Hide/Show Apps

Development and comparison of transforms for prediction residuals of markov-process-based intra prediction

Tekeli, Nihat
Intra prediction is an important tool used in modern intra-frame coding. In intra prediction, a block of pixels are predicted from previously reconstructed neighbor pixels of the block by copying the previously reconstructed neighbor pixels of the block along an angular direction inside the block. The prediction residual block is then transformed with the conventional 2-D Discrete Cosine Transform (DCT). Recently, it has been shown that transforming the intra prediction residuals with an Asymmetric Discrete Sine Transform (ADST) along the prediction direction and the well-known DCT along the perpendicular direction improves the compression performance. More recently, a recursive intra prediction algorithm, obtained by modeling image blocks with 2-D Markov processes, is proposed to improve the conventional copyingbased intra prediction methods. In this thesis, we develop transforms for the intra prediction residuals obtained with these new recursive intra prediction algorithms. Using the 2-D Markov process correlations of each intra prediction mode, we obtain the correlations of the prediction residuals, and numerically compute Karhunen Loeve Transforms (KLT) for each intra prediction mode. We present compression results to compare the derived transforms with the conventional 2-D DCT and the hybrid ADST/DCT within the H.264 reference software.