Tadaaki Hosaka

Statistical mechanics of lossy data compression using a nonmonotonic perceptron

The performance of a lossy data compression scheme for uniformly biased Boolean messages is investigated via methods of statistical mechanics. Inspired by a formal similarity to the storage capacity problem in neural network research, we utilize a perceptron of which the transfer function is appropriately designed in order to compress and decode the messages. Employing the replica method, we analytically show that our scheme can achieve the optimal performance known in the framework of lossy compression in most cases when the code length becomes infinite. The validity of the obtained results is numerically confirmed.

Lattice Constants, Electrical Resistivity and Specific Heat of RNiC2

Anomalous temperature dependence of electrical resistivity was found to be ubiquitous among the compounds RNiC 2 with R = La, Ce, Nd, Sm, Gd, Tb and Er. Lattice parameters were also observed to exhibit anomalous temperature dependence for some of the compounds studied by X-ray diffraction. It is argued that the existence of charge modulated structures is the cause of the former anomalies. The latter anomalies, on the other hand, are likely to be related to the coupling between the lattice and the f-electrons. Results of heat capacity measurements are used to estimate crystal field levels in NdNiC 2 and SmNiC 2 .

Statistical Mechanics of Source Coding with a Fidelity Criterion

We provide a method to evaluate the typical performance of lossy data compression schemes for general (discrete or continuous) memoryless sources using the replica method (RM). The proposed method reproduces a known formula to compute the rate-distortion function representing the optimal tradeoff in the limit of infinite data lengths between the compression rate and permissible distortion level, which is consistent with existing methods in information theory literature. The advantage of the RM-based method is the ability to accurately assess the performance of sub-optimal code ensembles, demonstrated here for Gaussian memoryless sources. The obtained result is used to construct a family of error correcting codes that are composed of practical-size codebooks and asymptotically achieve the capacity of the Gaussian channel.

HLAC Approach to Automatic Object Counting

Counting (identical) objects in images is a simple yet fundamental recognition task that requires exhaustive human effort. Automation of this task would reduce the human load significantly. In this paper, we propose a statistical method to automatically count objects in an image sequence by using higher-order local auto-correlation (HLAC) based image features and multiple regression analysis (MRA). This method is based on a simple computation, which enables fast and automatic object counting in real time. We propose several methods that have different preprocessing and image features and conduct comparative experiments of counting objects (ducks in this paper) in images captured by outdoor monitoring cameras. The experimental results demonstrated the effectiveness of the proposed methods.

Balancing with Noise and Delay

Motivated by recent studies in human balance control, we study a delayed random walk with an unstable fixed point. It is observed that the random walker moves away from the unstable fixed point more slowly than is observed in the absence of delay. It is shown that, for given a noise level, there exists an optimal delay to achieve the longest first passage time. Our observations support recent demonstrations that noise has a beneficial role for balance control and emphasize that predicitive strategies are not necessary to transiently control balance.

Statistical Mechanical Approach to Error Exponents of Lossy Data Compression

We present a scheme to accurately evaluate the error exponents of a lossy data compression problem, which characterize average probabilities over a code ensemble of compression failure and success ...

Image matting based on local color discrimination by SVM

Image matting is a technique used for extracting a foreground object in a static image by estimating the opacity, called alpha matte, at each pixel in the foreground image layer. The common drawback of the previous matting approaches is the decrease in performance when a foreground and its background have similar colors. In order to overcome this problem, we propose a method of estimating alpha mattes by using the color information of neighboring pixels and the support vector machine. We define a cost function on the basis of a Markov random field by considering not only a single pixel but also its neighboring pixels and utilizing the support vector machine to enhance the discrimination between the foreground and the background. This cost function is minimized by the belief propagation and the sampling methods. Qualitative and quantitative results have shown a favorable matting performance compared to the other methods.

Statistical mechanical approach to lossy data compression: Theory and practice

The encoder and decoder for lossy data compression of binary memoryless sources are developed on the basis of a specific-type nonmonotonic perceptron. Statistical mechanical analysis indicates that the potential ability of the perceptron-based code saturates the theoretically achievable limit in most cases although exactly performing the compression is computationally difficult. To resolve this difficulty, we provide a computationally tractable approximation algorithm using belief propagation (BP), which is a current standard algorithm of probabilistic inference. Introducing several approximations and heuristics, the BP-based algorithm exhibits performance that is close to the achievable limit in a practical time scale in optimal cases.

Object Detection Using Background Subtraction and Foreground Motion Estimation

A method for detecting moving objects using a Markov random field (MRF) model is proposed, based on background subtraction. We aim at overcoming two major drawbacks of existing methods: dynamic background changes such as swinging trees and camera shaking tend to yield false positives, and the existence of similar colors in objects and their backgrounds tends to yield false negatives. One characteristic of our method is the background subtraction using the nearest neighbor method with multiple background images to cope with dynamic backgrounds. Another characteristic is the estimation of object movement, which provides robustness for similar colors in objects and background regions. From the viewpoint of the MRF, we define the energy function by considering these characteristics and optimize the function by graph cut. In most cases of our experiments, the proposed method can be implemented in (nearly) real time, and experimental results show favorable detection performance even in difficult cases in which methods of previous studies have failed.

Image Matting in the Framework of Quantification IV

Image matting and segmentation, which are used to extract a foreground object from the background, are primary techniques for digital image and video editing. In digital matting, the transparency of the foreground object is considered, while segmentation performs a rigid extraction of the object. Recently, several algorithms for matting and segmentation problems have been proposed and have provided high-quality results. In this paper, we propose a unified formulation for image matting in the framework of the method of quantification IV based on a review of the previous studies from this framework. Our method also utilizes discriminative information provided by a user (a few strokes drawn by the user). The experimental results show a favorable matting performance.