Improving the k-svd facial image compression using a linear deblocking method

The use of sparse representations in signal processing is gradually increasing in the past several years. In a previous work we proposed a new method for compressing facial images using the K-SVD algorithm, which is a novel algorithm for training overcomplete dictionaries that lead to sparse signal representations. This method was shown to be most efficient, surpassing the JPEG2000 performance significantly. In this paper we present a significant addition to our compression algorithm in the form of image deblocking. Since the encoding is done in patches, a visually disturbing artifacts of blockiness appear in the reconstructed images. We eliminate these artifacts using a linear deblocking technique, which is based on local image filters. We construct a linear filter for each relevant pixel independently, and apply these filters as post-processing. This method is limited, but nevertheless it improves the PSNR of the reconstructed images and gives visually appealing results.