Shunji Mori

Research on Machine Recognition of Handprinted Characters

Machine recognition of handprinted Chinese characters has recently become very active in Japan. Both from the practical and the academic point of view, very encouraging results are reported. The work is described systematically and analyzed in terms of so-called feature matching, which is likely to be the mainstream of the research and development of machine recognition of handprinted Chinese characters. A database, ETL8 (881 Kanji, 71 hirakana, and 160 variations for each category), is explained, on which many experiments were performed. Recognition rates reported using this database can be compared, and so somewhat qualitative evaluation of these methods is described. Based on the comparative study, the merits and demerits of both feature and structural matching are discussed and some future directions are mentioned.

Algebraic description of curve structure

The authors propose a compact and concise method of describing curves in terms of the quasi-topological features and the structure of each singular point. The quasi-topological features are the convexity, loop, and connectivity. The quasi-topological structure is analyzed in a hierarchical way, and algebraic structure is presented explicitly for each representation level. The lower-level representations are integrated into the higher-level one in a systematic way. When a curve has singular points (branch points), the curve is decomposed into components, where each is a simple arc or a simple closed curve, by decomposing each singular point. The description scheme is applied to character recognition. >

Recognition of handprinted characters by an outermost point method

Abstract A recognition system for handprinted characters has been developed for reading sets of characters consisting of a mixture of numerical, alphabetic and Japanese “katakana” characters, a total of 82 symbols. This paper describes a simple convex hull method used for extracting concavity and convexity. The convex hull is constructed by finding the outermost points through tracing the contours of a character. When this system was applied to data consisting of 16,400 characters, a very high level of performance was achieved. The system also automatically compiles a dictionary.

An algebraic approach to automatic construction of structural models

We present algebraic approach to the inductive learning of structural models and automatic construction of shape prototypes for character recognition on the basis of the algebraic description of curve structure proposed by Nishida and Mori (1991, 1992). A class in the structural models is a set of shapes that can be transformed continuously to each other. We consider an algebraic representation of continuous transformation of components of the shape, and give specific properties satisfied by each component in the class. The generalization rules in the inductive learning are specified from the viewpoints of continuous transformation of components and relational structure among the components. The learning procedure generalizes a pair of classes into one class incrementally and hierarchically in terms of the generalization rules. We show experimental results on handwritten numerals. >

STRUCTURAL DESCRIPTION OF LINE IMAGES BY THE CROSS SECTION SEQUENCE GRAPH

In this paper, we propose the Cross Section Sequence Graph which describes line images in a simple and well structured form. It is composed of regular regions called cross section sequences and singular regions. A cross section sequence is a sequence of cross sections, each of which is constructed as a pair of boundary points almost perpendicular to the direction of the line. The sequence corresponds to a straight or curved line segment. The remaining regions are extracted as singular regions, each of which corresponds to an end point region, corner, branch, cross, and so on. The cross section sequence graph is useful for many kinds of feature extraction, especially for skeletonization since a singular region can be analyzed from adjacent regular regions. Experimental results show that the skeleton extracted from the cross section sequence graph is better than that of a pixel-wise skeletonization (thinning) in terms of both processing speed and the quality of the skeleton.

A MODEL-BASED SPLIT-AND-MERGE METHOD FOR CHARACTER STRING RECOGNITION

Recognition of handwritten character strings is a challenging problem, because we need to cope with variations of shapes and touching/breaking of characters at the same time. A natural approach to recognizing such complex objects is as follows: The object is decomposed into segments, and meaningful partial shapes (shapes which are recognized as some characters) are constructed by merging segments locally. Then, a globally consistent interpretation of the object is determined from the combination of partial shapes. This approach can be referred to as a model-based split-and-merge method. Based on this idea, we present an algorithm for recognition and segmentation of character strings. We give systematic performance statistics by experiments using handwritten numerals. This algorithm can be applied to character strings composed of any number of characters and any type of touching or breaking, whether the number of constituent characters is known or unknown.

Structural Analysis and Description of Curves by Quasi-Topological Features and Singular Points

We propose a compact and concise description method of curves in terms of the quasi-topological features and the structure of each singular point. By quasi-topological features, we mean the convexity, loop, and connectivity. The quasi-topological structure is analyzed in a hierarchical way, and algebraic structure is presented explicitly on each representation level. The lower-level representations are integrated into the higher-level one in the systematic way. When a curve has singular points (branch points), the curve is decomposed into components each of which is a simple arc or a simple closed curve by decomposing each singular point. The description scheme is applied to character recognition.

A Model-Based Split-and-Merge Method for Recognition and Segmentation of Character Strings

A Quick Way for Relational Matching: Morphology p. 47 The Use of Grammatical Characteristics in Syntactic Pattern Classification p. 63 Understanding Neural Networks for Grammatical Inference and Recognition p. 75 Identifying Regular Languages in Polynomial Time p. 99 Inference of k-Testable Tree Languages p. 109 On the Estimation of the Probabilistic-Distribution Functions of Stochastic Structural Models by Maximizing Merit Functions p. 121

Shape Description and Classification Based on Extremal Points and Their Relations

Humans classify shapes naturally. However, nobody knows this classification mechanism which is known as a very profouned problem. Here shapes on 2-Dimension are treated. Mathematics is essential to approach this open problem, but it seems that topology is too rough and differential geometry is too precise. In this sense, this paper proposes a new description of binary images, based on the fundamentals of topology and differential geometry, and gives a natural classification of shapes according to the description. A contour of shape is described by concatenation of four symbols which correspond to four kinds of extremal points where we assume that the contour of shape is a simple closed curve. Some properties of the strings/words consisting of these symbols are shown. Operations on the string, named horn making and horn removing are introduced. A set of such strings is closed under these operations. The positional relations of extremal points are investigated and it is shown that there exist positional orders of them which are invariants for continuous mapping under the condition of keeping the symbol representation. Based on the relations, a more detailed, but not so precise classification, of shapes which meets our intuition is given.