SEMINAR: Artificial Bee Colony based Patch Match

Yang Li
M.Sc. Thesis Proposal
Supervisor: Dr. Minglun Gong

Artificial Bee Colony based Patch Match

Department of Computer Science
Thursday, October 16, 2014, 1:45 p.m., Room EN 2022


The patch match concept provides a way of finding correspondences of small squares (NNF, nearest neighbor fields) between two images or within the same image. It is useful for various computer vision applications such as image inpaint/completion and texture synthesis. Finding the ground truth patches via brute force method is too expensive and slow, which limits the usage of patch match based application. However, the approximate patch match result is often good enough to produce a decent application result, so currently researchers are focusing on improving the speed of finding approximate patch match. Barnes, etc. proposed a randomized algorithm which can find the approximate correspondences quickly. We propose an ABC (artificial bee colony) based algorithm that can generate k approximate correspondences and the best correspondences of them are very close to the ground truth result. In addition, ourmethod is parallel friendly. Based on the patch match concept of 2D image, my thesis is going to apply the same idea to 3D point cloud. The main research direction will be focused on finding the appropriate features to measure the distance of 3D patches, fast approximate point cloud patch match searching and its applications.