Hierarchical vibrations for part-based recognition of complex objects

We propose a technique for localization of complex shapes in images using a novel part–based deformable shape representation based on finite element vibration modes. Here, our method gives an extension for Finite Element Models to represent elastic co–variations of discrete variable shapes. It avoids misregistration by resolving several drawbacks inherent to standard shape–based approaches, which cannot detect structural variations and occlusions. Our algorithm uses a hierarchical shape model, involving an evolutionary deformable shape search strategy. The different levels of the shape hierarchy can influence each other, which can be exploited in top–down part–based recognition. It overcomes drawbacks of existing structural approaches, which cannot uniformly encode shape variation and co–variation, or rely on exhaustive prior training. We applied our method to two different example applications, which include shape detection and discrimination, as well as localization of the desired object under occlusions. Experimental results are promising and show the good performance of our approach. It is robust to changes in the values of parameters used and requires no prior training with regard to shape variation and image characteristics. By utilizing a quality–of– fit function the model explicitly recognizes missing discrete parts of a complex shape, thus allowing for categorization between shape classes.