Morphological and statistical approach for boundary based shape representation methods

Progress in shape based object recognition methods involving either the boundary based or region based methods and their relative popularity is presented. Prevalence of boundary in almost of types and their shapes discriminating features are discussed. Limitations to boundary based methods viz. sensitivity to noise and variations are detailed. The present paper proposes to overcome these limitations by a novel method to represent boundary viz. using an integrating appearance basis both morphological and statistical nature. A part from the morphological approach that extracts the boundaries for binary images only, inclusion of the statistical approach will take care of the gray images also. The proposed boundary based method includes the reconstruction algorithm (for the extracted boundary) using the morphology. The present reconstruction method is found to reconstruct exactly same as the original shape from the extracted boundary. The experimental results evince relative efficacy (through error rate) for the boundary extraction system.