Haruo Asada

Understanding multi-articled documents

A document understanding method based on the tree representation of document structures is proposed. It is shown that documents have an obvious hierarchical structure in their geometry which is represented by a tree. A small number of rules are introduced to transform the geometric structure into the logical structure which represents the semantics. The virtual field separator technique is employed to utilize the information carried by special constituents of documents such as field separators and frames, keeping the number of transformation rules small. Experimental results on a variety of document formats have shown that the proposed method is applicable to most of the documents commonly encountered in daily use, although there is still room for further refinement of the transformation rules.<<ETX>>

An iterative prediction and correction method for automatic stereocomparison

In image processing of aerial photographs in cartography and of remote sensor data in earth resource surveys, one of the most basic parameters is the elevation of the ground estimated by means of parallax measurements made on stereo pairs of aerial images. In conventional methods, measurement by manual stereocomparators is precise but timeconsuming. Existing automatic stereocomparators do not work well because they use too simple a correlation method to get sufficiently precise measurements. Each of the two corresponding patterns from a stereo pair is so distorted by parallax that the calculated correlation value does not always attain the maximum at the point from which we can obtain the correct parallax. As a result of the present study, it appears that an iterative prediction and correction method can be suitable for the efficient and accurate measurement of elevation from stereo imagery. The conventional methods do not fully utilize the information which is presented in the image itself or is a priori known. The first prediction of elevation is carried out only for well-contrasting parts of the image, which usually means roads, coast, forest edges, etc. Then this prediction is expanded to the neighboring parts, making use of the continuity of elevation. Using a temporary predicted elevation map, one of the stereo pair patterns is modified to lessen the distortion compared with the other pattern. Correction of the elevation map is performed using the correlation values calculated from the modified patterns. This feedback procedure is iterated until no correction is needed. This new method is applied to digitized aerial photographs taken at a height of 3,000 m. Finally, a planned automatic image processing system is explained briefly, consisting of a newly developed scanner which can access every point in a scan field at high positional accuracy.

Resolving Ambiguity in Segmenting Touching Characters

This paper presents an efficient and powerful character segmentation method which enables touching characters in a document to be read accurately at high speed. The character segmentation phase extracts characters from a text line. Connected components in a text line image may have to be segmented or combined to form recognizable characters. For example, the character ‘i’ is formed by combining two components. Touching characters are segmented to identify each character. Segmenting touching characters is an open problem, whose solution would advance the field. Touching characters have several candidates for their break position, which are then confirmed by recursive segmentation and recognition, and finally by the determination of the linguistic context. There are several possible candidates at each stage. For example, several candidates for the break position of touching characters are nominated. Any segmented area might possibly fit several alternative characters. Therefore, an efficient resolution of ambiguity at each stage is significantly critical and indispensable for practical text reading. The authors’ approach is based on knowledge about character composition (e.g. an ‘m’ is like a combination of an ‘r’ and an ‘n’), as well as knowledge about omni-fonts. Knowledge about character composition compresses the number of recursive segmentation and recognition.

A visual navigation system using a multi-information local map

A visual navigation system for a robot working in a nuclear power plant is presented. Such a system should have two functions. One is to provide a rough environment map or directions towards a goal, such as turning left at the second corner. The other is to avoid obstacles and to find a passage for continuous movement of the robot. The operation of the system is divided into two stages. In the first stage, a multi-information local map (MIL map) describing information necessary for self-location measurement is created interactively from stereo images collected during remote-controlled navigation. The second stage is autonomous navigation. The robot moves autonomously by using the map with the aid of the second function of the system. The system consists of five subsystems: an environment teaching subsystem, to create the MIL map; a self-location measurement subsystem, to find the robots own location; an obstacle detection subsystem; a path planning subsystem; and a path following subsystem. Experimental results from a mobile robot show the usefulness of the system.<<ETX>>

New Image Processing Hardwares And Their Applications To Industrial Automation

Image processing hardware, a high-speed image processor and an intelligent image memory unit, are presented and their applications to the industrial image processing are discussed. An image processor has been realized on the basis of local parallel computation architectures, which contribute to speeding up major calculations and to reducing the hardware cost and complexity. An image memory unit has been implemented to manipulate and display the image data with a highly intelligent control, which helps man-machine interaction to be effective. Although these processors are conceptually designed for general image processing purposes, several feasibility studies have revealed that they can be utilized as com-petent simulation tools for research works on shape recognition, particle measurement, real-time signal processing, etc.

Obstacle detection by disparity prediction stereopsis method

The authors devised a method (disparity prediction Stereopsis method) which calculates the range of the disparity which corresponds to the space where a robot is due to move and predicts the existence or non-existence of the object by matching the edge segment extracted from the stereo images with the range of the disparity. The method is applied to the obstacle detection in the navigation of the mobile robot. The object to be detected is a still object standing on a floor. First, the existence of the object is predicted by the stereo matching of vertical edges. The remaining edges are examined only if the matched vertical edge exists. As a result, the efficiency is improved. In this method, the range of search is restricted and local matching is tried. Consequently, an incorrect match (false match) is produced easily in the background, in the environment with low contrast. To remedy this adverse effect, the floor region extraction process is executed by assuming the spatial continuity and the similarity of the gray-level distribution in the near region. If the lower end of the matched edge is contained in the floor region, the object is decided as a candidate for the obstacle. By this decision scheme, the probability of misdetection of the object which does not interfere with the space where the robot is due to move such as the object far and high above the viewpoint and the object hanging on the front wall can be reduced. This improves the reliability. By the indoor evaluation experiment, such as the scene of laboratory, the usefulness of the method is verified. The motion control of the robot is attempted in an environment simulating a nuclear power plant.

Image Processing Unit Hardware Implementation

Conventional digital computers are inefficient for even simple local image processing operations in terms of cost-effectiveness. Image processing by serial computers requires a large amount of computing time. Programs and data are stored in the same memory and all operations are serially executed in the so-called “von Neuman” architecture, even though image processing could usually be performed mostly in parallel. Two-dimensional image data require a large amount of memory storage and usually exceed the main memory capacity. This results in much overhead time transferring image data between the main memory and secondary storage, e.g., magnetic disk. Therefore, we have elected to implement cost-effective image processing by developing special hardware optimized to overcome the above problems.

Visual Navigation System for a Mobile Robot

This paper p r e s e n t s a v i s u a l n a v i g a t i o n system f o r a r o b o t working i n a n u c l e a r power p l a n t . Such a system should have two f u n c t i o n s . One i s t o p r o v i d e a rough environment map o r d i r e c t i o n s towards a g o a l , such a s t u r n i n g l e f t a t t h e second c o r n e r . The o t h e r is t o a v o i d o b s t a c l e s and t o f i n d a passage f o r a cont inuous movement o f t h e r o b o t . The o p e r a t i o n of t h e system is d i v i d e d i n t o two s t a g e s . I n t h e f i r s t s t a g e , a MIL(mu1ti-information l o c a l ) map d e s c r i b i n g t h e s e v e r a l in format ion necessary f o r a s e l f l o c a t i o n measurement i s c r e a t e d i n t e r a c t i v e l y from s t e r e o images c o l l e c t e d d u r i n g remote-cont ro l led n a v i g a t i o n . The second s t a g e i s autonomous n a v i g a t i o n . The r o b o t moves autonomously by u s i n g t h e map w i t h t h e a i d of t h e second f u n c t i o n of t h e system. The system c o n s i s t s of f i v e subsystems : an environment t e a c h i n g subsystem, a s e l f l o c a t i o n measurement subsystem, an o b s t a c l e d e t e c t i o n subsystem, a p a t h p l a n n i n g subsystem, and a p a t h fo l lowing subsystem. The environment t e a c h i n g subsystem c r e a t e s an environment map: t h e s e l f l o c a t i o n measurement subsystem g i v e s t h e l o c a t i o n o f t h e r o b o t by comparing t h e i n p u t image and t h e map i n f o r m a t i o n . The o t h e r t h r e e subsystems s u p p o r t autonomous n a v i g a t i o n . The system h a s been implemented on a mobile r o b o t . Exper imenta l r e s u l t s have shown t h e u s e f u l n e s s of t h e proposed system.