Compression Of Visual Objects Based On Perceptual Validity

The physical validity based lossy compression algorithms are usually accompanied with VOP structural distortion, leading to the degradation of the reconstructed image quality. Considering that many real applications such as Industrial streamline surveillance, traffic flow counting and some web-based interactive digital entertainment only deal with relatively monotonic and similarly posed objects, whose instantaneous physical change of the VOP texture and shape can be overlooked without affecting the system functionality, in this dissertation, the notion of Julesz visual equivalence is extended to the image contents with deterministic structures, e.g. the stationary and/or rigid video objects, and the respective shape and texture equivalent metrics are constructed, and consequently, a novel approach of video compression for rigid or stationary VOP sequences is proposed. The main advantages of this approach are that, the structural completeness can be guaranteed, and under the same parametrical conditions, simultaneous extra gain of the visual quality for each single frame and the compression efficiency, comparing with the existing MPEG-4 counterparts can be achieved. The disadvantages of this approach are that some instantaneous VOP textural and shape information is actively ignored, causing significant difference from its physical origins. The main contributions of this work are as follows.(1)Video compression is generally formulated as the optimization of the curve fitting or point approximation in a high dimensional space, that is, the optimal reconstructed video should be derived with minimal physical image samples and minimal number of fitting parameters. The current physical validity based or Julesz perceptual validity based approaches are only special solutions to this generalized problem.(2)The physical validity based compression uses the fitting strategy of dense point approximation. The limitation of this approach is that it's very hard to find a quantizing model to match perfectly with the structure of generic images, leading to the degradation of the quality after improving its efficiency. Julesz visual equivalence based texture parametrical expressions take the fitting strategy of using statistical constraints over the whole image,...