A Study on Video Coding Algorithms based on Region Segmentation using Motion Vectors
Grant-in-Aid for Scientific Research (C).
|Research Institution||Kanazawa University|
HASHIMOTO Hideo Kanazawa University, Faculty of Engineering, Professor, 工学部, 教授 (00251934)
|Project Fiscal Year
1994 – 1995
Completed(Fiscal Year 1995)
|Budget Amount *help
¥2,100,000 (Direct Cost : ¥2,100,000)
Fiscal Year 1995 : ¥300,000 (Direct Cost : ¥300,000)
Fiscal Year 1994 : ¥1,800,000 (Direct Cost : ¥1,800,000)
|Keywords||Motion vector / Region segmentation / Contour extraction / Subband coding / Inter-band correlation / gain / shape VQ / 動きベクトル / 領域分割 / 輪郭抽出 / サブバンド符号化 / バンド間相関 / 多数ベクトル量子化 / ブロックマッチング / ゲイン・シェイプベクトル量子化|
A video coding algorithm based on region segmentation has been studied. Segmentation coding methods are seemed to be better than the conventional waveform coding ones in points of coding efficiency and coded picture quality at a given coding rate. In our study, moving bjects are detected using motion vectors between successive frames, and segmented into different regions. Each object is coded by using the optimally method region by region. Obtained results in this study are as follws.
1. Optimal detection of motion vector
A hierarchical block matching method is applied to subband splitted images in order to obtain motion vectors in high accuracy. Parameters for block matching, for example blocksize, hierarchical stage number, detection thresholds and so on, are optimized. Motion vector detection with a half-pel precision is effective for rapid motion including panning, but block matching in small blocksize is better for fine pictures.
2. Segmentation and contour extraction
Two methods extr
acting moving object's contour are examined.
(1) Motion vectors are detected sequentially by reducing block size from larger one to smaller, and finally obtained in pel-base domain. The discontinuous points of pel-base motion vectors are recognized as moving object's contour.
(2) First, candidate contour blocks which include moving object's contour inside are determined in larger block size matching. Then Chrominance information (Hue) in those blocks is used for detection of exact pel-level contour.
Compared to the former in which noise causes error detection for motion vector, the latter has advantaged in extracting exact contour by employing both temporal and spatial correlation in picture. Coding of contour is performed by applying chain coding etc, but it is left for further study.
3. Subband coding
Subband image coding has also been investigated in relation to the hierarchical motion vector detection. Higher frequency components can be selectively coded without overhead information by using inter-subband correlation in images. In another method, gain/shape vector quantization is applied to a small block constituted over subband images, and high performance as well as complexity reduction are attained. Less
Research Output (15results)