Showing papers on "Handwriting recognition published in 1995"

Real-time American Sign Language recognition from video using hidden Markov models

Abstract: Hidden Markov models (HMMs) have been used prominently and successfully in speech recognition and, more recently, in handwriting recognition. Consequently, they seem ideal for visual recognition of complex, structured hand gestures such as are found in sign language. We describe a real-time HMM-based system for recognizing sentence level American Sign Language (ASL) which attains a word accuracy of 99.2% without explicitly modeling the fingers.

Automatic handwriting recognition using both static and dynamic parameters

Abstract: Methods and apparatus are disclosed for recognizing handwritten characters in response to an input signal from a handwriting transducer. A feature extraction and reduction procedure is disclosed that relies on static or shape information, wherein the temporal order in which points are captured by an electronic tablet may be disregarded. A method of the invention generates and processes the tablet data with three independent sets of feature vectors which encode the shape information of the input character information. These feature vectors include horizontal (x-axis) and vertical (y-axis) slices of a bit-mapped image of the input character data, and an additional feature vector to encode an absolute y-axis displacement from a baseline of the bit-mapped image. It is shown that the recognition errors that result from the spatial or static processing are quite different from those resulting from temporal or dynamic processing. Furthermore, it is shown that these differences complement one another. As a result, a combination of these two sources of feature vector information provides a substantial reduction in an overall recognition error rate. Methods to combine probability scores from dynamic and the static character models are also disclosed.

A Hough based algorithm for extracting text lines in handwritten documents

Abstract: The method herein proposed detects text lines on handwritten pages which may include either lines oriented in several directions, erasures, or annotations between main lines. The method has a hypothesis-validation strategy which is iteratively activated until the end of the segmentation is reached. At each stage of the process, the best text-line hypothesis is generated in the Hough domain. Taking into account the fluctuations of the text-line components. Afterwards, the validity of the line is checked in the image domain using a proximity criteria which analyses the context in which is perceived the alignment hypothesized. Ambiguous components belonging to several text lines are also marked.

LeRec: a NN/HMM hybrid for on-line handwriting recognition

Abstract: We introduce a new approach for on-line recognition of handwritten words written in unconstrained mixed style. The preprocessor performs a word-level normalization by fitting a model of the word structure using the EM algorithm. Words are then coded into low resolution "annotated images" where each pixel contains information about trajectory direction and curvature. The recognizer is a convolution network that can be spatially replicated. From the network output, a hidden Markov model produces word scores. The entire system is globally trained to minimize word-level errors.

Method and apparatus for recognizing and performing handwritten calculations

Abstract: A pen-based calculator recognizes handwritten input. The calculator comprises a display simulating a sheet of paper, and a stylus simulating a pen. The user writes a calculation on the calculator as if it were a piece of scratch paper. The calculator uses handwriting recognition to identify the various elements of the calculation, performs the calculation, and then displays the result at an appropriate location.

Variable duration hidden markov model and morphological segmentation for handwritten word recognition

Abstract: This paper describes a complete system for the recognition of unconstrained handwritten words using a continuous density variable duration hidden Markov model (CDVDHMM). First, a new segmentation algorithm based on mathematical morphology is developed to translate the 2-D image into a 1-D sequence of subcharacter symbols. This sequence of symbols is modeled by the CDVDHMM. Thirty-five features are selected to represent the character symbols in the feature space. Generally, there are two information sources associated with written text-the shape information and the linguistic knowledge. While the shape information of each character symbol is modeled as a mixture Gaussian distribution, the linguistic knowledge, i.e., constraint, is modeled as a Markov chain. The variable duration state is used to take care of the segmentation ambiguity among the consecutive characters. A modified Viterbi algorithm, which provides 2 globally best paths, is adapted to VDHMM by incorporating the duration probabilities for the variable duration state sequence. The general string editing method is used at the postprocessing stage. The detailed experiments are carried out for two postal applications; and successful recognition results are reported.

Method and apparatus for recognizing handwritten inputs in a computerized teaching system

Abstract: A pen-based teaching system recognizes handwritten input. The system comprises a display simulating a sheet of paper, and a stylus simulating a pen. The system uses handwriting recognition to interact with the student, so that the student enters input into the system as though it were a piece of scratch paper.

Combined dictionary based and likely character string method of handwriting recognition and display

Abstract: In a handwriting recognition process, a list of candidate recognized words is identified (202) as a function of both comparison of dictionary entries to various combinations of recognized character combinations, and through a most likely character string and most likely string of digits analysis as developed without reference to the dictionary. The process selects (301) a word from the list and presents (302) this word to the user. The user then has the option of displaying (303) this list. When displaying the list, candidate words developed with reference to the dictionary are displayed in segregated manner from the most likely character string words and the most likely string of digits. The user can charge the selected word by choosing from the list, or edit the selected word. When the user selects the most likely character string as the correct representation of the handwritten input to be recognized, the process automatically updates (310) the dictionary to include the most likely character string The same process can occur when the user selects the most likely string of digits.

Pen computing: a technology overview and a vision

Abstract: This work gives an overview of a new technology that is attracting growing interest in public as well as in the computer industry itself. The visible difference from other technologies is in the use of a pen or pencil as the primary means of interaction between a user and a machine, picking up the familiar pen and paper interface metaphor. From this follows a set of consequences that will be analyzed and put into context with other emerging technologies and visions.Starting with a short historical background and the technical advances that begin making Pen Computing a reality, the new paradigms created by Pen Computing will be explained and discussed. Handwriting recognition, mobility and global information access are other central topics. This is followed by a categorization and an overview of current and future systems using pens as their primary user interface component.

Handwritten character input device allowing input of handwritten characters to arbitrary application program

Abstract: Based on handwriting coordinate data input by a pen at an arbitrary position on a screen, a handwriting display portion displays a handwriting at the writing position. A recognition portion recognizes a written character or an editing symbol. A handwriting erase portion erases the handwriting, and returns the screen to a state before writing. A recognition result output portion supplies the recognition result to an application program having a keyboard focus in the same form as that of keyboard input. As a result, a character or the like can be directly written at an arbitrary position on a window displayed by the application program or on the screen, and input or editing can be carried out.

Real-time on-line unconstrained handwriting recognition using statistical methods

Abstract: We address the problem of automatic recognition of unconstrained handwritten text. Statistical methods, such as hidden Markov models (HMMs) have been used successfully for speech recognition and they have been applied to the problem of handwriting recognition as well. We discuss a general recognition system for large vocabulary, writer independent, unconstrained handwritten text. "Unconstrained" implies that the user may write in any style e.g. printed, cursive or in any combination of styles. This is more representative of typical handwritten text where one seldom encounters purely printed or purely cursive forms. Furthermore, a key characteristic of the system is that it performs recognition in real-time on 486 class PC platforms without the large amounts of memory required for traditional HMM based systems. We focus mainly on the writer independent task. Some initial writer dependent results are also reported. An error rate of 18.9% is achieved for a writer-independent 21,000 word vocabulary task in the absence of any language models.

Gap metrics for word separation in handwritten lines

Abstract: The problem of separating words in a handwritten line is made difficult by the presence of non-uniform spacing between words and between characters within a word. A central sub-problem in word separation is the estimation of gaps between adjacent components in a line. We present a new technique to estimate inter-component distances that is based on the gap between their convex hulls. The technique evolved through a study of the drawbacks in previous approaches to gap estimation, and is shown to be better in terms of performance and robustness.

A comparison of approaches to on-line handwritten character recognition

Abstract: Speech and handwriting are manifestations of a common need for linguistic communication. The similar nature of speech and handwriting recognition problems suggests that a largely shared solution may be possible. Recent advances in speech recognition can be partly attributed to changes in the research paradigm. These changes include using large corpora of common training and testing data, adopting statistical modeling over rule-based approaches, and ensuring meaningful comparisons between candidate technologies. The resulting improvements in system performance and robustness permit the study of increasingly difficult recognition tasks. The primary goal of my thesis is to compare handwriting representations for on-line, printed, alphanumeric character recognition without striving to construct the highest-performance system. My studies are based on a carefully collected body of data containing some 87,000 characters from 150 writers. Material was selected automatically to ensure compact coverage of significant letter sequences. Subjects were instructed and prompted so as to minimally influence the writing they produced. A time-aligned transcription was entered for all of this data. I conducted an authentication study to understand better the classification difficulty of this writing. Only 81.7% of testing characters were identified correctly. I examined a number of potential representations for handwriting classification including bitmaps, projections, transforms, chain codes, and point-sampling, paying particular attention to pen motion as an information source. All experiments were based on Gaussian mixture models because of their flexibility. The best representation features Cartesian coordinates of 10 equally-spaced samples along the pen trajectory. Without the benefit of relative size information, this representation resulted in 77.2% correct character classification on testing data. Finally, I adapted the scSUMMIT segment-based speech recognition system developed at MIT to handwriting. Segmentation is based primarily on pen-lifts, but strokes are divided to account for connected character pairs. The parameter described above is computed for each segment and the resulting graph passed to the recognition engine for classification and search. This system was able to correctly recognize 65.1% of the test-set characters. Incorporating a bigram character grammar with perplexity 11.3 improved this performance to 76.4%. (Copies available exclusively from MIT Libraries, Rm. 14-0551, Cambridge, MA 02139-4307. Ph. 617-253-5668; Fax 617-253-1690.)

Design of a linguistic postprocessor using variable memory length Markov models

Abstract: We describe a linguistic postprocessor for character recognizers. The central module of our system is a trainable variable memory length Markov model (VLMM) that predicts the next character given a variable length window of past characters. The overall system is composed of several finite state automata, including the main VLMM and a proper noun VLMM. The best model reported in the literature (Brown et al., 1992) achieves 1.75 bits per character on the Brown corpus. On that same corpus, our model, trained on 10 times less data, reaches 2.19 bits per character and is 200 times smaller (/spl sime/160,000 parameters). The model was designed for handwriting recognition applications but could also be used for other OCR problems and speech recognition.

A set of handwriting families: style recognition

Abstract: In this paper, we analyse the variability of handwritings. The aim is to determine what sort of observations gives a first degree of handwriting characterization before initiating a text recognition process. In the case of handwriting consisting of few words, such literal amounts on cheques, this first degree of characterization can be obtained for each word, independent of signification, by extracting the measures of some pertinent observations. Outcomes of this characterization are, to a certain extent, a distinction between significants which characterise the author and signification which is the semantic aspect. Based on an analysis of 980 different handwritten amounts, it is shown that these measures define a variability space of non-uniform density. A fuzzy partition of the set of 3788 words of the database is proposed which allows to regroup handwriting styles into a small number of specific families.

A system for machine-written and hand-written character distinction

Abstract: In applications of character recognition where machine-printed and hand-written characters are involved, it is important to know if the character image, or the whole word, is machine- or hand-written. This is due to the accuracy difference between the algorithms and systems oriented to machine- or handwritten characters. Obviously, this type of knowledge leads to the increase of the overall system quality. In this work a classification system is presented which reads a raster image of a character and outputs two confidence values, one for "machine-written" and one for "hand-written" character classes, respectively. The proposed system features a preprocessing step, which transforms a general uncentered character image into a normalized form, then the feature extraction phase extracts relevant information from the image, and at the end, a standard classifier based on a feedforward neural network creates the final response. At the end, some results on a proprietary image database are reported.

Cursive script recognition applied to the processing of bank cheques

Abstract: A method for recognizing unconstrained handwritten words belonging to a small static lexicon is proposed. Our computational theory is based on a psychological model of the reading process of a fast reader. The method we propose is global in its nature and avoid the difficult segmentation stage of common word recognition techniques. Our computational theory has been applied to the processing of handwritten bank cheques, whose problem domain is that of unconstrained handwriting, unlimited writers in a small static lexicon. Current results seem comparable to those published in the literature and support our computational theory.

An automatic reading system for handwritten numeral amounts on French checks

Abstract: We present an automatic recognition system applied to handwritten numeral check amounts. This system is based on a segmentation-by-recognition probabilistic model. The application is described from the field amount localization to the hypothesis generation of amounts. An explicit segmentation algorithm determines cut regions on digit links and provides a multiple spatial representation. The best path for the segmentation is determined by the combination of the recognition scores, segmentation weights and the outputs of a probabilistic parser. Training is done by a bootstrapping technique, which significantly improves the performances of the different algorithms. It also allows the use of a reject class at the recognition step. The system was evaluated on 10000 database images to show its robustness.

Segmentation and recognition of symbols within handwritten mathematical expressions

Abstract: An efficient on-line recognition system for symbols within handwritten mathematical expressions is proposed. The system is based on the generation of a symbol hypotheses net and the classification of the elements within the net. The final classification is done by calculating the most probable path through the net under regard of the stroke group probabilities and the probabilities obtained by the symbol recognizer based on hidden Markov models.

NPen/sup ++/: a writer independent, large vocabulary on-line cursive handwriting recognition system

Abstract: In this paper we describe the NPen/sup ++/ system for writer independent on-line handwriting recognition. This recognizer needs no training for a particular writer and can recognize any common writing style (cursive, hand-printed, or a mixture of both). The neural network architecture, which was originally proposed for continuous speech recognition tasks, and the preprocessing techniques of NPen/sup ++/ are designed to make heavy use of the dynamic writing information, i.e. the temporal sequence of data points recorded on an LCD tablet or digitizer. We present results for the writer independent recognition of isolated words. Tested on different dictionary sizes from 1,000 up to 100,000 words, recognition rates range from 98.0% for the 1,000 word dictionary to 91.4% on a 20,000 word dictionary and 82.9% for the 100,000 word dictionary. No language models are used to achieve these results.

A soft-decision approach for structural analysis of handwritten mathematical expressions

Abstract: An efficient system for structural analysis of handwritten mathematical expressions is proposed. To handle the problems caused by handwriting, this system is based on a soft-decision approach. This means that alternatives for the solution are generated during the analysis process if the relation between two symbols within the expression is ambiguous. Finally a string containing the mathematical information is generated and syntactical verified for each alternative. Strings failing this verification are considered as invalid.

Radical definition and dictionary creation for a handwriting recognition system

Abstract: The system described herein automatically defines a set of radicals to be used in a Kanji character handwriting recognition system and automatically creates a dictionary of the Kanji characters that are recognized by the system. In performing its functionality, the system described herein first obtains representative handwriting samples for each Kanji character that is to be recognized by the system. The system described herein then evaluates the samples to identify a set of subparts ("radicals") that are common to at least two of the Kanji characters. These radicals represent component roots from which the characters are formed. Each Kanji character is formed by one or more of these radicals. The radicals that are identified by the system described herein are not constrained to any preset definition (e.g., the traditional set of radicals used to organize Japanese dictionaries). Thus, the radicals utilized by the system described herein may include some of the traditional radicals or may include none of the traditional radicals. After identifying the set of radicals, the system described herein generates a dictionary with a mapping of each Kanji character that is to be recognized by the system to its component radicals. After the set of radicals and the dictionary have been created, these components can be utilized during handwriting recognition. When performing handwriting recognition, the system described herein identifies the radicals within the handwriting and then uses the mapping to determine which Kanji character the handwriting most closely matches.

Name and Address Block Reader system for tax form processing

Abstract: The reading of names and addresses is one of the most complex tasks in automated forms processing. The paper describes an integrated real time system to read names and addresses on tax forms of the Internal Revenue Service of the United States. The Name and Address Block Reader (NABR) system accepts both machine printed and hand printed address block images as input. The application software has two major steps: document analysis (connected component analysis, address block extraction, label detection, hand print/machine print discrimination); and document recognition. Document recognition has two non identical streams for machine print and hand print; key steps are: address parsing, character recognition, word recognition and postal database lookup (ZIP+4 and City-State-ZIP files). Real time throughput (8,500 forms per hour) is achieved by employing a loosely coupled multiprocessing architecture. The functional architecture, software design, system architecture and hardware implementation are described. Performance evaluation on machine printed and handwritten addresses are presented.

Forward-backward retraining of recurrent neural networks

Abstract: This paper describes the training of a recurrent neural network as the letter posterior probability estimator for a hidden Markov model, off-line handwriting recognition system. The network estimates posterior distributions for each of a series of frames representing sections of a handwritten word. The supervised training algorithm, backpropagation through time, requires target outputs to be provided for each frame. Three methods for deriving these targets are presented. A novel method based upon the forward-backward algorithm is found to result in the recognizer with the lowest error rate.

A fuzzy logic system for the detection and recognition of handwritten street numbers

Abstract: Fuzzy logic is applied to the problem of locating and reading street numbers in digital images of handwritten mail. A fuzzy rule-based system is defined that uses uncertain information provided by image processing and neural network-based character recognition modules to generate multiple hypotheses with associated confidence values for the location of the street number in an image of a handwritten address. The results of a blind test of the resultant system are presented to demonstrate the value of this new approach. The results are compared to those obtained using a neural network trained with backpropagation. The fuzzy logic system achieved higher performance rates. >

Recognition of handwritten Persian/Arabic numerals by shadow coding and an edited probabilistic neural network

Abstract: A system for recognition of segmented handwritten Persian/Arabic numerals irrespective of size and translation is developed. The image is represented by invariant features obtained from a new shadow coding scheme designed for the considered shapes. Classification is performed by a modified version of a four-layer probabilistic neural network (PNN) called the edited PNN (EPNN). Due to an editing and condensation procedure on the training samples, the EPNN has better performance and the network size is smaller. The performance of the system is evaluated on a database consisting of 2600 digits written by 10 different people. The obtained recognition accuracy is 97.8 percent. The developed system can process approximately two digits per second on a Intel 486 based PC with a 66 MHz clock.

The handwritten trie: indexing electronic ink

Abstract: The emergence of the pen as the main interface device for personal digital assistants and pen-computers has made handwritten text, and more generally ink, a first-class object. As for any other type of data, the need of retrieval is a prevailing one. Retrieval of handwritten text is more difficult than that of conventional data since it is necessary to identify a handwritten word given slightly different variations in its shape. The current way of addressing this is by using handwriting recognition, which is prone to errors and limits the expressiveness of ink. Alternatively, one can retrieve from the database handwritten words that are similar to a query handwritten word using techniques borrowed from pattern and speech recognition. In particular, Hidden Markov Models (HMM) can be used as representatives of the handwritten words in the database. However, using HMM techniques to match the input against every item in the database (sequential searching) is unacceptably slow and does not scale up for large ink databases. In this paper, an indexing technique based on HMMs is proposed. The new index is a variation of the trie data structure that uses HMMs and a new search algorithm to provide approximate matching. Each node in the tree contains handwritten letters, where each letter is represented by an HMM. Branching in the trie is based on the ranking of matches given by the HMMs. The new search algorithm is parametrized so that it provides means for controlling the matching quality of the search process via a time-based budget. The index dramatically improves the search time in a database of handwritten words. Due to the variety of platforms for which this work is aimed, ranging from personal digital assistants to desktop computers, we implemented both main-memory and disk-based systems. The implementations are reported in this paper, along with performance results that show the practicality of the technique under a variety of conditions.

Handwritten word recognition for real-time applications

Abstract: A fast handwritten word recognition system for real time applications is presented. Preprocessing, segmentation and feature extraction are implemented using chain code representation. Dynamic matching between each character of a lexicon entry and segment(s) of input word image is used for ranking words in the lexicon. Speed of the entire recognition process is about 200 msec on a single SPARC-10 platform for lexicon size of 10. A top choice performance of 96% is achieved on a database of postal words captured at 212 dpi.

Graph-based handwritten digit string recognition

Abstract: The article presents a handwritten digit string recognition algorithm based on matching input subgraphs with prototype symbol graphs. The article defines a set of acceptable graph transformations corresponding to typical variations of the handwritten symbols. The search for a match between the input subgraph and prototype graph is conducted using this set of transformations. This approach allows us to solve the problems of structure recognition methods caused by a high variability of handwritten symbol topology. The article presents experimental results of the handwritten digit string recognition system.

A simplified attributed graph grammar for high-level music recognition

Abstract: This paper describes a simplified attributed programmed graph grammar to represent and process a-priori knowledge about common music notation. The presented approach serves as a high-level recognition stage and is interlocked to previous low-level recognition phases in our entire optical music recognition system (DOREMIDI++). The implemented grammar rules and control diagrams describe a declarative knowledge base to drive a transformation algorithm. This transformation converts the results of symbol recognition stages to a symbolic representation of the musical score.