A just noticeable distortion based perceptually lossless image compression codec

Abstract

In this study thesis a JND (Just-Noticeable-Distortion)-Measurement will be implemented on top of JPEG-LS while only considering a grayscale bit depth of 8 Bit. This is sufficient to show a proof of concept of combining the JND approach with JPEG-LS. JPEG-LS is a widely used and relatively simple coding mechanism for lossless and near-lossless image compression. The JND measurement will be defined, implemented and integrated into JPEG-LS. Therefore a modified approach of [CL95] will be used. The quantization step size (QSS) will be adapted dynamically according to the JND value so that the compression ratio compared to a standard JPEG-LS implementation can be improved. By that, a near-lossless variable bit-rate (VBR) is introduced into the coding flow. The encoder and decoder are implemented in MATLAB and after defining the visual quality criteria the results are evaluated and analysed in matter of compression quality and performance. Two test picture sets are used and the perceptual quality of the codec will be evaluated by an Mean-Opinion-Score (MOS) test and Multi-Scale Structural Similarity (MS-SSIM) test.