Research On Image Restoration Algorithm Of Sample Block For Stable Filling

Digital image inpainting technology,which uses the known information to fill in the damaged region according to some criteria,intends to restore integrity of images and improve visual effect.The technique has not only been applied in the fields of cultural relics protection,film special effects,error concealment,image compression and super resolution reconstruction,but also been extended to aerospace,biomedical,communications,industrial engineering,military public security and other domains,which has become a research hotpot in computer vision.In the future,with the development of the times and technology,digital image inpainting method will have a broader application prospect.Matching-based exemplar image inpainting algorithm in image restoration technology can achieve good results in both small damaged region and large degraded area.However,problems such as structure fracture,excessive texture extension and error accumulation are prone to occur during the repair process.Therefore,for the defects of the matching-based block image inpainting method,this paper starts with filling order and matching norm and explores to get it more reasonably to maintain stable filling sequence and reasonable matching norms,two algorithms are proposed.They are:1.An image inpainting method using gradient features and color consistency is proposed.Firstly,average gradient is introduced in the priority norm,which characterizes the image structure and texture change feature,to preferentially fill the structure region and extend the texture information moderately.And appropriate parameters are selected according to the experiment results.Then,updating the confidence value of the boundary in damaged area on the basis of S-type function.Afterwards,the color consistency judgment is added to the matching rule,combined with color information to find the best exemplar.Finally,experiment results demonstrate that gradient features can make the filling order more stable,color consistency can obtain more reasonable matching block,and get a satisfactory repair effects.2.In order to obtain stable filling sequence,a spherical convergence and Manhattan distance-based image inpainting algorithm is raised.Firstly,spherical convergence-based priority rule is presented to suppress the excessive texture extension.Furthermore,inspired by Stirling's theory,confidence updating norm based on it is proposed to restrain the rapid decay of confidence value.At last,in order to reduce mismatch and error accumulation phenomenon,the Manhattan distance-based matching criterion is used for the similarity measure.In the light of experiment results,the threshold value and corresponding scale value are selected to acquire the best repair quality.In terms of objective and subjective evaluation indicators,algorithm in this paper can attain better inpainting effects than other compared algorithms.