In the hybrid coding of a simple temporal prediction (frame difference) and transform coding, the DCT(Discrete Cosine Transform) is adopted in many existing codecs because of it's good performance. The DCT is superior to other suboptimal transforms in the original image domain, but in the prediction error domain, it may not have as good performance as in the original image domain because the statistical properties of the prediction errors are quite different from those of original image.
In this thesis, the statistical properties of prediction errors are derived and compared with those of original image. Four transforms-KLT, DCT, WHT, HCT, are compared in the prediction error domain by means of measures such as variance of transformed values and rate distortion functions. To confirm the result obtained from above ineasures, some simulations are carried out for the hybrid coding with different quantization schemes.
As a result of simulations, it was confirmed that in both FD(Frame Difference) and DFD(Displaced Frame Difference) domain, the four transforms have very similiar performance because of very low inter-corrlation of prediction errors.