Showing papers on "Intelligent word recognition published in 1998"

The advantage of using an HMM-based approach for faxed word recognition

Abstract: A method for word recognition based on the use of hidden Markov models (HMMs) is described. An evaluation of its performance is presented using a test set of real printed documents that have been subjected to severe photocopy and fax transmission distortions. A comparison with a commercial OCR package highlights the inherent advantages of a segmentation-free recognition strategy when the word images are severely distorted, as well as the importance of using contextual knowledge. The HMM method makes only one quarter of the number of word errors made by the commercial package when tested on word images taken from faxed pages. © 1998 Springer-Verlag Berlin Heidelberg.

Real-time probabilistic segmentation for segment-based speech recognition.

Abstract: In this work, we investigate modifications to a probabilistic segmentation algorithm to achieve a real-time, and pipelined capability for our segment-based speech recognizer [4]. The existing algorithm used a Viterbi and backwards A search to hypothesize phonetic segments [2]. We were able to reduce the computational requirements of this algorithm by reducing the effective search space to acoustic landmarks, and were able to achieve pipelined capability by executing theA search in blocks defined by reliably detected phonetic boundaries. The new algorithm produces 30% fewer segments, and improves TIMIT phonetic recognition performance by 2.4% over an acoustic segmentation baseline. We were also able to produce 30% fewer segments on a word recognition task in a weather information domain [11].

Handwritten chinese character recognition with directional decomposition cellular features

Abstract: A new feature extraction approach based on elastic meshing and directional decomposition techniques for handwritten Chinese character recognition (HCCR) is proposed in this letter. It is found that decomposing a Chinese character into horizontal, vertical stroke, left slant and right slant directional sub-patterns is very helpful for feature extraction and recognition. Three kinds of decomposition methods are proposed. A minimum distance classifier is trained by 3755 categories of characters using the new features. Testing on a total of 37,550 untrained handwritten samples produces the recognition rate of 92.36%, showing the effectiveness of the proposed approach.

Method and apparatus for on-line handwritten input character recognition and recording medium for executing the method

Abstract: The invention aims at recognition of a handwritten input character on-line with quite high accuracy. The method of on-line handwritten input character recognition of the invention is characterized in that ordinary strokes and transition strokes and/or start-end (s-e) strokes of a handwritten input character sampled on-line are compared with ordinary strokes and transition strokes and/or s-e strokes of dictionary's characters previously registered in a dictionary and the character corresponding to the handwritten input character is recognized on the basis of the result of the comparison. Otherwise, when the dictionary's character most similar to the input character corresponds to a preset character, the handwritten input character is identified by using characteristic features of the corresponding character. Thus, by comparing ordinary strokes and transition strokes and/or s-e strokes of a handwritten input character sampled on-line with ordinary strokes and transition strokes and/or s-e strokes of dictionary's characters previously registered in a dictionary and recognizing the character corresponding to the handwritten input character on the basis of the result of the comparison, a handwritten input character which was difficult to recognize only by means of ordinary strokes can be recognized with high accuracy.

Automatic detection of italic, bold and all-capital words in document images

Abstract: We propose simple and fast algorithms for detection of italic, bold and all-capital words without doing actual character recognition. We present a statistical study which reveals that the detection of such words may play a key role in automatic information retrieval from documents. Moreover, detection of italic words can be used to improve the recognition accuracy of a text recognition system. Considerable number of document images have been tested and our algorithms give accurate results on all the tested images, and the algorithms are very easy to implement.

Elastic structural matching for on-line handwritten alphanumeric character recognition

Abstract: In this paper, we will propose a simple yet robust structural approach for recognizing on-line handwriting. Our approach is designed to achieve reasonable speed, fairly high accuracy and sufficient tolerance to variations. Experimental results show that the recognition rates are 98.60% for digits, 98.49% for uppercase letters, 97.44% for lowercase letters, and 97.40% for the combined set. When the rejected cases are excluded from the calculation, the rates can be increased to 99.93%, 99.53%, 98.55% and 98.07%, respectively. On the average, the recognition speed is about 7.5 characters per second running in Prolog on a Sun SPARC 10 Unix workstation and the memory requirement is reasonably low.

Handwritten character input device and storage medium

Abstract: PROBLEM TO BE SOLVED: To display the operations desired by a user in the smooth and easy-to- understand ways and to improve the operability of a handwritten character input device by attaining the free input operations of handwritten characters with no input frames required and also displaying a menu window etc., according to the characters, instructions, etc., which are inputted via a touch panel. SOLUTION: When the handwritten characters are inputted by a user at an optional position on an input panel, a CPU displays a transparent blue vertical or horizontal writing recognition window 612 in response to the direction of the inputted handwritten characters. The window 612 functions to change the recognition result of the inputted handwritten characters and also to convert the recognition result into KANJI (Chinese character) from KANA (Japanese syllabary) in a character string. Then ○ or × is inputted by handwriting to decide or cancel the recognition result or the converted character string candidates.

A recognition-based Arabic optical character recognition system

Abstract: Optical character recognition systems improve human-machine interaction and are widely used in many government and commercial departments. After forty years of intensive research, OCR systems for most scripts are well developed. However, not for Arabic script. Since Arabic is a popular script, Arabic OCR systems should have great commercial value. Thus a recognition-based Arabic OCR system is proposed in this paper. It consists of the image acquisition, preprocessing, segmentation, character fragmentation, combination of character fragments, feature extraction, and classification. A signal is fed back to improve and determine the segmentation/recognition result. The system has been implemented and it has 90% recognition accuracy with a 20 chars/sec recognition rate.

Multilinguistic handwritten character recognition by Bayesian decision-based neural networks

Abstract: In this paper, we present a Bayesian decision-based neural network (BDNN) for multilinguistic handwritten character recognition. The proposed self-growing probabilistic decision-based neural network (SPDNN) adopts a hierarchical network structure with nonlinear basis functions and a competitive credit-assignment scheme. Our prototype system demonstrates a successful utilization of SPDNN to the handwriting of Chinese and alphanumeric character recognition on both public databases (CCL/HCCR1 for Chinese and CEDAR for the alphanumerics) and in-house database (NCTU/NNL). Regarding the performance, experiments on three different databases all demonstrated high recognition (86-94%) accuracy as well as low rejection/acceptance (6.7%) rates. As for the processing speed, the whole recognition process (including image preprocessing, feature extraction, and recognition) consumes approximately 0.27 s/character on a Pentium-100 based personal computer, without using a hardware accelerator or coprocessor.

Correcting English text using PPM models

Abstract: An essential component of many applications in natural language processing is a language modeler able to correct errors in the text being processed For optical character recognition (OCR), poor scanning quality or extraneous pixels in the image may cause one or more characters to be mis-recognized, while for spelling correction, two characters may be transposed, or a character may be inadvertently inserted or missed out, This paper describes a method for correcting English text using a PPM model A method that segments words in English text is introduced and is shown to be a significant improvement over previously used methods A similar technique is also applied as a post-processing stage after pages have been recognized by a state-of-the-art commercial OCR system We show that the accuracy of the OCR system can be increased from 963% to 969%, a decrease of about 14 errors per page

Automatic allograph selection and multiple expert classification for totally unconstrained handwritten character recognition

Abstract: We introduce a new method for online character recognition based on the co-operation of two classifiers, respectively operating on static and dynamic character properties. Both classifiers use the nearest neighbour algorithm. References have been selected previously using an unsupervised clustering technique for selecting, in each character class, the most representative allographs. Several co-operation architectures are presented, from the easier (balanced sum of both classifier outputs) types to the most complicated (integrating neural network) one. The recognition improvement varies between 30% and 50% according to the merging technique implemented. We evaluate the performance of each method based on the recognition rate and speed. Results are presented on 62 different character classes, and more than 75000 examples are from the UNIPEN database.

Japanese OCR Error Correction using Character Shape Similarity and Statistical Language Model

Abstract: We present a novel OCR error correction method for languages without word delimiters that have a large character set, such as Japanese and Chinese. It consists of a statistical OCR model, an approximate word matching method using character shape similarity, and a word segmentation algorithm using a statistical language model. By using a statistical OCR model and character shape similarity, the proposed error corrector outperforms the previously published method. When the baseline character recognition accuracy is 90%, it achieves 97.4% character recognition accuracy.

Amount translation and error localization in check processing using syntax-directed translation

Abstract: Introduces a method for the comparison of legal and courtesy amount recognized by OCR programs on bank checks. Based on the method of syntax-directed translation (SDT), the result of legal amount recognition is translated into a digit string that can be directly compared with the result of courtesy amount recognition. SDT turns out to be a systematic, simple and flexible technique for this step. It is especially useful for languages where the translation cannot be done in a straightforward way and many different word representations correspond to the same digit amount. The information provided by SDT in our system is used to infer the relation between each digit and the corresponding part of the word amount. Thus if an inconsistency between the recognized word and digit amount occurs, the mismatching parts within each amount can be localized. The method has been implemented and tested on real postal checks. It is part of a check reading system that is currently under development.

A new scheme for off-line handwritten connected digit recognition

Abstract: A scheme is proposed for off-line handwritten connected digit recognition, which uses a sequence of segmentation and recognition algorithms. First, the connected digits are segmented by employing both the gray scale and binary information. Then, a new set of features is extracted from the segments. The parameters of the feature set are adjusted during the training stage of the hidden Markov model (HMM) where the potential digits are recognized. Finally, in order to confirm the preliminary segmentation and recognition results, a recognition based segmentation method is presented.

A neural based segmentation and recognition technique for handwritten words

Abstract: Artificial neural networks (ANNs) have been successfully applied to optical character recognition (OCR) yielding excellent results. In this paper a technique is presented that segments difficult printed and cursive handwriting, and then classifies the segmented characters. A conventional algorithm is used for the initial segmentation of the words, while an ANN is used to verify whether an accurate segmentation point has been found. After all segmentation points have been detected another ANN is used to identify the characters which remain following the segmentation process. The C programming language, the SP2 supercomputer and a SUN workstation were used for the experiments. The technique has been tested on real-world handwriting scanned from various staff at Griffith University, Gold Coast. Some preliminary experimental results are presented in this paper.

Direction-change features of imaginary strokes for on-line handwriting character recognition

Abstract: We found suitable direction-change features of the imaginary strokes in the pen-up state for online handwritten cursive character recognition. Our method simultaneously uses both directional features, otherwise known as off-line features, and direction-change features, which we designed as on-line features. Direction-change features express both written strokes in the pen-down state and unwritten imaginary strokes in the pen-up state. It is important to get suitable direction-change features when using this method. We tried to examine the influence on character recognition rates when changing the functions used to get each direction-change feature based on the imaginary stroke lengths. Then, we found that the best function is the function which puts no weight on the imaginary stroke lengths. The recognition rate for freely-written Japanese characters was improved from 82.37% to 86.32% by our new method using the best function as opposed to our old method using a function which gets each direction change feature in inverse proportion to the imaginary stroke lengths.

A new scheme for off-line handwritten connected digit recognition

Abstract: We introduce a scheme for an off-line handwritten connected digit string recognition problem, which uses a sequence of segmentation and recognition algorithms. The proposed system assumes no constraint in writing style, size or variations. First, a segmentation method, which combines the gray scale and binary information, is proposed to find the nonlinear character segmentation paths. Each segment is then, recognized by a hidden Markov model. Finally, in order to confirm the segmentation paths and recognition results, a recognition based segmentation method is presented. The proposed scheme is tested on 4000 handwritten connected digits, collected from 16 different persons. The experiments yield 97.2% recognition rate.

Improving OCR accuracy through combination: a survey

Abstract: Optical character recognition is perhaps the most studied application of pattern recognition. Recent work has increased accuracy in two ways. Combination of individual classifier outputs overcomes deficiencies of features and trainability of single classifiers. OCR systems take page images as input and output strings of recognized characters. Due to character segmentation errors, characters can be split or merged preventing output combination character-by-character. Merging of output strings is done using string alignment algorithms.

Comparison of Adaptive Strategies for On-Line Character Recognition

Abstract: Results on a comparison of adaptive recognition techniques for on-line recognition of handwritten Latin alphabets are presented. The classification strategies compared are based on first compressing or distilling a large database of handwritten characters to a small set of character prototypes. Each adaptive classifier then either modifies the original prototypes or conditionally adds new prototypes when they become available from the user of the system. In each case, the classification decision uses the 1-Nearest Neighbor (1-NN) rule for the distances between the input character and the stored prototypes. The distances are calculated using Dynamic Time Warping (DTW). One of the adaptive learning strategies features an extension of the neural Learning Vector Quantization (LVQ) algorithm to the DTW distance metric. All the methods concerned exhibit automatic unsupervised learning from user input simultaneously with the normal mode of operation. The presented experiments show that the assessed methods produce different tradeoffs between the accuracy and complexity of classification. Every version is, however, able to adapt to the user’s writing style with only a very few — say some tens of — handwritten characters.

Segmentation of Japanese handwritten characters using peripheral feature analysis

Abstract: A method of character segmentation of Japanese handwritten characters has been developed. It is effective especially in character recognition with the over-segmentation process. The method is based on the credibility measurement of each presegmented pattern for being a true character by analyzing peripheral features such as gaps between patterns and widths and heights of patterns. A heuristic statistical method is applied to the analysis. Experimental results show that the developed method increases the accuracy of character segmentation. It is effective especially when no linguistic feedback to segmentation is available and/or the character classifier ability is not high enough. In such cases, the recognition accuracy is increased from 30% to 70% by the new segmentation method.

Recognising Embedded Words in Connected Speech: Context and Competition

Abstract: Onset-embedded words (e.g. cap in captain) present a problem for accounts of spoken word recognition since information coming after the offset of the embedded word may be required for identification. We demonstrate that training a simple recurrent network to activate a representation of all the words in a sequence allows the network to learn to recognise onset-embedded words without requiring a training set that is already lexically segmented. We discuss the relationship between our model and other accounts of lexical segmentation and word recognition, and compare the model’s performance to psycholinguistic data on the recognition of onset-embedded words.

The AddressScriptTM Recognition System for Handwritten Envelopes

Abstract: This paper presents AddressScript - a system for handwritten postal address recognition for US mail. Key aspects of AddressScript technology, such as system control flow, cursive handwriting recognition, and postal database are described. Special attention is paid to the powerful character recognizer and the intensive usage of context, which becomes available during the recognition process. The algorithm of confidence level calculation is presented. Laboratory test results on a blind test set of 50,000 images of live hand-written mail pieces demonstrate a 64% finalization rate for error rates below USPS restrictions.

Combining multiple OCRs for optimizing word recognition

Abstract: We present a method of combining multiple classifiers for optimizing word recognition. The proposed method combines the results of individual classifiers in such a way that the correct word is more likely to be hypothesized. This method provides a solution to the crucial issue of assigning reliable cost to the edges of the segmentation graph in the popular over-segmentation followed by dynamic programming approach for word recognition. Three combination functions are proposed and implemented. Experiments show that proposed method has a significant improvement on the word recognition accuracy.

Character reading method and optical character reader

Abstract: PROBLEM TO BE SOLVED: To reduce misreads while holding the correct read rate by deciding character patterns that respective techniques are skillful at by the individual techniques as much as possible and integrating the results of plural techniques as to a pattern that the techniques are not skillful at so much. SOLUTION: A 1st character recognition part 104 performs character recognition by extracting features from a character image. If the 1st character recognition part 104 fails to recognize the character, a character code obtained as a candidate and its certainty are outputted to a memory 108 and a 2nd character recognition part 105 is actuated. The 2nd character recognition pat 105 is actuated when the 1st character recognition part 104 decides dismissal. The 2nd character recognition part 105 also operates almost similarly to the 1st character recognition part 104 and recognizes a character by extracting features from the character image. A recognition result integration part 106 is actuated when the two character recognition parts decide dismissal. A final character code is outputted by referring to the candidate character codes of the 1st and 2nd character recognition parts 104 and 105 and their certainty which are recorded in the memory 108.