Enhancement of Image and Video using In-Painting Technique

Abstract

A comprehensive image in-painting method was proposed to enhance the two critical task in the prior hybrid method, which are setting up the best application order for in-painting textural and structural missing regions and extracting the sub-image containing best candidates of source patches to be used to fill in a missing region. By integrating our ‘execution-order analysis based solution ‘to task one and our image ‘context –driven source image extraction solution ‘to task two. We were able to consistently improve in-painting quality compared with that of the previous non-hybrid in-painting method while even spending much shorter processing time compared with the conventional hybrid in-painting methods. Image in-painting is process of restoring or removing object in an image. The basic task is to fill the surrounding information to inner sides. This technique boost numerous application like restoring or removing degraded part in image , text removal , stamp or symbol removal and disocclusion in image based rendering (IBR). The problem definition in image in-painting is that it is ill-posed inverse problem. It means that there is no well-defined particular technique. Image in-painting techniques are broadly categorized in two types. First, texture based in-painting and another is the structure based in-painting. The main motivations related to this technique are that in-painting results are degraded for images with combination of texture and structure features. Another motivation is that it consumes more computation time. The working principle of image in-painting is that assumption of pixels in the known and unknown parts of image that share the similar statistical and geometrical structure. In past literature, diffusion-based in-painting was introduced that are best suited for straight line, parabolic curve and for small region. The main drawback of diffusion method is that are not work on unconnected edges and also produces Gradient Reversal artifacts after restoration. With advancement of technology, sparse based in-painting and examplar based in-painting are considering for eliminating problem.
In this digest, sparse based in-painting is introduced on basis of discrete wavelet transform technique based on finding the region pixel, calculating pixel priority and normalizing the in-paint region boundary.
An image can be mathematically represented as [1] =⊂ → → (), Where x is a vector indicating spatial-domain pixel, which in the case of gray scale image (n=2) and is defined as x = (x, y). For color image (m=3) and is defined in (R, G, B) color space. The goal of image in-painting is to calculate the (R, G, B) components of the pixels situated at position x in the unspecified region U, from the pixels situated in the known region S, to finally form the in-painted image. The purpose in term of quality is that reconstructed part seems natural to human naked eye, and is physically imaginable as possible.