Image quality evaluation method based on structural similarity

Aiming at solving the limit of cur-rent distortion sensitivity analysis(HVS is a complicated non-linear system, while the vision models current are linear and simple), we research a new image quality evaluation method based on structural similarity, that is, to get a general similarity from luminance, contrast and image construction, as an objective quality evaluation criteria. In this way, the method fully considers both image structure information and human vision characteristics. Based on human visual comprehension of image content, the method evaluates the subjective human visual perception to image quality by arithmetic modeling, so it ensures the structural similarity model matches the application purpose of image processing. After theory deduction and algorithm validation, the method provides reasons to select a proper image compression algorithm and gives a way to evaluate image quality efficiently. Experiments show that, to evaluate reconstructed images encoded by compression algorithm Set Partitioning in Hierarchical Trees (SPIHT), compared with the traditional evaluation method based on Peak Signal-to-Noise Ratio (PSNR), the method proposed in this paper is more effective to the perception of people's eyes.