An intelligent image inpainting

Abstract

Image inpainting is a method of filling in the best conceivable artifacts in the newlinemissing, damaged or blurring portions of a digital image. It is a process of filling in the newlineholes as opposed to eliminating the objects that exist in the image utilizing the data from newlinethe best region neighboring the holes. Image inpainting can be used in concealing data, newlineidentifying tampering, and recovering digital records. The existing inpainting methods newlinehad rendered the immeasurable result in reconstructing the damaged area in the picture. newlineFor the filling, however, the missing parts, which affect the complex structure and newlinetexture, are still a challenge. newlineThis research work addresses the significant challenges associated with existing newlineinpainting methods. The work focuses on 1) completion of missing complex structure newlineand texture in fast efficient manner, 2) inpainting on free form mask of variable size and newline3) reconstruction of image with best visual plausible after removal of object present in newlinethe image. newlineA hybridization approach of patch diffusion and propagation of tensor structure newlineis implemented to complete the missing structure and texture. This method is extended newlinewith successful implementation of texture synthesis using spiking neural network. Both newlinemethods are tested on the standard image dataset of Technische Universität München- newlineImage Inpainting Database (TUM-IID) which contains the diversified images. The image newlinedataset has variety of images of complex building structure, different textures of objects, newlinecomplicated pattern of grass, field and textual information. In this researched, four newlinedifferent types of masks are used to evaluate the proposed model which masked the 5% newlineand 10% of the image area. Inpainting model is also tested over the random images with newlinerandom masking and removal of objects. newlineFurther the approach is enhanced with deep learning autoencoder for irregular newlinemask. The autoencoder model converts the high dimension feature to low dimension that newlinerecreate an original one. Proposed auto encoder uses the partial convoluti