Abstract:

Effective contour fragments are valuable for accurately recognizing the occluded object. To solve the improper contour fragments obtained by the existing recognition algorithms, a partially occluded object recognition algorithm based on feature description integrity is proposed. Firstly, the preliminary contour fragments are obtained through the local curvature distribution. Then, a multi level fragment merging operation is carried out on those preliminary contour fragments. To ensure the feature description integrity, an evaluation of the importance degree of each contour fragment is performed. And a set of contour feature fragment (CFC), representing completely object features at various levels, is obtained. Then an evaluating function of reliability, which reflects the relationship between CFC and its corresponding object, is introduced to decrease the mismatch error between CFCs. Finally, the similarity of different CFCs, in combination with their reliability, is jointly used to get the best recognized result. Simulation verifies that this algorithm describes completely the feature and increases effectively the recognition accuracy.