Generalized region based transform coding for video compression

Abstract

Summary form only given. Block based transform coding (BBTC) is among the most popular coding method for video compression due to its simplicity of hardware implementation. At low bit rate transmission however this approach cannot maintain acceptable resolution and image quality. On the other hand, region based coding methods have been shown to have the capability to improve the visual quality by the acknowledgment of human perception. In order to take the advantages from both of the coding methods, a novel technique is introduced to combine BBTC and region based coding. Using this technique, a new class of video coding methods are generated and termed region based transform coding (RBTC). In the generalized RBTC, we represent regions containing motion in terms of texture surrounded by contours. Contours and textures are then coded separately. The novel technique is that the pixel values of the regions are scanned to form a vector. Then the vector is further converted to a number of fixed size image blocks. Using this technique, conventional transform coding can be applied on the blocks of texture directly. Contour can be coded using traditional contour coding methods or any other bit plane encoding methods. To prove the idea of this new class of video coding methods, a scheme called segmented motion transform coding (SMTC) is simulated. In SMTC, chain codes are used for contour coding. The simulations are performed using the first 60 frames of both of the CIF formatted "Miss America" and "Salesman" video sequences.