This paper realize a new coding in which images are constructed by a set of many kinds of unit curved surfaces using a vector quantization technique, and presents a design of vector quantization coding based on PQS(Picture Quality Scale) as a distortion measure. When we chose typical and essential curved surface units as initial codebook to construct an image, we have obtained a coding performance better than the performance in the case of splitting algorithm both in the case of open data and closed data. Our idea will develop into enhancement and modification of images. Furthermore we can expect the consideration of the structure of images and synthesis of images in this report will develop to Computer Graphics.
It is one of the important problems on subband image coding how to encode high frequency subband signals efficiently as possible, under bit allocation constraints. High-frequency subbands have some features preventing from efficient coding, i.e. their signal energy is smaller than that of low-frequency band, but widely spread in local area on spacial domain. In this paper, a subband image coding using inter-band correlation of signal distribution is described. Simulation results show the proposed method has attained good coding performance.
In this paper, biorthogonal wavelet transform of an image is realized by a 2-D octave splitting, separable multirate filter bank. Then vectors are constructed by samples picking up from the same positions in each subband and quantized. This coding is compared with the uniform splitting subband coding with vector quantization. The result is that the former gives the same level quality of an image as the latter with reduced computation complexity.
This paper presents a method to acquire high resolution images by integrating corresponded pixels over a series of images. To estimate accurate correspondence between the pixels in the two temporally serial image, we propose a new block matching method applicable to the motion with affin transformation. The method employs an iterative technique which analysis blocks in the image is transformed toward decreasing mean square error(MSE) in the block, and thus enable analysis blocks to track over more frames compared to usual method. In this paper, we apply this method to a series of 50 frames and improved the resolution in a image.
This paper proposes a new region segmentation scheme of the colored still image in order to apply to the structural description. At first, RGB image signal is mapped into HVC color space which is uniform and inherent to human perception. The region segmentation algorithm applies color difference defined in HVC as the scale. We establish the new algorithm of self-organizing clustering so as to get rid of conventional initial value problem. By the segmentation described above, it is clarified that our new proposed scheme is stable and perform a good segmentation.
HVC color space(Munsell Renotation Color Space) consists of three atributes(Hue, Value, Chroma) according to the color perceptual property of human beings. This paper proposes systematic examination for Colorization based on HVC color space. Colorization is the method of coloring monochrome image. To genarate color images from monochrome images, it must add Hue and Chroma to Value of segmented region in monochrome image. Therefore, the analysis of HVC features in segmented regions of color image is important examination, the results are used to knowledges for Colorization.