Tatsunari Sakurai

This paper proposes a stereo algorithm utilising a reaction-diffusion system defined in a three-dimensional domain. A previous reaction-diffusion stereo algorithm provides a stereo disparity map by utilising multi-layered reaction-diffusion systems defined in a two-dimensional domain. The previous algorithm assumes that three-dimensional structures of scenes consist of unslanted planar surfaces and approximately describes any slanted or curved surfaces with piecewise unslanted planar surfaces. However, in real scenes there are highly slanted or curved surfaces, which violate the assumption of the previous algorithm and cause inaccurate results of disparity distribution, in particular, with respect to subpixel accuracy. The reaction-diffusion system consists of reaction-diffusion equations, which are described with partial-differential equations and solved with a numerical computation technique such as a finite difference method. Thus, we revise the reaction-diffusion stereo algorithm by utilising a reaction-diffusion system defined in a three-dimensional domain consisting of the two-dimensional domain plus a disparity domain. Discretisation of the disparity domain brings subpixel accuracy and helps the algorithm to detect accurate disparity for slanted or curved surfaces. In addition, this paper proposes a two-stage algorithm for fill-in of disparity undefined areas caused by little image feature. Finally, this paper demonstrates performance of the proposed algorithm for slanted or curved surfaces with the Middlebury stereo vision data-sets. © 2012 ACM.

This paper presents a quick review of reactiondiffusion systems and the application of a discretized version of a reaction-diffusion system to edge detection in image processing. A reaction-diffusion system refers to a system consisting of diffusion processes coupled with reaction processes. Several reaction-diffusion systems exhibit pattern formation processes, in which the systems self-organize spatio-temporal patterns of target and spiral waves propagating in two-dimensional space. In addition, some of the systems having strong inhibitory diffusion self-organize stationary patterns; the Turing pattern is one of the typical examples of the stationary patterns observed in reaction-diffusion systems under strong inhibitory diffusion. We have previously found that the discretized version with strong inhibition has a mechanism detecting edges from an image intensity distribution. The mechanism divides an image intensity distribution into brighter or darker intensity areas with a threshold level, and organizes pulses along edges of the divided areas. By searching an output distribution of the version for pulses, we can achieve edge detection. However, since the threshold level is usually fixed at a constant value in the version, the mechanism is not applicable to gray level images. Thus, this paper furthermore proposes an edge detection algorithm consisting of two pairs of the version with a variable threshold level. We apply the edge detection algorithm and a representative algorithm proposed by Canny to several artificial and real images in order to confirm their performance.

This paper presents a computer algorithm of detecting edges from a grey scale image with FitzHugh-Nagumo type excitable elements discretely spaced at image grid points. A previous edge detection algorithm utilising the elements is not applicable to darker intensity areas surrounded by brighter ones the algorithm fails in detecting edges in the areas. In order to solve the problem in detecting edges in relatively dark areas, we proposed to utilise an intensity inverted image as well as its original one. The proposed algorithm firstly provides a tentative edge map from the original image, and simultaneously provides an additional tentative edge map from the inverted image. Then, the algorithm provides a final edge map by merging the two edge maps. We quantitatively confirm performance of the proposed algorithm, in comparison with that of the previous one and that of the Canny algorithm for an artificial grey scale image not having noise. We furthermore confirm robustness and convergence of the proposed algorithm for a noisy image and real ones. These results shows that the performance of the proposed algorithm is much higher than the previous one and is comparable with the Canny algorithm for a noise-less image, and the proposed algorithm converges for all of the images. However, the proposed algorithm is vulnerable for additive noise, in comparison with the Canny algorithm and the anisotropic diffusion algorithm. © 2011 IEEE.

This paper presents a reaction-diffusion stereo algorithm with anisotropic diffusion. The reaction-diffusion stereo algorithm detects the stereo disparity, by finding a one-to-one stereo correspondence between the left and right images. There is often ambiguity in finding stereo correspondences because of a lack of texture or similar brightness patterns in objects. In addition, in occluded areas, there is no information by which to find stereo correspondence. Thus, it is necessary to fill in undetected areas of stereo disparity from neighbouring detected areas on a stereo disparity map according to the two constraints of uniqueness and continuity. The reaction-diffusion stereo algorithm realises the continuity constraint with the two different diffusion processes of activation and inhibition. Motivated by psychological evidence from actual human stereo depth perception, this paper proposes the imposition of anisotropic inhibitory diffusion on the algorithm. We apply the proposed anisotropic reaction-diffusion algorithm to several pairs of stereo images and evaluate its performance in comparison to the original reaction-diffusion stereo algorithm. Results of the evaluations show that the addition of anisotropic inhibitory diffusion is partially effective in depth discontinuity areas. © 2011 IEEE.

Low cost and attractive equipment for use in experimental higher physics education has been desired by teachers, especially those in developing world. A system of novel experimental apparatus named Personal Desk Lab (PDL) was developed [1]. Each apparatus is miniaturized to one-fifth of the conventional one or less, and built on a steel plate; therefore, it is portable and can be used on a classroom desk. Each set is constructed by some parts divided according to their functions, and some of the parts are used in a number of experiments; which saves material, cost, and storage space. All parts are designed to be easy to make, maintain and repair. Almost all apparatus are battery driven [2]. After ICPE 2006, we have improved the system continuously; consequently, the experimental themes cover the field of mechanics, electromagnetism and optics. The number of these themes in use exceeds ten. The performance of PDL has been tested at Chiba University, Japan and Royal University of Phnom Penh (RUPP), Cambodia. In Chiba University, physics education with PDL is currently conducted individually to 80(max.) students in a classroom at the same time, and to more than 900 students per year. Experiment class using PDL in Cambodia started on October 2008 with 120 students of physics department, RUPP. They were divided into three classes, and conducted four experimental themes in pairs. The advantages confirmed from the practices at two universities are as follows: (1) the use of PDL arouses learner's interest, promotes their deep understanding extensively, and inspires to learn further; and (2) costs for introduction and running of PDL system are fairly small compared to the traditional one. Furthermore, the instruction for distant learners having PDL on each hand was conducted successfully through internet.