Hybrid Image Compression by Using Vector Quantization (VQ) and Vector-Embedded Karhunen-Loève Transform (VEKLT)

Abstract

In this paper, a new block-transform-based image compression scheme is proposed by combining vector quantization (VQ) and two transformations, discrete cosine transform (DCT) and vector-embedded Karhunen-Loève transform (VEKLT). First, 8×8 blocks from an input image are normalized and vector-quantized. Then, the difference between the original block and its vector-quantized block is transformed by VEKLT. In parallel, the original block is transformed by DCT. All blocks are classified into two categories (DCT and VEKLT) to minimize arithmetic code length. After that, quad tree decomposition is performed on the binary index image which indicates where a block belongs to one of the two categories. Experimental results show that the proposed scheme outperforms JPEG in peak signal-to-noise ratio (PSNR) and visual quality at high detail.