The conference starts in:

Early submissions deadline:

Join us Vkontakte and follow news in Russian

You are here

Recognition of panorama parts using OpenCV

Background and Motivation: 

The more data you store, the harder it becomes to manage it. This project is dedicated to classification of photos. We take 6000 image series sourced from panoramic images. Each series contains matching fragments of a panoramic image with a partial overlap (possibly in the wrong order). Some series may contain one or two images from other series. The goal is to figure out the odd ones.

Project Summary: 

The main difficulty of automated classification is to work out an algorithm which would give highly differential results on image comparison. There are some standard methods of analyzing images: template matching, image stitching and optical flow, but the last one is hardy suitable for this problem. The idea is to use OpenCV library where these methods are already implemented and to estimate the quality of such preliminary classification. Then starts the main part of work: to adjust preprocessing algorithms and parameters according to the current problem. For the first step we tried to use the template matching approach. For each pair of images in the set we searched the segments of target image on the source image. To achieve the best result several target-into-segments splitting strategies were developed. For the second step we tried image stitching approach. Specifically, we used the SURF algorithm to find the key points for each image and then found matching key points for each pair if images in the set. The last step was combination of two implemented approaches in their best configurations.

Project goals and future research directions: 
<p>The goal of this project is to develop an automated system for finding the odd images in series of panorama parts. This system must be able to process large amounts of data. For the first step we tried to use the template matching approach. For each pair of images in the set we searched the segments of target image on the source image. To achieve the best result several target-into-segments splitting strategies were developed. For the second step we tried image stitching approach. Specifically, we used the SURF algorithm to find the key points for each image and then found matching key points for each pair if images in the set. The last step was combination of two implemented approaches in their best configurations.</p>
List of team members and their organizations: 

Ekaterina Polishchuk, SPbAU University supervisor: Kirill Krinkin, SPbAU Industrial consultant: Ekaterina Tuzova, JetBrains

Status: 
Graduate
Project Timeline and Expected Deliverables: 

April-May 2011: template matching algorithm implementation May 2011: image stitching algorithm implementation June 2011: presentation on local University seminar November 2011 : paper and report on 10th FRUCT conference

Final deadline: 
Saturday, April 30, 2011 (All day)
Users: 
Group: