Visual Object Detection Using Deformable Sparse Coding Model

Object detection is a challenging task in the field of pattern recognition. The objective of object detection is to locate the target objects in the testing images. In this paper, we use SVM trained active basis model as a sparse coding model for representing objects. The sparse coding model represents each image as the linear superposition of a small number of Gabor wavelets selected from an over-complete Gabor dictionary. We divide the learned template into several parts, and allow each part and each Gabor wavelet to locally shift to account for shape deformation in detection process. The model is trained from the roughly aligned images. The detection is achieved by sliding windows from a multi-scale image pyramid. The experiment shows a good performance of the detection method in some testing images.