The Super-resolution Algorithm Based On Image Structure Learning Of The Robot Visual

Robot vision system is a simulation of the human visual system,collecting all kinds of scene image information,which plays a key role in the practical application.In the process of the image acquisition on robot vision system,which is disturbed by the noise,light and storage,most image resolution is low,reducing the quality of the image.The low quality and low resolution images acquired by the robot vision system can not meet the needs of the application.The problemof the low quality and low resolution images of various scenes has become the bottleneck of the practical application.Through research on the image segmentation and reconstruction of the image processing,in view of the practical application on the robot,the corresponding improvement methods is provide.In the low quality and the low resolution of the robot images,the structure information is used to improve the quality of images.Through image segmentation percepts the image region,the structure information is used to improve the image resolution by the super-resolution methods.The image acquisition of the robot system is affected by noise environment,reducing the quality of image segmentation,robust image segmentation via low-rank matrix recovery is proposed.The algorithm based on the specific method of overlapping image block uses the low-rank matrix recovery to obtain the teature image.Then,using the graph cuts based on min-cut/max-flow,the feature space image is segmented to obtain the best results.Experiments results on Berkeley Segmentation Data Set show that:compared the Berkeley method and the graph cuts,the algorithm has some advantages of the robustness of noise.It has better segmentation on subjective and objective.For robot vision system to obtain low resolution images,a novel Locality-constrained low-rank representation based FH method has been proposed.The algorithm enforces the low-rank constraint on choosing dictionary atoms that belongs to a subspace corresponding to same cluster.Then,it imposes a locality constraint on choosing dictionary atoms that are nearest neighbor atoms.Experiments have been conducted on face image dataset to evaluate the proposed algorithm.Our algorithmoutperforms SR and LCR based methods both on subjective and objective image quality,thus denoting that exploiting the structure information of data is feasible in face hallucination.