comparing the effects of auditorally similar letters to visually similar letters in deaf children. Refers to phonetic coding in memory.
Cited in Fitzgerald76 patent, example of sketch recognition / sketch processing.
Resistive sheet/film digitizer, using a single sheet and a conductive stylus and wire. Mentions integration with a display by applying the measured voltage directly to the deflection plates of a CRT (telautograph).
Theory for perceptual interpretation, subjective impression for context.
Kuklinski has margin notes on psycho-physical (perception), judgmental, and linguistic context.
Refers to other work on semantic (language) context.
Page 239 - relates to inadvertent bias in testing recognition results, since judgment affected by the rest of the characters the subject looks at (or writes).
Page 240 - Kuklinski's margin notes on all the forms of context bias.
Cited in FoleyJD82: for user-interface for mathematical input, two-dimensional.
Illustration shows use of character extents, spacing, size, position, for parsing two-dimensional hand-written mathematical formulae.
Compares "abstract" issues of pattern recognition with heuristic/linguistic approach specific to problem area.
Feature selection by statistics "gives no insight into dependencies among features".
Inter-set and intra-set feature distances an ill-defined concept.
"curse of dimensionality": how feature interact and overlap a hard problem.
Project features onto one dimension and you cannot tell how features inter-relate.
Human recognition/perception: human readers (non-Arabic, non-Hebrew, non-Chinese) perceive characters to the left of the fixation better than to the right.
CRT displays are too large, and resolution too low for many application (ca. 1971).
Pressure-sensitive conductive rubber: conductive particles in an elastomeric binder.
Nominally optical digitizing tablet using hologram patterns printed in a glass surface.
Character recognition for simplified character set using endpoints/termini only, which had to touch one of six reference points/zones in a hexagonal pattern.
Optical drawing input (analog digitizer) using optical prisms that can respond to any touching object: rubber stamps, fingers, brushes: light affects flying-spot scanner for a "paint-pulse" input. Also shows piano-type keyboard for selecting visual effects. Later cited by Greene for multi-finger/object (multi-touch) drawing system using same principle.
OCR by projecting in X and in Y direction for histogram.
80.4% character recognition rate claim on static (OCR) handwriting character recognition using (unspecified) multilevel logic after skew normalization (slate).
Sketch recognition as early problem in artificial intelligence. Includes recognition of gestures, adaptation to user, pen pressure on Sylvania tablet. Uses report rate from tablet as clock to determine dwell. Pressing harder makes lines darker (user wants to show that line is correct), Z proximity sensing. Transparent touchscreen, first pressure-sensing pen/stylus? System asks for user sketching samples (e.g. slow and fast squares) for initial training to user. Squiggles/rub-out gesture. Adaptive thresholds on pressure/force.
Cited in FoleyJD82.
Sketch recognition as a form of data compression: primary example is recognizing straight lines, simple curves. Constraints (two lines parallel, same length, etc.).
Indexing by handwritten electronic-ink user-interface using sketch recognition / sketch matching.
Speculative, comic-book style publication on future computer technology: long section on PLATO touchscreen terminals for education. PDP-11 GRASS system with Vector General tablet. Text editing with light pen (Parallel Textface application) showing multiple windows with sections of same document, for editing hyperlinks. Cites to Ledeen character recognizer.
Signature verification, mis-called signature identification, using pressure pattern without position. Pressure transducer is strain gauge or piezoelectric sensor in stylus/pen, etc, alternative the tablet/platen is a force/pressure transducer.
Cited in Sinha88.
The handwriting/gesture symbol recognition system used by Applicon CAD/CAM system: everyone used to copy it. Trainable. Refers to training to recognize a full alphabetic character set, but with some neography required.
Chapter eleven: cannot use mouse for tracing, compared with digitizing tablet. Mentions RAND tablet (electromagnetic), voltage-gradient (resistive tablet), Sylvania tablet (electrostatic/capacitive), acoustic tablet (SAC: see Science Accessories Corporation).
Scan of book excerpt: RAND tablet, etc.
Digitizing tablet, nominally optical: holograms embedded at different positions on the tablet.
Electromechanical 3-D rod (3DOF) with force feedback to simulate surfaces haptically. Based on PhD Thesis.
Electromechanical 3-D rod (3DOF) with force feedback to simulate surfaces haptically.
From Kuklinski's old files.
Dynamic time-warping / non-linear time alignment via dynamic programming for signature verification: published later in 77/78.
Architectural overview and design of the MULTICS system. Paging using associative memory tables. Runtime (delayed) linking and loading. Ring/bracket protection of entry points. Interprocess communication. Hardware support for access rights (access control), ACL / access control lists on files and objects. Access and mode control on I/O devices. Not much on networking.
Historical notes on Papenmeier BRAILLEX, early/first tactile Braille display.
Best piece-wise fit of line segments to a polygon: split and merge segments to drive error below a threshold.
Piece-wise line segment fit helps on feature extraction, data compaction, and noise filtering.
Chain codes - cite this one so Pavlidis gets his licks in.
Logo graphical turtle, with control interface with physical buttons for toddlers: buttons labeled with command and numbers, with a memory, so see whether toddlers (ages 4-6) would learn to program. Cited in Minsky84.
Touch-panel digitizer using force gauges/sensors at three or four corners of a plate: compute location of touch from corner forces. Plate can be transparent, and sensors are only at the corners.
Cited in FoleyJD72.
Deformable resistive sheet digitizer with insulating mesh between so that resting hand does not cause sheets to make contact. Palm rejection.
Describes character recognition based solely on chain codes / stroke direction / arc segments.
Describes character recognition based solely on chain codes / stroke direction.
Edge-defining algorithms for OCR of handprinting/handwriting, maximum likelihood statistical classifier.
Cited in Burr81.
Rear-projection display on translucent acoustic time-delay touchscreen digitizer, receivers for sounds of tap/contact of stylus with the digitizer screen. Sound generator in stylus tip.
TV/Video display with transparent spatial signal generator (touchscreen) on front: grid of conductive sensing wires, or pen attached by rigid or string members from stylus to potentiometers.
Study showing that culture affects optical perception: Singapore Chinese more subject to Mueller-Lyer line-length illusion than Singapore British.
Cited in Goshtasby88 for binary letter pairs (digrams) for spelling context, following by spelling dictionary look-up for what's left, compare with Viterbi algorithm or with trigrams.
Cited in DosterW77 on a spelling dictionary: important point of diagram is only zero/non-zero value.
Automatic determination of useful features for OCR character recognition (on thinned images) by mathematical determination of clusters matching desired outcomes. Notes that the test data set was unrealistically small training set.
Preliminary abstract of Fisher75.
Cited in DosterW77 on a spelling dictionary. File also contains list of numerous reference *to* Riseman for context information in handwriting and speech recognition.
Electrostatic grid digitizer, same signal at same frequency fed on one side and other side but at different phase, stylus detects phase angle across tablet: resistor ladder to apply phase shift on each conductor. Vertical and horizontal elements of grid/matrix operate at different frequencies simultaneously, so X and Y determined simultaneously.
Stylus design of capacitive/electrostatic grid digitizer, tablet is the transmitter. Cord on stylus.
Cited in Teh89.
Cited in Teh89.
Cited in Teh89.
Cited in Teh89.
Background history of Chinese/Japanese Kanji, Katakana, hiragana: educated Japanese cannot recognize Kana scripts.
Overview of protection mechanisms in Multics and file system: ACL access control lists, list of user/project/compartment (extendible) and access rights, associated with object. (Capability system would associate with user.) Description of "*" wildcard/don't-care syntax.
Overview of protection mechanisms in secure computer systems: ACLs access control lists embedded in indirect objects -- separate from object for which permission needed, separate access controller granting access, looks for match. Mentions encipherment/encryption (notes that security classifications from military restrict publication of ideas): authentication by handwriting signature verification or rhythm of keyboard typing (biometrics), physical tokens (unduplicatable identity card/token). Multi-node networked systems.
IEEE journal version of 1974 ACM paper. Overview of protection mechanisms in secure computer systems: ACLs access control lists embedded in indirect objects -- separate from object for which permission needed, separate access controller granting access, looks for match. Mentions encipherment/encryption (notes that security classifications from military restrict publication of ideas): authentication by fingerprint reader, handwriting signature verification or rhythm of keyboard typing (a.k.a. keystroke dynamics) (biometrics), physical tokens (unduplicatable identity card/token). Multi-node networked systems. Quote: "Work in progress is not well represented by published literature.".
79% handwriting recognition on 84 cursive script words (OCR).
Does cursive script recognition as part of handwriting recognition (OCR).
Statistical recognition (digrams) can make recognition worse: "fo" more frequent than "fa" in English, so ambiguous "far" would always be recognized wrong as "for".
90% per character recognition accuracy gives 81% per pair, 72% per triplet, etc.: fix with context.
Used handwriting data samples from Bell Telephone Labs (OCR).
Cites very early work by Harmon, internal Bell Telephone Labs reports.
Cited in Tappert88b: on segmentation combined with character recognition.
OCR scanner for handwritten characters: hardware to detect inflection points, endpoints, sub-stroke (horizontal and vertical segments) of strokes.
Describes two-level recognizer with (generic) first-level and detail-level feature analysis.
Standard physics reference and textbook.
Confusion data matrix on phoneme recognition/discrimination in speech recognition: degree of similarity, confusibility, association, psychological "proximity": how to find out a distance metric for pattern recognition. proximity analysis, signal-to-noise ratio in pattern space, multi-dimensional scaling.
functional attributes reliable (e.g. tail) on block character recognition: graphical context biases the attribute equally for all characters. Did not include context where only certain characters with the attribute were shifted?
General references on character recognition, especially cognitive psychology aspects.
Includes references on insight through study of character formation / generative variability.
Includes references on psychological descriptions of characters.
Brief user study of apparent line shortening for wider/narrow characters: graphical context on character recognition.
Paper copy only. Lists features that are not perceptually significant. Shillman's PhD Thesis: also published as BlesserB73.
OCR for discrete boxed handwriting character recognition of digits with cost and accuracy of keypunch typing transcription manually.
Rotationally invariant (at 90 degree increments) recognition of simple geometric shapes, apparently by learning shapes in each orientation: not specific to character recognition.
Cited in Baird84: for template matching in feature space using minimum distance?
Trainable system using subsets of curved line segments as the 'branch feature set'.
Template matching vs feature analysis is a continuum, not disjoint.
Cites several papers on low-level feature extraction in human perception.
Argument against dynamic matching: "theoretical and empirical reasons for feature extraction".
Cites different recognition mechanism for handwritten and printed characters.
Cites no real consensus on actual features humans use in recognizing letters (which is why functional attribute tests tell you what features).
Cited in Blatt88: 97.8% character recognition rate.
Biometric authentication of users based on typing pattern while entering standard entry code (password/PIN): time pattern and key pressure (force) characteristics. See also Saltzer 1975. Compares to telegraph key signal ("Fist").
Capacitive sensor, used to detect presence of water (or similar fluid) on the other side of a car windshield, and turn on the windshield wipers. Distinguishes between different materials, e.g. does not respond to debris on the windshield.
512x512 cross-wire (capacitive?) touch panel digitizer on CRT display, menu buttons arranged 10 x 13. Refers to "panel" as term for hierarchy of menus.
OCR recognition of handwriting affected by scanner, lighting, pencil type. Handwritten data collection done in the field, where computer not available for data entry.
Study of legibility of 30 different handwriting styles taught in North America. Block printing more legible than cursive (in general).
Refers to unpublished treatise on handwriting style variations in North America.
Clipping of drawn lines to a viewport window by binary division of the lines to edges of window.
multiple-pen user input on large-area tablet digitizer , three dimensional 3D input by providing input for multiple perspective views simultaneously. Compare with Buxton multi-hand/two-hand input, multi-touch.
General essay on human pattern recognition. Visual recognition of objects: segmentation of images, grouping of features, work for varying rotation and orientation. Cites multiple studies that recognition is by syllable rather than by phoneme.
Cited in DosterW77 on spelling dictionary.
Elastic matching applied to general strings of characters/codes: sped up by assuming initial part of pattern to be matched is not garbled, bracketing occurences of that sequence in input string.
Data-reduction in sketch recognition. Corners are important feature, detected better when taking into account drawing speed -- jitter when drawing at low speed badly affected line-straightening. Proposal to use pressure: slowly drawn or heavy pressure means user input is more deliberate. Sylvania data tablet, transparent tablet/touchscreen, two distance levels for proximity sensing. Storage-tube display (hard to erase, compared to drawing with physical ink). Replay sketching data at same timing as input -- compare with interactive system such as Wang Freestyle? Proposal for graphical compiler to process (recognize) sketches: compare to 2D mathematical input?
diode-based DC touch panel / tablet for teaching machine, operates as cross-bar switch, can sense two distinct contact (multi-touch). Compare with Kaplow switching matrix for touchscreen?
Recognition of Japanese/Chinese character using chain codes (direction codes) for first level classification of fundamental strokes.
proof-reading symbol recognition, writing directly over the strokes of a character on a tablet.
character recognition input in a special area of the tablet.
Cited in Sklarew99 prior art.
General UNIX reference.
Two-level handwriting recognition (off-line/OCR?). Two stages: first level classifies for similar topological configurations, second stage local analysis for the particular category. Cited in Goodale83.
Mathematical treatment of using confusion matrix with three different recognition algorithms: choice model, and finite sensory state models. Suggest that better character recognition will need a model of specific distinguishing features: compare to Pencept?
Resistive-film digitizer, ratiometric measurement, mounted on front of CRT display.
Cited in Goshtasby88.
using edge directions for recognizing hand-printed character if the continuity information between characters is utilized (stroke hooks and continuation marks?).
Copy on file has best image of RAND tablet and display hardware: shows photographic image and electronic text/ink displayed on same surface.
Multiple users on one graphics system: 32 consoles, each with up to 8 different input devices.
Mentions development of all-digital video.
Compare with later Schumer patent.
Describes a tablet-based text editor from Carnegie-Mellon University (Coleman69), using symbols/gestures on a RAND tablet: decision tree to recognize symbols. Ease of correction makes requirement for perfect recognition be less. Cost of computers made handwriting recognition not cost-effective in 1971.
Capacitive matrix keyboard, similar to capacitive touchscreen. Physical keys move bridging conductor near capacitive elements.
Analysis of line-drawing images using combinatorial algorithms: compare permutation-free search with chain codes in handwriting recognition and structural grammars in sketch recognition.
Tutorial system for German grammar doing automatic grammatical analysis according to the two-prong word-order grammar of Lohnes and Strothmanm.
Thesis (?) on proposed low-cost handwriting recognition product? Voltage-gradient (resistive sheet) tablet. Lists stroke directions for on-line characters.
Essays on the role of personality and real-world-practice versus methodology-theory in software development. Anniversary edition 25 years later compared state of progress with that of 1971 publication.
Aid for the blind: hand-held wand with optical sensor, device generates different pitch tones to indicate sensor reading. Slit camera (10 sensors vertical) user scans across printed material, generates tones for each sensor. Compare to Optacon.
Two-dimensional programming languages: textual, tabular, and pictorial (flow diagrams, etc.), Refers to keyboard languages if typed, pen language if handwritten or tablet language (light pen, acoustical pen, RAND tablet with stylus, mouse, joystick, trackball, etc.) MadCap early two-dimensional pen language. Cites to Grail language (Ellis), with scrubbing motion (gesture) to erase. Scratchpad, Ambit/L mathematical language.
SAC Science Accessories Corporation acoustic stylus for sonic digitizer: generated shock energy sound wave by spark.
Summagraphics original design of tablet for character recognition, zone based, numerics of single-strokes and double-strokes.
Summagraphics magnetostrictive tablet, using sheet of magnetostrictive material.
Cited in Burr81.
pattern recognition using a template (e.g. shape/classifier of chromosomes in images) by allowing non-exactly-matching signatures: elastic matching on handwriting (?) recognition.
User writes in two dimensions, compiler translates to linear form (shown at top of display) an evaluates. Characters are translated into recognized characters in two-dimensional form for editing in real time. Uses bounding rectangles in parsing. Gesture recognition described, handwriting recognition is in a separate publication. "Scrub mode" is a scratch-out gesture. Arrow gesture indicates move/edit. Horizontal line gesture for "open space/move" editing operation. Double horizontal line for close up. Dot (tap) plus second line to move a character. Two perpendicular stroke gesture for select/move group.
The Assistant Mathematician (TAM), two-dimensional hand-printed mathematical input for computer programming: visual programming language VPL. Transparent/translucent digitizer tablet with rear-projected CRT display. Mathematic expression is displayed as recognized by system for confirmation by user in semantic context: semantic ambiguity of mathematical language (e.g. "i" as SQRT(-1), variable, or number). Scrubbing (rubout) gesture, editing of electronic ink.
Cited by ClarkJH76 as 3D digitizer.
User's guide for AMBIT/L visual programming language / digrammatic language for list processing. Refers to Drawing of AMBIT/L Programs and Their Encodement. See AMBIT/G for lightpen/tablet user interface.
INTERCAL: The world's most bizarre, yet clearly defined, programming language (Geek humor). Compare with spoof on Basic at Data General, 1974?
Product announcement on SRI/Xebec handwriting character recognition using accelerometer pen stylus. See Hewitt Crane.
Cited in WangPSP85.
32x32 and 8x8 pixel bit-map character recognizer for OCR of Chinese.
Mathematical model for pen dynamics in script, but not print or stroke connection.
40-key special keyboard for input of Chinese Kanji characters, keys used to input shape of the character.
Biswas81 cites on fuzzy sets for recognition. Focus is on linguistic variables: e.g. height is a variable with possible values short, tall, very tall, extremely tall, medium, etc. A calculus of fuzzy sets instead of a precise calculus of numbers.
Biswas81 cites on fuzzy sets for recognition. Compendium / graduate text of papers on fuzzy sets and fuzzy logic. Specific applications include handwriting recognition, fuzzy grammars and pattern recognition: compare with chain codes and parsing rules for sketch recognition.
Tanaka82 on fuzzy sets. Linguistic variable is variable who value is (overlapping/qualitative) set membership: e.g. age: young, not young, old, very old vs. 21, 22, 23. Applications to pattern recognition.
Post-processing of a distribution of features points in 2D Bayesian space, attempt to cover "noise" from translation, rotation, and noise displacement of the individual points. Cited in Baird84. Only tested with small number of artificial samples.
Dynamic description of character drawing and its use for on-line recognition of handwritten characters (in German).
zone and chain-code recognition for handwriting/hand-printing of symbols. Based on Doctoral thesis, TU Aachen 1972. Describes use of direction information in recognition, light pen on display.
biblio70 (Prev page)
(Next page) biblio80
This compilation and all annotations are copyright © Jean Renard Ward, 1993, 2004, 2011, 2013, 2016.
I can be contacted at firstname.lastname@example.org.