Quality Assessment of Screen Content Images Based on Convolutional Neural Network Considering Binocular Visual Mechanism

In this paper, a dual-pathway convolutional neural network (CNN) considering binocular visual mechanism is proposed to simulate the perceiving characteristics from binocular vision. Because the different sensitivity of retinal photoreceptor cells to RGB colors and the characteristics of human visual attention are more important in binocular vision mechanism, an attention module is used to weight the RGB channels and theirs patial position on each channel. Meanwhile, considering that 3D convolution can learn inter-frame information well, we use 3D convolution to extract the correlation features between RGB channels. Furthermore,considering the importance of optic chiasm in human visual system (HVS), a new cross-method is design to simulate the optic chiasm. Besides, since the characteristics of multi-scale and multi-level are indispensable to perception of any objects, a new multi-scale and multi-level feature fusion (MSMLFF) module is constructed to extract the perceptual features at different scales andlevels. The experimental results illustrate that the proposed method outperforms several state-of-the-art SCIs quality assessment methods.