Improvements to the JPEG-LS prediction scheme

摘要

In JPEG-LS, simple edge detection techniques are used in determining the predictive value of each pixel. These techniques only detect horizontal/vertical edges and have only been optimized for the prediction of pixels in the locality of such edges. Thus, JPEG-LS produces large prediction errors in the locality of diagonal edges. We propose a low complexity technique that accurately detects diagonal edges and efficiently predicts pixels, based on the information available within the standard predictive template of JPEG-LS. We show that the proposed technique outperforms JPEG-LS in terms of predicted mean squared error, by margins of up to 15%.