Detecting deformable objects with flexible shape priors

We address the problem of detecting objects/shapes with large deformation and articulation in cluttered images. The approach requires a shape prior that describes the approximated outline and articulation property of a given model. While dynamic programming is often used in solving shape detection, our focus is on formulating a more effective energy function to evaluate the optimality of a matching between the shape prior and image features. For efficiency, the detection via optimization is carried out over a non-uniform elastic grid based on referencing the edge information. Experimental results are included to illustrate our method.