Abstract:

A noreference image quality assessment metric for JPEG2000 code images is presented. First, the edge points in the structuretexture region of the reconstructed picture are detected based on the principle of Human Visual System (HVS). Then according to the gradient direction of the edge point, the edge feature is extracted, which is comprised of the pixels in the neighborhood of the edge point and is perpendicular to the gradient direction. Finally, a noreference quality assessment model based on OLSRBF network is trained to predict the quality of JPEG2000 coded images by using the image feature and the image subjective quality rating. Experimental results show that, compared with the metrics proposed by Ong and by Marziliano, the performance of the proposed metric is more consistent with subjective evaluation.

Key words: communciation, image quality assessment, noreference, human visual system, radial basis function, JPEG2000