Image Quality Assessment Based on Binary Structure Information

An accurate assessment of image quality is crucial to the success of many image processing systems, where the numerical outcome evaluated by objective measurement is expected to be consistent with the subjective assessment made by human being. In this paper, a new image quality metric based on measuring the similarity of the structural information inherent in digital images is proposed. The structural information is defined geometrically and statistically in a block basis, where geometrical structure information is obtained from the binary quantization that preserves the first two moments of the image block. Statistical structure information includes the luminance mean and contrast calculated from the pixels of two different groups after the binary classification. To verify the validity of the proposed metric, the correlation between objective and subjective scores is inspected and compared with that obtained by famous MSSIM, in which a large amount of test images in LIVE database are assessed. The cross-distortion test results show that the proposed metric outperforms MSSIM in judging the distorted images corrupted by JPEG2000, Gaussian blurring, white noises and fast fading and has the performance close to MSSIM in judging the distorted images corrupted by JPEG.