Miki Haseyama

Audio signal segmentation and classification for scene-cut detection

A scene is regarded as a basic unit of audiovisual material, and thereby the boundaries between two adjacent scenes, which are called scene-cuts, must be detected in advance for audiovisual indexing. This paper proposes a scene-cut detection method. Since scene-cuts are associated with a simultaneous change of visual and audio characteristics, both audio and visual analyses are required for the scene-cut detection. For the audio signal analysis, the proposed method utilizes an audio signal segmentation and classification method using fuzzy c-means clustering, which has been proposed by the authors. For the visual signal analysis, the proposed method utilizes some visual segmentation methods. By using these methods simultaneously, the proposed method can accurately detect the scene-cuts, and thereby it is highly valuable for the preprocessing for audiovisual indexing. Experimental results performed by applying the proposed method to real audiovisual material are shown to verify its high performance.

Restoration method of missing areas in still images using GMRF model

This paper proposes a GMRF (Gaussian Markov random field)-model based restoration method of missing areas in still images. The GMRF model used in the proposed method is realized by a new assumption that reasonably holds for an image source. This model can express important image features such as edges because of the use of the new assumption. Therefore, the proposed method restores the missing areas using the modified GMRF model and can correctly reconstruct the missing edges. Consequently, the proposed method achieves more accurate restoration than those of the traditional methods on both objective and subjective measures. Extensive experimental results demonstrate the improvement of the proposed method over previous methods.

Moving object extraction using a shape-constraint-based splitting active contour model

This paper proposes an efficient moving object extraction method based on an active contour model, which is usually called a Snake. The previous Snakes cannot extract multiple objects inside one contour, since they require that one initial contour can only include one object. Such a restriction becomes a problem for moving object extraction application, because an initial contour may contain more than one object including not only the extraction target but also other objects. To deal with this kind of problem, the proposed method utilizes a new splitting mechanism. By incorporating this mechanism, the proposed Snake can successfully extract each of the multiple objects located inside one initial contour. The experiments of extracting moving vehicles from actual image sequences verify the effectiveness and high performance of the proposed method.

A robust human-eye tracking method in video sequences

An accurate tracking method of the human eyes in video sequence is proposed. The method consists of the following two systems. (1) The first system extracts a region including the both eyes by using a statistical processing. By the processing, the extracted region does not include the other features, especially such as the hair. (2) The second system, which is realized with the circle-frequency filter (CFF), locates each eye position from the region extracted by the first system. Since the output of the CFF is robust to luminescent noise near the eyes, it can successfully obtain the exact eye-location inside of the above region without any noise effect. Consequently, the combination of these two systems makes the proposed method correctly track the eyes without any learning schemes, templates, and geometric relations of the other facial features. Experimental results verify that the proposed method achieves accurate eye-tracking.

Quality improvement technique for JPEG images with fractal image coding

This paper proposes a quality improvement technique for JPEG images by using fractal image coding. JPEG coding is a commonly used standard method of compressing images. However, in its decoded images, quantization noise is sometimes visible in high frequency regions, such as the edges of objects. Hence, in order for the JPEG coding to become a more powerful image-coding method, the JPEG image quality must be improved. Therefore, our method solves this problem by adding the obtained codes by the fractal image coding to improve the image quality. Some simulation results verify that the proposed method achieved higher coding-performance than the traditional JPEG coding.

Reconstruction method of missing texture using error reduction algorithm

This paper presents a novel reconstruction method of missing textures using an error reduction algorithm which is one of phase retrieval methods. The proposed method estimates the Fourier transform magnitude of the missing area from another area whose texture is similar in the obtained image. In order to realize this, a novel approach that monitors the errors caused by the error reduction algorithm is introduced into the selection scheme of the similar texture. Further, the proposed method estimates the phase of the target area by using the error reduction algorithm modified for the texture reconstruction and can restore the missing area accurately. Some experimental results show that the proposed method achieves more accurate restoration than that of the traditional methods.

An improved GRAS algorithm using Archimedes's spiral

In this paper an improvement is proposed to solve a problem in the object recognition using GRAS: when an object is rotated, the change may not be acceptable by GRAS in most cases. The proposed algorithm searches the shape of a given object to define some nodes by Archimedess spiral, and transforms the nodes into a directed graph. Then, it defines some interrelated graph characteristics to compare the given object with others. The performance of the proposed algorithm is presented via experimental results which compare the recognition accuracy of GRAS and proposed algorithm

GA-based applications for routing with an upper bound constraint

This paper presents a method of searching for the shortest route via the most designated points among the routes whose lengths are less than the upper bound using a genetic algorithm (GA). If chromosomes whose route lengths exceed the upper bound are simply screened out in the GA process, the optimization probability gets worse. For the purpose of solving this problem, this paper proposes a new fitness function including an upper bound constraint which can be flexibly changed during the searching process. By using this function, the optimum is efficiently obtained and the optimization probability can be raised. Furthermore, the effectiveness of the proposed method is verified by experiments, applying it to the actual map data.

Convergence Properties of a CORDIC-Based Adaptive ARMA Lattice Filter

This paper presents a theoretical convergence analysis of a CORDIC-based adaptive ARMA lattice filter. In previous literatures, several investigation methods for adaptive lattice filters have been proposed; however, they are available only for AR-type filters. Therefore, we have developed a distinct technique that can reveal the convergence properties of the CORDIC ARMA lattice filter. The derived technique provides a quantitative convergence analysis, which facilitates an efficient hardware design for the filter. Moreover, our analysis technique can be applied to popular multiplier-based filters by slight modifications. Hence, the presented convergence analysis is significant as a leading attempt to investigate ARMA lattice filters.

Steady-State Properties of a CORDIC-Based Adaptive ARMA Lattice Filter

This paper analyzes the steady-state properties of a CORDIC-based adaptive ARMA lattice filter. In our previous study, the convergence properties of the filter in the non-steady state were clarified; however, its behavior in the steady state was not discussed. Therefore, we develop a distinct analysis technique based on a Markov chain in order to investigate the steady-state properties of the filter. By using the proposed technique, the relationship between step size and coefficient estimation error is revealed.