Yoshiki Mizukami

An off-line signature verification system using an extracted displacement function

This paper proposes an off-line signature verification system based on a displacement extraction method. The optimum displacement functions are extracted for any pair of signatures using minimization of a functional. The functional is defined as the sum of the squared Euclidean distance between two signatures and a penalty term requiring smoothness of the displacement function. A coarse-to-fine search method is applied to prevent the calculation from stopping at local minima. Based on the obtained displacement function, the dissimilarity between a questionable signature and the corresponding authentic one is measured. The proposed system achieved error rate of 24.9% in a experiment.

Optical Flow Computation on Compute Unified Device Architecture

In this study, the implementation of an image processing technique on compute unified device architecture (CUDA) is discussed. CUDA is a new hardware and software architecture developed by NVIDIA Corporation for the general- purpose computation on graphics processing units. CUDA features an on-chip shared memory with very fast general read and write access, which enables threads in a block to share their data effectively. CUDA also provides a user- friendly development environment through an extension to the C programming language. This study focused on CUDA implementation of a representative optical flow computation proposed by Horn and Schunck in 1981. Their method produces the dense displacement field and has a straightforward processing procedure. A CUDA implementation of Horn and Schuncks method is proposed and investigated based on simulation results.

A handwritten Chinese character recognition system using hierarchical displacement extraction based on directional features

A recognition system using displacement extraction based on directional features is proposed for handwritten Chinese characters. In the system, after extracting the features from an input image, the displacement is extracted by the minimization of an energy functional, which consists of the Euclidean distance and the smoothness of the extracted displacement. The coarse-to-fine strategy is adopted to escape local minima and reduce computational costs. The statistical classification is performed based on the estimated variance. In addition, the smoothness of the extracted displacement is utilized. An improvement in recognition performance is achieved as compared with the method without displacement extraction.

A handwritten character recognition system using hierarchical displacement extraction algorithm

A handwritten character recognition system using the hierarchical algorithm to extract displacement between a template pattern and an input pattern is proposed. In the proposed system, the displacement can be computed by Gauss-Seidel iteration derived from Euler-Lagrange equations of the energy functional, which consists of a correspondence error between patterns and a smoothness constraint of the extracted displacement. To extract both global and local deformations included in input patterns, the hierarchical structure is introduced. In computer experiments, the recognition performance is clarified. In addition, the relation between the ability of displacement extraction and the recognition performance when the correspondence error is used as the distance is discussed. Finally, we show that it is possible to improve the recognition performance by using both the correspondence error and the smoothness of the extracted displacement as the distance.

CUDA Implementation of Deformable Pattern Recognition and its Application to MNIST Handwritten Digit Database

In this study we propose a deformable pattern recognition method with CUDA implementation. In order to achieve the proper correspondence between foreground pixels of input and prototype images, a pair of distance maps are generated from input and prototype images, whose pixel values are given based on the distance to the nearest foreground pixel. Then a regularization technique computes the horizontal and vertical displacements based on these distance maps. The dissimilarity is measured based on the eight-directional derivative of input and prototype images in order to leverage characteristic information on the curvature of line segments that might be lost after the deformation. The prototype-parallel displacement computation on CUDA and the gradual prototype elimination technique are employed for reducing the computational time without sacrificing the accuracy. A simulation shows that the proposed method with the k-nearest neighbor classifier gives the error rate of 0.57% for the MNIST handwritten digit database.

Intelligent control for pneumatic servo system

This paper presents a novel model reference adaptive control (MRAC) incorporating neural network (NN) for the pneumatic servo system. In the proposed method, the input of the NN are different from which in the conventional method. And there is no need to use the inner parameters of the NN during the learning of the NN. The effectiveness of the proposed scheme is confirmed by experiments using the existent pneumatic servo system.

Image edge detection with discretely spaced FitzHugh-Nagumo type excitable elements

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.

Model reference adaptive control using delta-operator with neural network for pneumatic servo system

Pneumatic servo systems have non-linear elements essentially caused by air compressibility, various frictions and so on. Therefore it is difficult to accomplish savory control performance by conventional adaptive control. In this paper we propose a design scheme which combines a model reference adaptive control using delta-operator with neural network for pneumatic servo system to overcome the influence of non-linearity.

Handwritten digit recognition by hierarchical displacement extraction with gradual prototype elimination

This paper investigates the performance of a handwritten character recognition method by hierarchical displacement extraction based on NIST special database (HSF7). The method is composed of a displacement extraction technique and a coarse-to-fine search strategy. In the displacement extraction technique, the displacement (correspondence) between an input pattern and a prototype is iteratively computed by minimizing a functional defined in the framework of regularization theory. In the coarse-to-fine search strategy, the above-mentioned displacement is determined with multiresolution images so as to avoid the pitfalls of local minimum and make the number of iterations low. In addition, a new idea is proposed for reducing the cost of computation without degrading the recognition performance, in which the number of candidate prototypes is eliminated gradually through the hierarchical procedure of the coarse-to-fine search strategy.

Statistical displacement analysis for handwriting verification

In this paper, it is assumed that each writer has his or her own statistics of handwriting displacement, therefore a statistical displacement analysis for handwriting verification is proposed. Here, a regularization method with the coarse-to-fine strategy computes the displacement function in questionable handwritten letters, and then it is normalized to remove the noisy displacement that arises from the position drift and scaling variation. Finally, the normalized displacement function and the statistics of displacement obtained in advance from registered authentic letters are used to calculate the distance from a standard handwritten letter to a questionable one. A fundamental simulation was conducted in order to evaluate the performance of the proposed method.