Tetsuya Izumi

Characters recognition method based on vector field and simple linear regression model

In order to obtain a low computational cost method (or rough classification) for automatic handwritten character recognition, the paper proposes a combined system of two feature representation methods based on a vector field: an autocorrelation matrix; a low frequency Fourier expansion. In each method, the similarity is defined as a weighted sum of the squared values of the inner product between the input pattern feature vectors and the reference pattern ones that are normalized eigenvectors of a KL (Karhunen-Loeve) expansion. The paper also describes a way of deciding the weight coefficients using a simple linear regression model, and shows the effectiveness of the proposed method by illustrating some experimental results for 3036 categories of handwritten Japanese characters.

Extended SPRT for structural change detection of time series based on a multiple regression model

It is important to detect a structural change in a time series quickly as a trigger to remodeling the forecasting model. The well-known Chow test has been used as the standard method for detecting change, especially in economics. However, we have proposed the application of the sequential probability ratio test (SPRT) for detecting the change in single-regression modeled time-series data. In this article, we show experimental results using SPRT and the Chow test when applied to time-series data that are based on multiple regression models. We also clarify the effectiveness of SPRT compared with the Chow test in its ability to detect change early and correctly, and its computational complexity. Moreover, we extend the definition of the point at which structural change is detected with the SPRT method, and show an improvement in the accuracy of change detection.

Early structural change detection as an optimal stopping problem: solution theorem and its proof using reduction to absurdity

The change point detection (CPD) problem in a time series is when it is found that the structure of the data being generated has changed at some time and for some reason. We have formulated structural change detection in a time series as an optimal stopping problem using the concept of dynamic programming (DP), and we present the optimal solution and its correctness by numerical calculations. In this article, we present the solution theorem and its proof using reduction to absurdity.

Fast Kansei Matching Method as an Algorithm for the Solution of Extended Stable Marriage Problem

This paper proposes a fast iteration algorithm for Kansei matching as an algorithm for solving the general stable marriage problem, which is easier and more transparent than the conventional (extended) Gale-Shapley (GS) algorithm in the sense of programming and debugging. This paper also presents a fast version of the iteration algorithm and describes the result of comparative experimentation in execution time. The result shows that the proposed algorithm executes more than six times faster than the GS one, while it requires the same memory storage as GS one, and this proves the effectiveness of the iteration method.

Experimental Investigation of Feature Quantity in Sound Signal and Feeling Impression Using PCA

This paper describes experimental investigation of the relationship between feature quantity of sound signal and feeling impression using PCA (Principal Component Analysis). As the feature quantity, we use Fluctuation value and sum of squared errors (Residual) which is calculated by regression analysis of sound signal, in the same way as our previous paper. In order to investigate the feeling impression and effect from sound signal, we use a questionnaire survey method, that is, we ask some examinees to evaluate their feeling impression about sound (music) that we provide. As a result, we have found that the feeling response of examinees can be classified into three groups by a clustering analysis. And we have verified the feeling impression effects depending on each group of examinees and four kinds of frequency zone of sound signal from the results of PCA. In this paper, we also discuss the analysis results on the Kansei (or feeling) effect.

Application of SPRT to an image data sequence for a remote monitoring system

Recently, remote monitoring camera systems have been widely used for security. In such systems, one important function is that the system automatically detects any change in the scenes from the monitoring cameras. In wireless remote monitoring camera systems, the images of the scenes are generally transmitted as compressed data (e.g., JPEG file), because of the capacity of the wireless channel. This article shows the automated detection of the change point in time-series data of compressed JPEG file quantity (Kbytes) from the monitoring camera by applying the sequential probabilistic ratio test (SPRT) and the Chow test, which is well known as a standard method for detecting structural change in time-series data.

Face identification for people image with general background using vector phase field

This paper proposes an automatic identification method of an acquaintancepsilas face from people image with general background. In this method, we assume that a face in a given image approximately equals to be an Affine transformed (rotated, enlarged/reduced and translated) pattern of registered original one, and that the face pattern is also perturbed by a lighting variation and noise. The recognition method deals with a vector phase field (VPF) that is a normalized gradient vector field obtained from input grey level image. The VPF shows a kind of feature representation for face pattern, while it gives an insensitive feature to lighting variations on the input image. In addition to the representation, we use a region-weighted similarity over the VPF in order to improve the identification accuracy. This paper also presents the experimental results of the proposed method, and illustrates its effectiveness by comparing with non region weighted case.

Characters recognition using vector field and linear regression model

In order to obtain a low computational cost method (or rough classification) for automatic handwritten character recognition, this paper proposes a combined system of two feature representation methods based on a vector field: one is autocorrelation matrix, and another is a low frequency Fourier expansion. In each method, the similarity is defined as a weighted sum of the squared values of the inner product between input pattern feature vector and the reference pattern ones that are normalized eigenvectors of KL (Karhunen-Loeve) expansion. This paper also describes a way of deciding the weight coefficients using a simple linear regression model, and shows the effectiveness of the proposed method by illustrating some experimentation results for 3036 categories of handwritten Japanese characters.