|M.Sc Student||Gershikov Evgeny|
|Subject||On Subband Transform Coding for Color Image Compression|
|Department||Department of Electrical and Computer Engineering||Supervisor||ASSOCIATE PROF. Moshe Porat|
Subband transform coders are vastly used for image coding. The performance of such coders, however, has not been analyzed so far for color images, especially when the selection of color components is considered. The common choice for such a color space is YUV or YIQ which is in most cases not optimal. Another aspect of the coding is the allocation of subband rates, which in some coding systems is performed in a way that is not optimal for a given image such as using fixed quantization tables.
In this work we introduce a new rate-distortion model for color image compression, and employ it to find the optimal color components transform (CCT) and optimal rates allocation for the compression. It should be noted that the optimal CCT is image dependent. We thus propose an image independent transform, namely the DCT (Discrete Cosine Transform), motivated by its similarity to the obtained optimal CCT. We show that the DCT can be used to transform the RGB components into an efficient set of color components suitable for subband coding, so that the coding performance is superior to that of the commonly used CCT or the Karhunen Loeve Transform.
The optimal rates, derived from our model, can be used to design adaptive quantization tables for the coding stage with results superior to those presently used. We show that if the encoder complexity is of primary importance, the knowledge of the statistical distribution of the subband coefficients can be employed to reduce the amount of computations required for the calculation of a quantization tables. We also show that our theoretical rate-distortion connection can be used to design applications for rate-control and quality-control of the compression. Accuracy and performance measures are presented for these applications.
Based on both theoretical considerations and simulation results, our conclusion is that the new approach to color image compression can improve presently available methods. The model can be used to achieve both efficient coding and rate-control or quality-control of the compression, and thus contribute to the implementation and theory of color image coding.