This page details our the recent work in image inpainting. It is more in the realm of curve inferencing/filling/completion algorithms which has direct applications to global image inpainting.
Image inpainting is a technique to fill the missing region, or the hole, based on the surrounding image statistics. A majority of the inpainting techniques attempt to inpaint by propagating surrounding information into the hole using information from the local neighborhood.
Recent efforts in image inpainting has focused on a two step process,
1. The first stage involves segmentation and structure completion and
2. The second stage involves texture synthesis to
complete the inpainting process.
Objective:
To complete the missing regions of a symmetric curve under severe occlusions.
Impact:
Contour completion and reconstructing symmetric objects under severe occlusions offer tremendous opportunities in many areas of computer vision applications such as digital inpainting, machine vision of robots, object recognition and identification.
Idea:
The intuitive idea behind our algorithm is to make use of existing contours in the image to complete the missing structures. This can relatively be more homogenous and more structurally compatible in completing structures than inpainting algorithms which tend to "evolve" curves based on local properties.
Methodology:
We can use similarity transforms of existing contour structure to complete the missing regions of the symmetric objects due to occlusions. Our algorithm works by exploiting the invariant nature of the curvature under similarity transform of the underlying symmetric object and also by utilizing the periodicity of curvatures of symmetric objects.
Figure 1(a) Symmetric
object with
occlusion;
Figure1(b) Curvature of the contour of the object
Figure(1)(c) Autocorrelation of the curvature; Figure1(d) Normal vectors of corresponding points nx(a) and nx(b) on the contour
To Be updated...