Abstract:

A twoclass classification model is established to evaluate the performance and application prospect of the novel linear SVM (Support Vector Machine) correlation filter. The SVM filters based on different training samples are synthesized by computer as well as other 3 correlation filters. The ability to resist inplane rotation, outofplane rotation and noise of those filters are tested and compared. The experimental results indicates the SVM correlation filter exhibites the best ability to resist inplane rotation, and favorable ability to resist outofplane rotation and noise. It is concluded that the linear SVM correlation can be applied helpfully in the inplane rotated image recognition and 3D object recognition within a medium distortion range.

Key words: orrelation filter, pattern recognition, SVM (Support Vector Machine)