Hisashi Ikeda

Form reading based on form-type identification and form-data recognition

Form reading technology based on form-typeidentification and form-data recognition is proposed. Thistechnology can solve difficulties in variety for readingdifferent items on fairly large number of different types offorms. The form-type identification consists of two parts:(i) extraction of targets such as important keywords in aform by matching between recogised characters and wordstrings in a keyword dictionary, and (ii) analysis ofpositional or semantic relationship between the targets byconstellation matching between these targets and wordlocation information in the keyword dictionary. The formdatarecognition consists of two parts: (i) extraction of aregion of interest (ROI) contained a character string of theitem by using a layout knowledge of the very form-type,and (ii) character string recognition of the item by usingthe linguistic constraint which can be obtained from acontent knowledge of the form-type. A experiment using642 sample forms with 107 different types in totalconfirmed that the form-type identification method cancorrectly identify 97% of 642 form samples at a rejectionrate 3%. Another experiment confirmed that the form-data recognition method can correctly read 95% of thenumber of items on the form samples.

A recognition method for touching Japanese handwritten characters

Describes a method for recognizing Japanese handwritten characters, including touching ones. The touching characters are segmented by cutting the connected components in a pre-segmentation process. The proposed segmentation method consists of the following steps: (1) identification of the touching components by checking the size of candidate character patterns; (2) estimation of the breakpoint and cutting of the components; and (3) generation of candidate character patterns by merging the components with adjacent patterns. The proposed method has the feature that the touching components can be identified even if they are comprised of relatively small patterns. The touching components are cut by using stroke width analysis and projection image analysis. This cut-and-merge procedure was able to segment 134 (88.2%) address lines correctly out of 152 Japanese address lines having touching portions, and could generate correct candidate character patterns.

Pattern Recognition and Synthesis for a Sign Language Translation System

Abstract Sign language is one means of communication for hearing-impaired people. Words and sentences in sign language are mainly represented by hand gestures. In this report, we describe a sign language translation system which we are developing. The system translates Japanese Sign Language into Japanese text and vice versa . In this system, hand shapes and positions are provided by hand-based input in the form of time series. Then, hand gestures are recognized and translated into Japanese sentences using continuous DP (dynamic programming) matching. Japanese text is translated into sign language by generating three-dimensional computer-graphic animation of sign language gestures.

A Segmentation Method for Touching Handwritten Japanese Characters

The purpose of this paper is the segmentation of touching handwritten Japanese characters to enable each segmented character to be recognized. Though segmentation methods cooperating with recognition for the simple characters are known, they are not applicable to complicated characters such as kanji. To estimate the ability of segmentation itself, our method does not use any results of character recognition but segments touching characters using some specific features of character patterns. Linear components in patterns are extracted as features and a segmentation method based on them is proposed. A database, which contains character pattern and ground truth is constructed. In the database the ground truth data describe the correct areas of touching points judged by plural human subjects. As a result of the segmentation experiments, about 64% of touching patterns can be segmented at appropriate points. This correct segmentation rate is automatically calculated consulting the ground truths.

A Context-free Grammar-based Language Model for String Recognition

A context-free grammar-based language model for string recognition has been developed. The developed language model is implemented as a set of graphs which are equivalent to a recursive transition networks. This graph-structured model is not only more compact in memory size than the trie-structured model but also more powerful in describing variations of recognition-target phrases, such as use of characters, selection of words, and structures of phrases. This can improve recognition accuracy. We also developed a method to semi-automatically generate the language model from a collection of sample phrase texts and predefined variation rules. An algorithm to match this language model to the character-recognition results was devised. The model and the generation method have been successfully applied to practical applications such as address recognition. And experimental tests with freely-handwritten addresses show the effectiveness for implementation of the language model and the improvement of the recognitio...

A novel scheme for retrieval of handwritten textual annotations for information just in time (iJIT)

We have designed a novel query retrieval scheme for the information just in time (iJIT) system to retrieve handwritten annotations from digital documents based on typed/handwritten query. The two key components of the developed query retrieval system (QRS) are the character recognition engine and the query retrieval engine. The character recognition engine uses Tesseract 2.01 open source Optical Character Recognition (OCR) Engine under Apache License 2.0 and is trained with handwritten samples from different users. The character recognition engine receives real-time digital pen generated data, and produces segmented-recognition result. The query retrieval engine, resolves the index / query requests from the users for possible information update / retrieval. In case of a handwritten query, the query retrieval engine interacts with the recognition engine to create / update the inverted index table with recognized word labels with annotation indices. In the case of typed text query, the inverted index table is searched directly to retrieve the best matches of annotation indices using a q-gram based approximate string matching technique. A HMM - Viterbi algorithm is finally implemented to find the optimum recognized character sequence in each word using a fuzzy character confusion matrix.

Toward personalized digital library for providing Information JIT

The personalized digital library for providing users withproper information just in time (JIT) is proposed. This digital library can support not only document retrieval task, but also users’ reading and comprehension task. This function can be realized by recording the person-dependent information, which is about human behavior toward documents, such as highlighting, marking, annotating, etc. By utilizing the digital pen and paper with dot-pattern, handwritten annotations on paper documents can be simultaneously digitized and stored in the corresponding e-document. Furthermore, hybridization of real world and e-world in digitizing personal annotations is proposed. This information on users’ annotations makes it possible to retrieve document based on the implicit memory and to provide information support appropriate to the situation with users. The personalized digital library proposed in this paper is to be developed.