A new image compression via adaptive wavelet transform

The conventional two-dimensional wavelet transform is performed through one-dimensional filtering in horizontal and vertical directions, which fails to provide an effective representation for directional features, while natural images are rich in texture and directional features. The wavelet can not achieve ideal sparse representation for natural images, which has some defects in image compression. In this paper, we propose a new adaptive DWT via image texture. Partitioning the image based on local content, predicting the texture direction of local image, keeping the orthogonal property, applying directional wavelet filtering, we achieve our method by lifting structure with in-place operation. In order to protect the texture information, we get half-pixel by interpolating along predicted direction. The method can be used to construct a new image coder with the EBCOT that was adopted in JPEG2000, coding the coefficients and directional information respectively. The result of our method is more effective for representation of natural images. The experiments of compression show that the numerical results and visual effect are improved obviously by our method compared with that of conventional wavelet transform.