Yoshinori Izumi

Robust tracking of soccer players based on data fusion

Presents a technique for integrating multiple visual features for tracking moving objects. Our proposed method consists of observation (pattern-matching) units and prediction units, which form a ladder structure. The major feature of our proposed method is that each of the observation units with different pattern matching algorithms is executed step-by-step to innovate the state vector considering the reliability of the observation. The fusion of multiple observations makes the tracks robust to occlusion and to deformation. Experiments with soccer sequences are shown to validate the techniques robustness. Its applications to broadcasting services are also briefly discussed.

Morphological segmentation of sport scenes using color information

We propose a new method for segmenting image sequences in sports scenes containing brisk movement. An important feature of this method is the computation of color histograms in the areas for the turf, which is done interactively during rehearsal. Another important feature is automatic morphological segmentation during broadcast. The morphological segmentation consists of two operations: coarse segmentation by binary reconstruction based on the areas detected by thresholding the color histogram, and fine segmentation by watershed transformation with markers. It is shown that this new method achieves accurate segmentation in sport scenes.

Image enhancement based on estimation of high resolution component using wavelet transform

An image enhancement method based on multi-scale wavelet representation is discussed. Each edge in an image is analyzed by multi-resolution analysis and is classified according to the pattern of neighboring right and left edges together. A high definition image and its band-limited image are used for learning the edge relation between high resolution and low resolution. By switching the transition rule that is adapted to the input image, generating high resolution components, and using the most suitable enhancement method for each class, natural enhancement without artifacts is realized.

Detection of image orientation with mathematical morphology and orientation‐adaptive image processing for antialiasing

We propose a new method of detecting image orientations by using mathematical morphology, a type of nonlinear processing. In the new method, an opening filter with a one-dimensional structure along each orientation axis detects image orientations after detection of oriented edges in the images. Area widening is performed on the detected orientations, changing the widening parameters according to the length of the orientation. A combinational logic process, in which output values are determined in logical comparison with other orientation signals, makes the results stable and flawless. We obtained good orientation information when we applied this method to a MUSE decoder. Moreover, adaptive processing with the results of orientation detection has enabled a MUSE decoder to decode MUSE signals with fewer artifacts than a conventional decoder.

A Consideration about Improving the Picture Quality of the HDTV Broadcasting System

Abstract HDTV broadcasting, which uses a multiple sub-sampling method, is being conducted in Japan on an experimental basis. In this paper, we discuss a new method to improve the picture quality of the HDTV broadcasting system. We are proposing a new signal processing method for chrominance signals in moving areas. The concept of this method is to process the chrominance signal frame by frame. Here, we will refer to the MUSE system as an example of an HDTV broadcasting system, however it is easy to adapt the idea to other HDTV systems.