A. Ashtari (Iran) and R. Fazel-Rezai (Canada)
Image Quality Assessment, Human visual System, VisualAttention Modeling, color image processing
This paper presents a comprehensive image quality assess ment metric for still images. We take both early vision and visual attention models into account. The early vi sion model is a combination of the models of Wandell[1], Ahumada[2] and Lambrecht[2]. Device characteristics ef fects are also considered. An opponent color space is pro vided by applying true gamma corrections for red, green and blue components and using suitable different filters for each channel. The bottom-up visual attention model of Itti [4] is used to mimic later visual system effects. Several band-oriented selective Gabor filters are used to approxi mate neuron's behavior in cortex. At the end, these two parts of human vision model are combined to produce the new metric. The results are compared with a subjective test on jpeg and jpeg2000 compressed images. A correlation of 72.3% with the subjective test is achieved in comparison with a 66.8% correlation of PSNR.
Important Links:
Go Back
