Image quality evaluation method based on visual saliency and deep neural network

Abstract

The invention discloses an image quality evaluation method based on visual saliency and a deep neural network, and the method specifically comprises the following steps: building an image saliency detection model based on the visual saliency through the color saliency and central region saliency in the visual saliency; generating a color weighted saliency map by utilizing the characteristic that human eyes specially pay attention to colors and the center of the image; utilizing a convex hull principle to obtain a region of the salient object, and generating a convex hull saliency map; fusing the color weighted saliency map and the convex hull saliency map to obtain a final saliency map, and giving an effect picture; according to the method, adopting an LIVE3DIQDphase 1 database as an imagepreprocessing library and a subsequent training library; generating a fused left-eye image and a fused right-eye image: fusing the left image and the right parallax compensation image by taking the left view as a reference to synthesize a single-eye image; generating a visual saliency map for the three-dimensional distorted image, and fusing the generated monocular image and the saliency map thereof; and combining convolution with the neural network to obtain a convolutional neural network.

Description

technical field [0001] The invention belongs to the field of image processing, in particular to the objective evaluation of the quality of stereoscopically distorted images, and relates to an objective image quality evaluation method using a saliency map and a composite image of a stereoscopic image. Background technique [0002] In recent years, the vigorous development of virtual reality (Virtual Reality, VR) technology has brought consumers a more realistic visual experience. As an important part of VR technology, stereoscopic image technology plays an extremely important role in the further development of VR technology, while distortion restricts the progress of stereoscopic image technology. [0003] The problem of distortion of stereoscopic images has always been a research hotspot at home and abroad. Many researchers have put a lot of effort into researching the distortion of stereoscopic images in order to grasp the detailed causes of distortion so as to correct the ...

AI Technical Summary

Problems solved by technology

However, the SSIM algorithm does not take into account the weighting of the area of ​​interest of the human eye, and also ignores the relevant characteristics of the HVS (Human Visual System). Therefore, the objective quality evaluation method consistent with human perception has become a research hotspot.

Images