Super-resolution Technique For Visual Quality Enhancement Of Low-resolution Images And Videos

Many super resolution(SR)techniques have been proposed to obtain the high resolution images and videos by fusing the low-resolution images and videos respectively.In this thesis,we proposed an effective technique to increase the visual quality of low resolution image at low cost of complexity.We used linear regression and iterative back projection technique.In linear regression,future sign search algorithm is used to extract the image features and then couple dictionaries and iterative back projection techniques are used to obtain an initial SR reconstruction image which has some visual artifact like blurring and edges.To remove the visual artifacts,image refinement process is applied which contains Gaussian filter.After applying refinement process,final SR image is obtained.This method offers a cost-effective way to overcome some of the inherent resolution limitations of the low cost imaging sensors for better utilization of the growing capability of high-resolution displays.If we compared our method with others SR methods such as Yang [1] and [6],it gives reliable and efficient results.Due to the efficient performance of proposed method,we extend our work in video transcoder that converts low-resolution AVS videos to ultra-definition HEVC videos.For this end,the proposed SR method is used to increase the visual efficiency of transcoder at low cost.Video information such as motion vectors(MV),transform coefficients(TC),and determined macro block(MB)modes are very useful to classify the video frame into two types of blocks: blocks of the most interest(MBOI)and blocks of less interest(LBOI).After successful classification,we applied the proposed SR method with different parameters.For MBOI,we intend to achieve the better quality,while for LBOI the visual quality will be lower.Optimized parameters values are used for both MBOI and LBOI with different number of iteration due to the required quality and complexity cost.Experiment results show that the proposed method generates efficient visual quality as compared to full frame super resolution method.