Image quality metric with an integrated bottom-up and top-down HVS approach
Guangtao Zhai, Weikang Zhang, Xiaokang Yang, Yiling Xu · IEE Proceedings - Vision Image and Signal Processing · 2006
A novel image quality assessment technique is proposed that incorporates both a ‘bottom-up’ and a ‘top-down’ simulation of the mechanism of the human visual system (HVS) into a unified quality metric, called top-down/bottom-up quality metric (TBQM). The Importance Map and region-based structural similarity index is employed as the bottom-up and the top-down approach of HVS, respectively. These two approaches are then integrated into the TBQM. Extensive experiments indicate that the proposed metric is highly correlated with subjective quality grading and is significantly better than the widely used image distortion metric, such as mean squared error. This method also outperforms simple top-down structural similarity measures.