Local Variance-based Sparse Coding of Still Images

Image compression with learned dictionaries using the K-SVD algorithm has outperformed conventional transformed-based techniques. However, rate-distortion performance may worsen when applying K-SVD to image patches without considering the local features of the image. Within a picture, there are variations in visual attributes that can be efficiently utilized in adaptive image processing approaches. One characteristic that may be used in an adaptive image coding technique is the local variation of picture intensity, which is a significant image characteristic. In this study, the R-D performance of the compression strategy is enhanced by applying adaptive coding depending on the local variance of the image region. The outcomes of the suggested technique are compared with the conventional method of sparse coding. PSNR and the number of sparse coefficients at a given sparsity level are used to check the performance.