Deep Learning Based Research On Depth Map Super-Resolution Reconstruction

With the development of three-dimensional technology,the three-dimensional technology will be applied in the development of national science and technology and every aspect of our daily life.It has important significance to research the three-dimensional technology.Depth information is the basis for the application and development of threedimensional technology.Therefore,high-quality depth information is very important to guarantee the three-dimensional technology provide a comfortable experience,which can make the three-dimensional technology more widely used and rapidly developed.However,due to the limitation of hardware conditions,the depth maps captured by the existing depth camera have low resolution.The low-resolution depth maps cannot be well combined with the corresponding high-resolution color images,thus cannot obtain a high-quality virtual three-dimensional scene in the application of three-dimensional technology.Therefore,the study of depth map super-resolution reconstruction is of great significance in the development of three-dimensional technology.In this paper,think of the correlation information between the low-resolution depth map and the corresponding high-resolution color image,the depth map super-resolution reconstruction is studied based on the deep learning method.1)This paper proposes a depth map super-resolution reconstruction algorithm based on fully edge-augmented guidance.Considering the correlation between lowresolution depth map and high-resolution color image,a convolutional neural network structure with two branches is designed.The two branches extract features from lowresolution depth maps and high-resolution color images respectively.Between the layers with equal levels,the low-resolution depth map is fully guided by the features of the high-resolution color image.The features are fused in the convolutional neural network and the high-resolution depth map is reconstructed.In addition,considering the highresolution color image contains much redundant information,an edge augment method is proposed,which can accelerate the convergence speed of the network and improve the quality of the result.The experimental results show that the proposed method can complete the depth map super-resolution task and obtain sharp and clear reconstruction results.2)This paper proposes a depth map super-resolution reconstruction algorithm based on structured attention guided convolutional neural fields.Thanks to two designed convolutional neural network structures with strong expressive ability,features are extracted from low-resolution depth maps and high-resolution color images.In addition,the conditional random field is introduced in the proposed convolutional neural network structure,and the attention mechanism model is integrated into the conditional random field.The model can automatically focus on the texture structure of the depth map in the convolutional neural fields.The features in the two branches can be guided and fused with each other in the convolutional neural network,then obtain the final reconstruction result.Experimental results show that the method can reconstruct high-quality high-resolution depth maps with accurate texture details.