Parallax Attention Based Stereoscopic Image Color Correction
ZHENG Yu-ming1,FAN Yuan-yuan1,NIU Yu-zhen1,2
1(College of Computer and Data Science,Fuzhou University,Fuzhou 350105,China)2(Key Laboratory of Spatial Data Mining & Information Sharing,Ministry of Education,Fuzhou 350105,China)
Abstract:The difference between left and right views of stereo image affects the reconstruction of depth information by visual task.The color correction effect of the existing stereo image color correction methods is not balanced with the overall time efficiency.To solve this problem,we proposed an end-to-end stereoscopic image color correction method based on parallax attention.This method simultaneously performs stereo correspondence acquisition and target image color correction.First,a multi-scale feature extraction module with shared parameters is used to extract the features of the reference image and the distorted target image respectively.Then the corresponding relationship between features is matched and fused through the bidirectional matching strategy of the parallax attention module.Finally,the fused feature is reconstructed to obtain the corrected target image.The parallax attention mechanism only searches for matching pixels along the horizontal parallax direction,which can achieve efficient integration of the left and right views′ information.Experimental results show that the proposed model is more effective than the existing methods,and can obtain higher quality color correction results while maintaining high time efficiency.