Motion Segmentation of Color Images

Joan Borràs and Tomás Svoboda

Original image

Collinearity segmentation

GMM segmentation

Maximum segmentation

Segmentation of color images is a common research area, where good algorithms have been developed. In this project the algorithm will process each input frame to decide which pixels belong to the background or to an object detection. After an input image is read, it uses the adaptive background modelling by using a mixture of Gaussians, GMM [6]. It creates a model with several Gaussians. After that it decides which Gaussians belong to the background. That one will be the most similar to the intensity values of the pixel. It compares that values and its neighbourhood to decide if the point is labelled as foreground or background. The comparison can be made by collinearity [1] or maximum distance criterion. If that distance is smaller than a threshold, then the pixel will be labelled as a background, otherwise it will be labelled as foreground.The threshold has to be specified. It uses the GMM algorithm to know which pixels match background. With those pixels it will process an histogram to specify the threshold.

Code and report:

• Code.rar
• Report.pdf

Source images:

• Input Office.rar
• Input RedChair.rar

Video results:

Office.wmv [5.53 MB]
First section is input image, the others are the methods
segmentation: GMM, collinearity and maximum

RedChair.wmv [1.99 MB]
First section is input image, the others are the methods
segmentation: GMM, collinearity and maximum

References:

1. Rudolf Mester, Til Aach and Lutz Dümbgen. Illumination-invariant change detection using a statistical collinearity criterion. In B. Rading and S. Florczyyk, editors, Pattern Recognittion: Proceedings 23rd DAGM Symposium, Lecture Notes in Computer Science 2191, pages 170-177. Springer Verlag.
   
2. Image Processing, Analysis and Machine Vision. Milan Sonka, Vaclav Hlavac and Roger Boyle. Thomson 3rd edition, 2007. Chapter 16: Motion analysis.
   
3. Segmentation of color images. Adam Chwedyk. Czech Technical University, Prague. Version 1.0, 25th January 2007.
   
4. Real-time Segmentation of color images. Implementation and practical issues in the Blue-C project. Nicolas Galoppo von Borries, Tomas Svoboda, and Stefaan de Roeck. Swiss Federal Institute of Technology, Zürich, Computer Vision Lab. BiWi-TR-261, Version 0.9, 21st August 2003.
   
5. Image Processing, Analysis & Machine Vision – A MATLAB Companion, by Tomas Svoboda, Jan Kybic and Vaclav Hlavac. August 2007.
   
6. Adaptive background mixture models for real-time tracking. Stauffer, Chris and Grimson, W.E.L. EEE Computer Vision and Pattern Recognition, June 1999.