Image Completion Technique Based On Energy Minimization

Generally, Image completion technique is the process of restoring the integrity ofimage, which filling the defect area in visually reasonable way to achieve a variety ofspecific image editing purpose, and it is difficult to perceive image have been edited toobservers subsequently. Compared to the manually repairing way, digital imagecompletion techniques have higher efficiency in the implementation and wider range ofapplications, and therefore it has become an emerging research focus in the field ofmachine vision for the past few years. Now this techniques have good utility value andvast development prospects in calligraphy and painting relics of digital inpainting,unnecessary object removing, old film processing, video privacy protection, image lossycompression, film special effects and error concealment in multimedia communications,etc.Now global optimization technique based on energy minimization is becoming thenew research tendency in this field, for it is usually able to obtain fairly good repairingresults. Image completion framework based on energy minimization consists of two parts:extraction of repairing object based on the GrabCut technique and filling defect regionbased on the Priority-BP technique. The former is image segmentation techniques by theGraph cuts global optimization algorithm, and it accurately sketches the contours of therepairing object in less manual interaction cost. In order to improve the quality of objectextraction further, the color computation is changed the RGB space to the CIE space. Thelatter is image completion technique based on the belief propagation algorithm of globaloptimization, and specifically adopting the modified version of BP algorithm based onpriority message scheduling and dynamic label pruning, which is proposed by Komodakis.Moreover, using the interactive constraints information given by extraction of therepairing object can reduce the size of the label space of the Priority-BP algorithm further,thus to enhance the executing efficiency of the algorithm.The executing results demonstrate that segmenting target object in less interactivecosts can reduce the burden of manually identified, and increase the practicality andflexibility of the image completion system. In addition, color computations in CIELab space model improve effects of the image object extraction. What’s more, constraininglabel sample space based on the Priority-BP image completion algorithm making use ofinteractive information of object extraction, can improve the executing speed of algorithm.