Image Completion Algorithm Research And Implementation Of Intelligent Filling

Image completion technique is used for repairing damaged images with missing regions. Using other parts of complete images as references, according to certain rules to fill the missing regions and make repaired images visually plausible. With the popularity of digital media technology, digital image completion technology has been widely used, not only to be applied to repair damaged images, also for image compression, object removal, the super-resolution image analysis and video error concealment, etc.. However, since there is not enough information to ensure the correctness of the results, image completion stays a pathological problem. General judgment depends on people's subjective visual perception. Therefore, using algorithms to meet people's visual perception is a continuous learning, cognitive understanding and reproduction process. Among many image completion methods, Patch-based synthesis method based on sample attracts much attention due to the good results for structural image completion, and patch matching algorithm is the foundation of such methods.However, as patch matching image completion algorithm is good for structural image completion, it always ignores the texture information processing while matching errors often occur in the nearest neighbor search process. To solve these problems, we propose a layered image completion method with texture synthesis integrated. This method based on patch matching algorithm, using image pyramid mode and elimination mechanism to enhance the completion effects of image structure. Meanwhile, the sample-based texture synthesis method is assigned to further optimize the texture. Experiments show that, with the same input images, the proposed method can obtain better results than traditional ones.On the other hand, traditional nearest neighbor search matching process is costly and time-consuming. It cannot provide good user experience in practical application. To solve the problem, we propose a variable scope patch sampling method for fast image completion. This method can quickly find the right patches from the image, thereby reducing the cost of the nearest neighbor search. Experiments show that, to ensure the quality of image completion results, the proposed method is faster than traditional methods.Finally, combining two improved methods, we design and implement an interactive image intelligent filling application. This application is different from many other image processing applications. It considers the perspective of visual psychology, partitions different image completion areas according to visual focal point. Visual focus areas use high quality filling method while non-visual focus areas use quick filling method to complete the whole process. In this way, the application is able to meet the visual perception while improving the speed and reducing cost which has a good application prospect.