Research Of Quality Assessment For Compressed Images

Compressed image are widely used all over the world, for the compression algorithm approvethe small storage space and the ease of transmission. With the variation of the compressiondegrees, different distortions appear in the images. Quality assessment for digital images ishelpful for evaluating the compression algorithms, and thus improving these algorithms. Aimingat resolving the problems appear in quality assessment, this dissertation presents two methodsuseful for referenced and non-referenced assessment for compressed images. Total studycontains the following contents:1. Blind assessment based on multi-scale decomposition for compressed image. With theincrease of compression degrees, the image blocking effects also increase. Lossy of details andtextures in the image lead to reduction of sharpness of image edges, which is a kind of fuzzydistortion. An assessment method based on multi-scale decomposition is proposed in thisdissertation. We use the Non-subsampled Contourlet Transform on multi-scale decomposition ofimage edge, and investigate how the coefficients change due to different compression degrees.Combined with the smoothness characteristic factor extracted from one dimensional DCTcoefficient, we construct an MDSCORE system for compressed image assessment. Experimentalresults show that this system has a good assessment performance, and a better robustnesscorresponding to different processing operations such as low-pass filtering and image clipping.2. An improved method of referenced quality assessment for compressed image. PSNR isa common used image assessment method. However, it is not sensitive to blocking effectdistortions. PSNR-B filled this gap accordingly by introducing a blocking effect ration (BER)into PSNR. Nevertheless, PSNR-B is not effective if some blocking effects were already in theoriginal image. To resolve this problem, we propose a generalized metric IPSNR-B with a newconstructed BER factor. Experimental results show IPSNR-B has a better assessment capability.The work was supported by the National Natural Science Foundation of China (61103181)，and Education Innovation Fund (11YZ10).