Sparse convex combination of shape priors for joint object segmentation and recognition

In this paper, we introduce a novel model for simultaneously segment and recognize object using shape prior information. Given a set of training shapes including many different object classes, the target shape in a test image is represented approximately as a sparse convex combination of the training shapes. The proposed model is optimal in the L₂ criterion between the unknown true shape and the convex combination of the training shapes. Without explicitly imposing sparsity constraints, the convex combination coefficients obtained from minimizing the ISE are natural sparse. The proposed model is able to automatically select the reference shapes that best represent the object, and accurately segment the image taking into account both the image data and shape prior information. It is different from the existing shape prior based segmentation models, which are constructed by using linear combination of a data-driven term and a shape constraint term. In addition, an intrinsic registration of the evolving shape is introduced into the model for transformation invariance. Numerical experiments on synthetic and real images show promising results and the potential of the method for object segmentation and recognition.