Transform image coding using adaptive block size

Abstract

The block transform coding with a large block size has advantages of energy compaction and decorrelation between coefficients for low entropy image. Whereas a coding using a smaller block size is efficient for compression of high entropy and for computation simplicity. Therefore we will expect that adaptive block transform coding is better than the non-adaptive block transform coding. In this paper, a discrete cosine transform (DCT) coding based on two kinds of transform block is proposed and demonstrated. The scheme adaptively selects the transform block size according to the variance of luminance in each maco-block. If the variance (standard deviation) is not bigger than a threshold, the luminance block is transformed, quantized and coded using a 16x16 block (Algorithms II). The other macro-blocks are divided into four 8x8 blocks, which are transformed, quantized and coded using 8x8 block (Algorithms I). This paper used two rectangular block sizes, whose sizes are 8x8 block and 16x16 block. Simulation results are presented and compared to the results using a fixed block size. It is shown that 0.09 dB in Foreman and 0.67 dB in Miss America are better than conventional algorithms.