Mixed‐transform image coding with variable block sizes

Abstract

In the proposed image coding, variable block size and two transforms are used in image transform coding. A 16×16 image block is detected to be active or inactive according to activity indexes which are the values of three transform coefficients. If the image block is active, then it is partitioned to four 8×8 image blocks and then processed by a 2‐D 8×8 discrete cosine transform (DCT); otherwise it is processed by a 2‐D 16×16 Walsh‐Hadmard transform (WHT) directly since the WHT has simple structure and the same image compression performance as the DCT does for inactive image blocks. As a result, the proposed approach is to process image blocks by a 16×16 WHT or an 8×8 DCT. If the PSNRs for both the conventional DCT image coding and the mixed‐transform coding are kept the same in our experiments with 9 images, the number of additions of the latter increases by 4.18% on average, but the number of the multiplications decreases by 78.53% at bit rate 0.5.