关键词: 图像表示, 图像压缩, 梯度域重构, 小波变换, 四叉树, Bandelet变换

Abstract:

This paper presents a novel image representation and compression framework using edge and gradient fields. The original image was represented with the pixels in edge region and the gradient fields in non-edge region, and they were encoded separately. At the encoder side, the edge region map was encoded using an adaptive quadtree decomposition and 1D wavelet transform, and the gradient fields were encoded using standard 2D wavelet coder. At the decoder side, a gradient domain variation reconstruction method was utilized to restore the decoded image from the resulting edge region and gradient fields after inverse transform. The proposed framework can provide both lossless and lossy image compression in high compression rate, and can avoid the so-called ring phenomena near edges effectively.

Key words: image representation, image compression, gradient domain reconstruction, wavelet transform, quadtree, Bandelet transform