Japanese Character Recognition Research Articles

An optimizing method of Japanese character recognition based on improved support vector machine

An optimizing method of Japanese character recognition based on improved support vector machine

Classification of Japanese Handwritten Characters Using Biometrics Approach

The following paper presents a solution to the problem of offline recognition of Japanese characters. Minutiae and other features extractable from handwriting images have been used to recognize individual characters. The solution presented by the authors uses minutiae to recognise single Japanese characters. Due to the complexity of this typeface, the solution presented can be used to recognise archaic characters, from old documents or also works of art. Neural Networks and hybrid classifiers based on five basic types of classifiers, i.e., k-nearest neighbour method, decision trees, support vector machine, logistic regression and Gaussian Naive Bayes classifier have been developed for classification. The study was conducted on Hiragana, Katakana and Kanji characters (ETL9G Database). The accuracy value obtained was 99.934%. The authors present what is probably the first algorithm using minutiae to recognize Japanese handwriting.

Accuracy analysis of Japanese machine translation based on machine learning and image feature retrieval

At present, there are still many deficiencies in Chinese-Japanese machine translation methods, the processing of corpus information is not deep enough, and the translation process lacks rich language knowledge support. In particular, the recognition accuracy of Japanese characters is not high. Based on machine learning technology, this study combines image feature retrieval technology to construct a Japanese character recognition model and uses Japanese character features as the algorithm recognition object. Moreover, this study expands image features by generating a brightness enhancement function using a bilateral grid. In order to exclude the influence of the edge and contour of the image scene on the analysis of the image source, the brightness value of the HDR image is used instead of the pixel value of the image as the image data. In addition, this research designs experiments to study the translation effects of this research model. The research results show that the model proposed in this paper has certain effects and can provide theoretical references for subsequent related research.

畳み込みニューラルネットワーク(CNN)を用いた手書き日本語文字認識システムの試作

畳み込みニューラルネットワーク(CNN)を用いた手書き日本語文字認識システムの試作

An Algorithm for Japanese Character Recognition

In this paper we propose a geometry- topology based algorithm for Japanese Hiragana character recognition. This algorithm is based on center of gravity identification and is size, translation and rotation invariant. In addition, to the center of gravity, topology based landmarks like conjunction points masking the intersection of closed loops and multiple strokes, as well as end points have been used to compute centers of gravity of these points located in the individual quadrants of the circles enclosing the characters. After initial pre-processing steps like notarization, resizing, cropping, noise removal, synchronization, the total number of conjunction points as well as the total number of end points are computed and stored. The character is then encircled and divided into four quadrants. The center of gravity (cog) of the entire character as well as the cogs of each of the four quadrants are computed and the Euclidean distances of the conjunction and end points in each of the quadrants with the cogs are computed and stored. Values of these quantities both for target and template images are computed and a match is made with the character having the minimum Euclidean distance. Average accuracy obtained is 94.1 %.

Recognition of Japanese Phonographic Kana (Hiragana) and Logographic Kanji Characters by Passive Finger Tracing

The present study assessed the ability of normal Japanese adults to recognize kanji and hiragana characters through passive finger tracing without visual cues. We tested fifty-six right-handed Japanese university students using the left or right hand. Participants recognized approximately 50% of 44 kanji characters, regardless of the hand they used. The results are consistent with previous findings in Chinese speakers. In contrast, the average correct response to the 44 hiragana was almost 80%, again irrespective of hand. The results are discussed in terms of differential processing systems for Japanese writing and of differential cerebral hemispheric specializations.

An OCR Post-Processing Method Based on Dictionary Matching and Matrix Transforming

This paper describes a post-processing method for Chinese and Japanese character recognition based on dictionary. By the analysis results of recognition in the processing of OCR, we can find some segmentation and recognition errors do not conform to the rules of lexical and just recognized as the characters which its fonts approach to the scanned texts. For these errors we can deal with them by the Fix Length Segmentation Matching based on Dictionary and the Glyph Code Matrix Transforming. Through the above processing, most of the inaccurate recognitions can be corrected and by the experimental results, it can be proved that this method is an effective way to improve the recognition rate of Chinese and Japanese Character.

Relationships between Semantic Activation and Eye Fixation in Visual Word Recognition of Japanese Characters

Relationships between Semantic Activation and Eye Fixation in Visual Word Recognition of Japanese Characters

Prototype learning for structured pattern representation applied to on-line recognition of handwritten Japanese characters

This paper describes prototype learning for structured pattern representation with common subpatterns shared among multiple character prototypes for on-line recognition of handwritten Japanese characters. Prototype learning algorithms have not yet been shown to be useful for structured or hierarchical pattern representation. In this paper, we incorporate cost-free parallel translation to negate the location distributions of subpatterns when they are embedded in character patterns. Moreover, we introduce normalization into a prototype learning algorithm to extract true feature distributions in raw patterns to aggregate distributions of feature points to subpattern prototypes. We show that our proposed method significantly improves structured pattern representation for Japanese on-line character patterns.

An Application of Binary Decision Trees to Pattern Recognition

Decision rules are a key technique in decision making, data mining and knowledge discovery in databases. We introduce an application of decision rules, handwriting pattern classification. When decision rules are applied to pattern recognition, one rule forms a hyperrectangle in feature space, i.e., each decision rule corresponds to one hyperrectangle. This means that a set of decision rules is considered a classification system, called the subclass method. We apply decision rules to handwritten Japanese character recognition, showing experimental results.

Printed Japanese Character Recognition Using Multiple Commercial OCRs

This paper proposes two algorithms for maintaining matching between lines and characters in text documents output by multiple commercial optical character readers (OCRs). (1) a line matching algorithm using dynamic programming (DP) matching and (2) a character matching algorithm using character string division and standard character strings. The paper proposes a method that introduces majority logic and reject processing in character recognition. To demonstrate the feasibility of the method, we conducted experiments on line matching recognition for 127 document images using five commercial OCRs. Results demonstrated that the method extracted character areas with more accuracy than a single OCR along with appropriate line matching. The proposed method enhanced recognition from 97.61% provided by a single OCR to 98.83% in experiments using the character matching algorithm and character recognition. This method is expected to be highly useful in correcting locations at which unwanted lines or characters occur or required lines or characters disappear.

The state of the art in Japanese online handwriting recognition compared to techniques in western handwriting recognition

This paper compares the current state of the art in online Japanese character recognition with techniques in western handwriting recognition. It discusses important developments in preprocessing, classification, and postprocessing for Japanese character recognition in recent years and relates them to the developments in western handwriting recognition. Comparing eastern and western handwriting recognition techniques allows learning from very different approaches and understanding the underlying common foundations of handwriting recognition. This is very important when it comes to developing compact modules for integrated systems supporting many writing systems capable of recognizing multilanguage documents.

Handwritten Japanese Character Recognition Using Adaptive Shape Normalization by Global Affine Transformation

This paper proposes a new, promising character recognition system with a category-dependent shape normalization technique that normalizes an input pattern against each reference pattern adaptively using global affine transformation (GAT) as follows. (1) An input character pattern is fed to basic OCR, the most powerful of the conventional OCRs. (2) The basic OCR plays the role of rough classification and outputs a small set of candidate categories for the input pattern. (3) GAT normalizes the input pattern against each candidate's reference pattern adaptively. (4) Each adaptively normalized input pattern is fed again to the basic OCR. (5) The final recognition result is obtained using the updated distance values within candidate categories. In experiments, our basic OCR linked to GAT adaptive normalization is successfully applied to 28,694 patterns of totally unconstrained handwritten characters, including Kanji, Kana, and alphanumerics, written by 300 people, with substantial improvements in recognition accuracy.

日本語を学習する中国語母語話者の漢字の認知

日本語を学習している中国語母語話者 (上級学習者) が漢字を日本語で読み上げるとき, その漢字の音読みが母語の中国語の発音に似ていると, 似ていない発音の漢字を読み上げるときより反応時間が短くなった。また, 通常訓読みで読まれる漢字よりも通常音読みで読まれる漢字のほうが反応時間が短かった。これは, 漢字を日本語で読む際にも母語の音韻情報が活性化するためであると考えられ, 中国語音起源でない訓読みも, 音韻情報の一部として心内辞書に収められていることが要因であると考えられた。そして, 異言語間で共用されている漢字という文字にそれらの言語の音韻情報が備わっていることが示唆された。一方, 日本語学習歴がさらに長く, 超上級と言われるレベルに達している学習者は, エラーも少なく, 反応時間は, 日本語と中国語の発音の類似には影響されなかった。

Fast and precise discriminant function considering correlations of elements of feature vectors and its application to character recognition

During the last few years, research in recognition of handwritten Chinese and Japanese characters has matured significantly. However, in order to obtain high recognition rate, most character recognition systems have paid an extremely high computational cost. For high-performance character recognition systems, reduction of the expensive computational cost has become a very important goal. Discriminant function is a very important factor for precise pattern recognition. The Mahalanobis distance is considered as an effective function. However, precise calculation of the Mahalanobis distance requires extremely large numbers of training samples. In this article, by investigating the relationwhip of elements of feature vector, a new discriminant function, the vector-partitioned Mahalanobis distance, is proposed. With the proposed method, high recognition performance can be obtained with less computational cost. Because the proposed method partitions high dimensional feature vector into several small number dimensional vectors, the ratio of the number of training samples to the number of dimensions becomes larger. This method is especially effective in the case of lack of training samples. The effectiveness of the proposed method is shown by experimental results with the database ETL9B. © 1999 Scripta Technica, Syst Comp Jpn, 30(14): 33-42, 1999

Fast and precise discriminant function considering correlations of elements of feature vectors and its application to character recognition

During the last few years, research in recognition of handwritten Chinese and Japanese characters has matured significantly. However, in order to obtain high recognition rate, most character recognition systems have paid an extremely high computational cost. For high-performance character recognition systems, reduction of the expensive computational cost has become a very important goal. Discriminant function is a very important factor for precise pattern recognition. The Mahalanobis distance is considered as an effective function. However, precise calculation of the Mahalanobis distance requires extremely large numbers of training samples. In this article, by investigating the relationwhip of elements of feature vector, a new discriminant function, the vector-partitioned Mahalanobis distance, is proposed. With the proposed method, high recognition performance can be obtained with less computational cost. Because the proposed method partitions high dimensional feature vector into several small number dimensional vectors, the ratio of the number of training samples to the number of dimensions becomes larger. This method is especially effective in the case of lack of training samples. The effectiveness of the proposed method is shown by experimental results with the database ETL9B. © 1999 Scripta Technica, Syst Comp Jpn, 30(14): 33-42, 1999

A handwritten character recognition system using directional element feature and asymmetric Mahalanobis distance

This paper presents a precise system for handwritten Chinese and Japanese character recognition. Before extracting directional element feature (DEF) from each character image, transformation based on partial inclination detection (TPID) is used to reduce undesired effects of degraded images. In the recognition process, city block distance with deviation (CBDD) and asymmetric Mahalanobis distance (AMD) are proposed for rough classification and fine classification. With this recognition system, the experimental result of the database ETL9B reaches to 99.42%.

Reading of Japanese Kanji (morphograms) and Kana (syllabograms): a magnetoencephalographic study

Magnetoencephalograms were recorded from six healthy Japanese subjects in order to investigate the areas in the cortices which are involved in the recognition of Japanese characters (Kanji and Kana). Forty-four Kanji (morphograms), 44 Kana (syllabograms) and 20 alphabet letters were used as stimuli. They were presented randomly and the subjects were required to read each stimulus and count the number of letters. The magnetic responses were recorded with dual 37-channel SQUID (Superconducting Quantum Interference Device) gradiometers from the temporal, parietal and occipital areas of the brain. The magnetic responses to Kanji and Kana were similar and consequently the locations of equivalent current dipoles (ECDs) to Kanji and those to Kana did not differ at any recording site. In all the subjects, ECDs were found in the posterior-inferior temporal (PIT) areas approximately corresponding to Brodmann area 37 in the latency range of 150–300 msec. These activities were found in both hemispheres without consistent laterality. The location of the ECD moved forward from posterior to anterior in the PIT area as the latency increased in all but one subject. Only one subject showed activities in the left angular gyrus. Since activities in PIT areas were also found in alphabet letters, the bilateral PIT areas are considered to play an essential role in reading.

Form and sound similarity effects in kanji recognition

Four experiments are reported here to address the question of whether figurative and phonological processing based on sub-word components (radicals) interact in the recognition of Japanese kanji characters. A delayed matching task was used in which two briefly exposed ‘source’ characters (e.g., ), each made up by two radicals, were followed by a probe character (e.g., ) which in critical conditions was different from the source characters. The task of the subject was to decide whether the probe was one of the two source characters. When a probe was figuratively similar to the source display, the homophonic relatedness between source and probe characters elicited more false responses to the probe. However, no homophony effect was found when the probe was dissimilar to the source display. Further, the false alarm rates in the homophone condition with figurative similarity was shown to be sensitive to proportion of homophonous trials in negative sets. The results suggest that phonological information of both whole character and of components was automatically activated despite experimental tasks in which subjects were given little incentive to execute phonetic processing. It is concluded that the interaction of figurative and phonological processing is due to mutual activation of the whole character and its radical(s) in the process of word identification in kanji. The results are considered within an interactive-activation framework with fore- and background activation device in multilevels.

Improvement of handwritten Japanese character recognition using weighted direction code histogram

Several algorithms for preprocessing, feature extraction, pre-classification, and main classification are experimentally compared to improve the recognition accuracy of handwritten Japanese character recognition. The compared algorithms are three types of nonlinear normalization for the preprocessing, the discriminant analysis and the principal component analysis for the feature extraction, the minimum distance classifiers and the linear classifier for the high-speed pre-classification, and modified Bayes classifier and subspace method for the robust main classification. The performance of the recognition algorithm is fully tested using the ETL9B character database. The recognition accuracy of 99.15% at the recognition speed of eight characters per second is achieved. This accuracy is the best one ever reported for the database.