A full reference image quality assessment method for retina-like sensor

Abstract

Retina_like sensor is characterized by a space-variant resolution mimicking the distribution of photoreceptors in the human retina. It is devided into two areas-the central area and the peripheral area. Density of pixels is highest in the center and decreases monotonically toward the periphery area. Such space-varant image allows high-resolution tasks using the central region while maintaining a lower resolution part providing relevant information about the background. In high speed forward motion field, because the image system is approaching or leaving the object in a high speed, the recorded image will be blurred radially. However, this kind of radial blur can be reduced by changing the pixel layout of retina_like sensor. So image quality assessment studies carried out on the different structures of the retina_like sensor output image can provide theoretical guidance for the establishment of the optimal layout of the retina_like sensor. This paper first analyzes the distortion process of the retina_like sensor output image in high-speed forward motion and find that such a distortion of the image including under-sampled distortion and radial blur distortion. Based on the characteristics of the distorted image, the author puts forward a full reference image quality assessment method, named Qretina, for such images. The method is according to the distortion process. Do under-sampling evaluation and radial fuzzy evaluation firstly, then weigh the two parts to get the final image quality evaluation results. The experimental results show that Qretina has better performance than the SSIM.