Research On Color Image Assisted Very Low Resolution Depth Image Reconstruction

The research on depth learning based super-resolution reconstruction of depth images aims to reconstruct high-resolution images from low-resolution images in a short time.This method breaks the cost constraints of hardware devices,obtaining higher resolution images at a lower cost,and obtaining richer distance information.It is applied to multiple fields such as face recognition,geographic information,and medicine.Currently,the most popular depth image super-resolution imaging technology is used to obtain low resolution images for network training by simply downsampling high-resolution images.However,this sampling method has a small scale,and the resulting low-resolution images still contain a large amount of depth and distance information,which is somewhat different from actual application scenarios.The advantages of super-resolution reconstruction networks are not obvious.Moreover,the existing optimization directions mainly focus on mining single scale information from depth images,ignoring many cross scale details,and the reconstruction effect is not ideal.Therefore,the work of this study mainly focuses on the following two aspects.Firstly,in view of the limited sampling scale for low resolution images in existing technologies and the unicity of network mining image features,this paper proposes an optimized reconstruction method for very low resolution depth images,and studies a network that uses high-resolution two-dimensional color images to assist in reconstructing very low resolution depth images.Mining intra scale and cross scale features of color and depth images and their own prior information features in the network,and performing feature fusion.Through iterative operations on the network,the detail accuracy of very low resolution depth image reconstruction is improved.Secondly,due to the non depth information of color images,the reconstructed depth image contains too many non depth details,resulting in quality defects in the reconstructed image.On the basis of the proposed network,this article adds an edge matching module.First,the color image is filtered to remove some non edge information,and then the processed edge information is extracted to match the edge of the depth image.Different image blocks are divided and matched by blocks.For blocks that do not match,this article uses the method of neighborhood supplementation to fill in.The image quality evaluation index for final imaging has increased by approximately 8%.The method of high resolution color image aided reconstruction of very low resolution depth images proposed in this paper leads to higher accuracy and richer details of the reconstructed super-resolution depth images.Comparative experiments can prove that the method in this paper has obvious advantages in image quality indicators compared to several classic super-resolution reconstruction algorithms.