Image Completion Using Automatic Structure Propagation With Bézier Curves

This paper presents an algorithm developed to fill the missing region of salient structures specifically for the curve. The stages are as follows: segmenting the image to identify the salient structures along with knowing the first and last points of each curve, finding a way to match points that should connect for structure reconstruction in missing salient structures, finding control points from the curve on the known region to estimate the control points in the missing curve to create a Bézier curve, and filling the missing regions of salient structures following the created Bezier curve. To choose candidate patches, we apply form image inpainting via modified exemplar-based inpainting with two-stage structure tensor, and image sparse representation method changes priority step as a product of data and confidence terms. Finally, find the candidate patch for better fill structures by rotating that original image to $r$-degree until a full 360°.