A Study On The Reduction Of Compression And Specular Artifacts In2D And3D Images

Block discrete cosine transform (BDCT) based coding standards have been widely used for image/video compression. At low bit rates, the BDCT compressed images/videos suffer from some coding artifact, such as blocking artifact, ringing artifact. The higher visual quality of the decoded images/videos is needed. With3D becoming increasingly interesting for home entertainment or mobile applications, the application of image deblocking in3D images for visual quality improvement is also needed. Specular highlight removal is a hot problem in computer vision. Highlights in images actually are the combinations of diffuse and specular reflections, which make many problems in computer vision, object recognition, image content editing and so on. To properly acquire the diffuse only reflection, a method to separate the two components robustly and accurately is required.In the thesis, our works on the reduction of compression and specular artifacts in2D and3D images are as follows:(1) An image deblocking method based on sparse representation is proposed. Firstly we train a general dictionary using KSVD algorithm. Then from the JPEG compressed image, we can estimate a threshold automatically which is used in orthogonal matching pursuit (OMP). At last we can do the deblocking process like denoising to remove the blocking artifact based on the general dictionary and the estimated threshold. The performance including the visual quality and the evaluation criteria is improved a lot by experiments.(2) A specular highlight removal method using reflection component separation and joint bilateral filtering is proposed. Firstly we separate the diffuse and specular components using a theoretical method. Then we use joint bilateral filter to smooth the separated diffuse component for improving the performance. The effectiveness of proposed method is verified by experiments of the simulated images and real images.(3) A blocking artifact reduction method in3D images based on sparse representation is proposed. Firstly we train a3D dictionary using KSVD algorithm. Using the useful information obtained from the3D compressed images, we can estimate a threshold automatically used in OMP. Then we can do the deblocking process based on the error-based sparse representation method. And we make an initial system for3D images compression, transmission and postprocessing. Experimental results indicate that the visual quality and the evaluation results are improved a lot. This thesis is supported by the National Natural Science Foundation of China (No.60970097and No.610201101).