1992 年 16 巻 4 号 p. 9-16
In transform image coding such as DCT, an image to be encoded is usually partitioned into square-blocks. As this segmentation of the image is carried out independent of its contents, such coding schemes have some problems concerning coding noise which deteriorates reproduced image quality, e.g. blocking-effect and mosquito noise. In this paper. we propose a new coding scheme in which all square blocks are transfigured into variable shape blocks so that their boundaries can fit the contours of the image. And luminance component in each block is encoded using KLT basis-functions derived from the covariance matrix with respect to its block shape. Computer simulation indicates that this coding scheme provides not only higher coding performance but also better reproduced image quality compared to square-block based coding such as JPEG.