The Directional Adaptive Red-Black Wavelet Transform for Image Coding

We present a novel 2-D directional adaptive red-black wavelet transform (DA-RBWT). This approach uses three or four pixels to adaptively predict the value of middle pixel according to the correlation among its neighboring pixels. Compared to red-black wavelet transform (RBWT) on a rectangular grid, DA-RBWT shows more directions for its adaptability to local image features, especially for the regions of edges and textures. Experimental results show that DA-RBWT outperforms conventional RBWT in the application of image coding in PSNR. The construction from the DA-RBWT better represents the structure in the image and is visually more pleasing.