Sets of multi-view images that capture plenoptic information from different viewpoints are typically related by geometric constraints. The proper analysis of these constraints is key to the definition of consistent compact representations of such images. We propose an algorithm for joint sparse approximation of multi-view images driven by epipolar geometry considerations. The new algorithm, called the Multi-View Matching Pursuit (MVMP), decomposes multi-view images into linear combination of geometric atoms by balancing the approximation error and the geometric consistency. We further add a rate penalty constraint that favors representations with small entropy towards efficient coding applications.

We also propose a dictionary learning method that learn dictionary parameters for optimal representation of multi-view images under the geometric correlation model. The proposed approach brings significant benefits in applications such as camera pose estimation in omnidirectional camera networks and compression of stereo perspective images.

For more information, please see:
[MVMP-J1] I. Tosic and P. Frossard, Dictionary learning for stereo image representation, IEEE Transactions on Image Processing, Vol. 20, No 4, pp 921-934, 2011. [pdf]
[MVMP-C1] I. Tosic and P. Frossard, Learning of Stereo Visual Dictionaries, APSIPA Annual Summit and Conference 2009. [pdf]
[MVMP-C2] I. Tosic, A. Ortega and P. Frossard, MVMP: Multi-view Matching Pursuit with geometry constraints, IEEE ICIP 2010. [pdf]
[MVMP-C3] D. Palaz, I. Tosic and P. Frossard, Sparse Stereo Image Coding with Learned Dictionaries, accepted to ICIP 2011