Classifiers for Motion

In this paper we present a unsupervised learning based approach for sub-pixel motion estimation. The novelty of this work is the learning based method itself which tries to learn the shifts from a large training database. Integer pixel shift is sub-divided and discretized to levels in both the horizontal and vertical direction. We pose the problem of motion estimation in a polar coordinate system. Shift estimation in the x and y direction has been posed as a problem of estimating r and . The ordinal property of r has been used, and consequently, we employ a ranking based approach for estimating r. For estimation we employ multi-class classification techniques. We demonstrate how very simplistic features can be used to differentiate between different subpixel shifts. Further, we compare our method with logistic regression analysis and boosting based methods. Also we draw comparison with standard shift estimation techniques.

• Type: Conference
• Year: 2006
• Where: ICPR
• Authors: Mithun Das Gupta, Nemanja Petrovic, ShyamSundar Rajaram, Thomas S. Huang
• Keywords: Computer Vision, ICPR 2006, Integer Pixel Shift, Shift Estimation Techniques, Sub-Pixel Motion Estimation