Towards Accurate and Efficient Image Quality Assessment with Interest Points

In recent years, the dramatic development of cloud computing, referring to as the applications and services implemented over Internet, has been witnessed and draws many attentions from both academia and industry. Objective image quality assessment (IQA) is fundamental to a broad range of applications throughout the fields of image processing and computer vision. There is a huge desire in exploiting new design of IQA model which is not only accurate but also efficient for fitting the requirement under the background of big data. Many successful models have been built for accurate prediction of the perceptual visual quality, where some typical characteristics of human visual system (HVS) are utilized and incorporated in IQA systems. The well-known foveation effect assumes that the regions around the fixation points are much more attractive to human eyes, thus the quality of these regions would significantly influence the overall visual quality. In this paper, we analyze the correlations between the fixation point and quality assessment by integrating several state-of-the-art interest point detection algorithms into IQAs. Experimental results on public database demonstrate that the addictive information of interest point is helpful for improving accuracy of popular IQA models, and meanwhile dramatically reducing the computational complexity. Furthermore, the parameter impacts on IQA performance are thoroughly analyzed showing that the parameters should be carefully designed for different IQA models as well as viewing conditions.