Ordering and charging goods have been increasingly treated with faxed forms. FAXOCR system for specified forms has been used practically. But, the accuracy for unspecified forms is not enough, because of the effect of noise on the faxed forms during the facsimile transmission. In this paper, we propose a form position adjustment and identification algorithms which are important roles of the FAXOCR system for unspecified forms. We use the Hough transform which can realize robust line detection from noisy forms. The proposed algorithm can reduce heavy computational burden of the Hough transform by utilizing the property that line segments on the form consists of only horizontal and vertical components. Furthermore, we used a mathematical property that parallel transform and rotation by facsimile transmission can be expressed easily in the Hough space as the shifts of each axis in the Hough-space. The performance of the proposed method can be verified by experiments using actual form data.

View full abstract

A method for extracting feature amounts from the facial area and the construction of knowledge to recognize facial expressions are presented in this paper. In order to recognize the facial expressions exactly, each feature amount must be extracted by using an individual method suitable for each specific part of the face. From this point of view, The method suitable for the eye and the mouth is applied, respectively. In the knowledge for facial expressions recognition, the membership function which can respond to an ambiguous motion is used. Using computer simulation with the obtained knowledge for five people, experimental results show 86.8% accuracy on average for the expressive facial images.

View full abstract

“AWA ODORI" dance is one of the traditional dance festival of Tokushima Prefecture in Japan. Basic operation of “AWA ODORI" dance needs harmony of hands and legs, and the rhythm of duple time, but it is not possible that the beginners do these actions completely. In this paper, a method of quantifying evaluation of the skill of “AWA ODORI" dance is proposed. The image sequence of dance scenes are obtained by using color motion capture system. The dynamics of dance is detected by dynamic image processing. The data waveform and model data waveform of excellent dancers are compared, and the feature of dance are extracted.

View full abstract

We propose a joint detection method considering hierarchical contours. We also introduce Iterative Deepening into the method for detecting joint segment efficiently. We apply proposed method to reassembling potsherds which could not be reassembled by existent methods. We report on good performance results for the pot sherds. First, we describe purposes of this research in Chapter1. We describe a method for the hierarchy of contours in Chapter2. We propose the algorithm of joint detection considering hierarchical contours in Chapter3. We explain an outline of our earthenware reconstruction system in Chapter4. We give experimental results of our system in Chapter5., and finally, we conclude remarks.

View full abstract

This paper presents a new approach to blurred image restoration using an inverse filter optimized by Genetic Algorithm (GA). The feature of proposed restoration approach is to estimate a degradation level based on Discrete Fourier Transform spectrum prior to an inverse filter optimization using Real-coded GA. An addition of the degradation level estimation enables the adaptive image restoration according to its blurred level and the high-speed image restoration by efficient GA search of inverse filters. Experiments on several kinds of natural images demonstrate that the proposed restoration approach can provide a practical solution for a high-speed image restoration.

View full abstract

Acceleration methods based on bounding volumes have been widely recognized as an efficient approach for diminishing volume rendering time by eliminating unnecessary computations. Objects inside the volume data are enclosed by various shapes in order to delimit the range for volume rendering, thus avoiding unnecessary calculation. This paper presents an approach for accelerating pure software-based volume rendering by using object boundary projection and parallel processing techniques. Object boundary projection is a variant of the bounding volume method and works by extracting the tight object boundary information and projecting it onto the image space plane in order to obtain the effective range, both in image and object space for efficient volume rendering. We optimized the original object boundary projection algorithm and added other facilities such as macro-cell projection and parallel processing. An efficient load-balancing scheme for image-space data distribution which takes into consideration the boundary information is also presented and these were evaluated on a distributed-memory parallel architecture machine. We obtained some encouraging results which show the effectiveness of this acceleration approach by diminishing volume rendering time without sacrificing the rendered image quality.

View full abstract