Image quality assessment using block-based weighted SVD

Image quality is an important challenge in image processing. The quality measures should be designed in the direction where the correlation between the mathematical evaluation and subjective evaluation is high. We propose a new image quality assessment relying on block-based singular vectors. The corresponded distorted blocks are projected onto the singular vector matrices of the original blocks. These projection coefficients are the main quality attribute. The algorithm is further developed into the reduced reference method. Eigenvectors of the covariance matrix of all original blocks are used as the constant basis to compute the projecting coefficients of all original and distorted blocks. Simulation results on different databases with various distortion types and comparison to state-of-the-art methods show the proposed method in most cases gives the best correlation with human evaluation.