Image Inpainting Algorithm Based On Statistical Feature Subregion Division

Image inpainting is a process of completely filling the missing information of the image,and which is one of the hotspots in the field of image processing.The classical algorithm of image inpainting based on sample block,although on the image reparation can obtain good,the problem of structural line fault and matching error will appear in the process of repairing the large damaged image.Therefore,this paper mainly studies the problem of repairing the missing information in the damaged region with rich texture information and complex structure information,and designs an image inpainting algorithm based on statistical feature subregion division.The specific work is as follows:1)This paper proposes an image inpainting algorithm based on statistical characteristics of feature subregion division.First,according to the different features included in the image,a feature formula is constructed to extract features from the image,and then the feature subregions are divided by statistical feature values.Then,in the calculation of priority order,a priority calculation formula with structural factors is constructed,which increases the influence of structural information and avoids the occurrence of structural fracture.Secondly,searching in the corresponding feature subregion,searching for the initial candidate sample block set according to the constraints of the block to be repaired and the sample block itself,and then according to the constraint condition between the optimal neighborhood similar block and the sample block determined in the neighborhood window of the block to be repaired,secondary screening is performed to determine the best sample block set.Finally,assigning weights to synthesize the best sample blocks improves the matching accuracy.2)In this paper,the priority calculation formula,matching method and filling method of image inpainting algorithm based on the sample block are improved,and an image inpainting algorithm based on structure tensor feature partitioning is proposed.First,this paper design a priority calculation method based on feature partitioning,the structural tensor is used to classify the block to be repaired,and the structural feature factors are constructed according to the structural tensor eigenvalue,thus the priority formula is improved,so that the more structural information is repaired,the better the block can be repaired.Secondly,in the search block set,using the K-means algorithm for K sample collection,and then comparing similarities between the block to be repaired and the K candidate cluster center to select the best sample block.At the same time,under the constraint of "structure + color",we get the best sample block to fill the repaired block,which can not only avoid the occurrence of mis-match,but also improve the matching accuracy.Through experimental verification,the algorithm of this paper can repair images of damaged region with rich texture information and complex structure information,compared with the Criminisi algorithm,the 2~3dB is improved on the peak signal-to-noise ratio,and the value of structural similarity is also closer to 1,and the repaired image is more satisfying for people's visual connectivity.