Research On Image Inpainting Algorithm

At the age of information, image acquisition, processing, transmission andreception are all based on a digital device. In communication system, the imageinformation is extremely sensitive to noise due to image compression andtransmission characteristics. Particularly in the mobile communication channel,with various kinds of noise, it is easy for image signal to miss some information.Consequently, the purpose of researching this subject is self-evident in the trendof mobile Internet. In addition to the video error concealment technology, theinpainting technology is also widely used in retouching of old digital paintings orphotos, removing unwanted persons or things in the image, film production,image compression and reconstruction of amplification.In the field of image processing, image inpainting is a new branch which isactive in scholars’ research areas after2000. Belongs to the branch of the subjectof image restoration, for processing mathematical method, it was used the idea ofpartial differential equations when first proposed. Then, the evolution based onthe idea and the new algorithm was proposed one after another. Totally, theinpainting algorithm will be grouped into two categories: one based on thestructure; the other based on texture.The three models in the first class of algorithms was had simulation, and thecharacteristics of the algorithm were compared and summarized in this paper.Several shortcomings were found about Criminisi algorithm, and theimprovement of idea was put forward: adaptive selection of the template sizewhich optimize processing time and effect, the calculation of priority which islargely slowed the emergence of the fault, the addition of punishment item whichensures the best matching block loyal to the original image.A better fast algorithm than both FMM and Bornemann algorithm waspresented. The product of the direction factor about the weight function wasreplaced by the isolux, so the value of direction factor is larger when the point iscloser the isolux. Confidence factor of source pixel was adjusted to avoiding errors accumulation.The feasibility of the two improved algorithms is tested according to a lot ofbasic tests and analysis of results.