Casas J. Image compression based on perceptual coding techniques. Torres L. Signal Theory and Communications. [download link]: Universitat Politècnica de Catalunya (UPC); 1996. p. 178.

Abstract

This thesis studies image and video sequence coding methods from the point of view of the way the human visual system perceives and understands visual information. The relevance of such study is due, on the one hand, to the important role that visual signals have in our civilization and, on the other hand, to the problem of representing the large amount of data that image and video processing systems have to deal with. Three different approaches have been investigated for the coding of image textures in an advanced compression scheme relying in aspects of visual perception. The first approach is based on image transitions and the interpolation of smooth areas from such transitions. The second one, considers the extraction, selection and coding of meaningful image details. Finally, the third approach studies the efficient representation of homogeneous fine textures that give a natural appearance to the reconstructed images at high compression levels. In order to apply these techniques for still image and video coding, a three component model of the image, that matches the perceptual properties of the human vision, is put forward. The coding approaches subject of research have leaded to the design of three new image analysis and coding techniques. Using non-linear tools from the framework of Mathematical Morphology, three texture coding techniques are developed. In particular, - A "morphological" image interpolation method aimed at the problem of scattered data interpolation. - An empirical subjective criterion for the ranking and selection of image details according to visual perception. - The application of a conventional image coding technique, subband coding, to the coding of arbitrarily shaped image regions (region-based subband coding). These are new texture coding techniques in the field of object-oriented and Second Generation image and video coding schemes. Furthermore, the model of the image that has been investigated follows the line of the last proposals in the framework of MPEG4, the forthcoming coding standard for low bit-rate visual communications, which considers the possibility of content-based manipulation and coding of visual information.

The PhD thesis report can be downloaded from http://hdl.handle.net/10803/6920