Depth Map Super-resolution Reconstruction Algorithm Research

Depth information which as a key content in stereo vision is one of hot researches in the field of image. TOF as a new three dimensional imaging equipment has the advantages of good real-time, high frame rate and without scanning. It is widely used in pattern recognition, robot navigation, three dimensional visualization and other areas. As for the issue of low resolution depth map captured by TOF, we carry out super resolution reconstruction based on fusing depth.map with high resolution color image and image super resolution reconstruction based on sequence depth map captured by TOF. Our contribution is as follows:(1 Introducing the background and research status of super resolution reconstruction based on depth map, then describe the image degradation model, the technology of super resolution and related image quality evaluation standard.(2)As for the issue of edge blurring and texture copying in depth map captured by super-resolution method, we designs and achieve a super-resolution reconstruction method with edge feature-guiding on the RGB-Depth stereo vision system, we extract edge features from depth map through data fusion, then execute super-resolution reconstruction with low revolution depth map guiding. Experimental results demonstrate that our algorithm can protect the image edge structure effectively, also solve the problem of texture copying.(3)Improving a bilateral regularized super-resolution reconstruction method based on weighting term between L1 norm and L2 norm. On the basis of the degradation model of depth map, we use the weighted data item based on L1 norm and L2 norm and bilateral regularization item as the constraint item in the model, then construct an optimization objective function. Finally, we obtain high resolution depth map through iterative by the objective function with steepest gradient descent method. Our method preserve the detail of image effectively and, restrain the image noise to achieve a good result.