Structural Feature Extraction from Regular-Shaped Rigid Objects(2) - Enhanced by Prior Knowledge

Abstract

The structural feature extraction from regularshaped rigid objects is continuously studied. Some rules based on prior knowledge are proposed to improve the effects of the same subject(1). First, some rules are introduced for recovery of the missing true features. Then, some rules for distinguishment between true features and pseudo features are proposed, which is suggested be executed in a recursive way. Last, a graph structure that represents the connection of the corners is established, which group the corner features according to connectivity components. Experiments validate that the proposed approaches have higher detection rate and much lower error rate than the approach in the same subject(1). The proposed approaches make complete the true features on the silhouette of the objects and remove many pseudo features. All these pave the way for future structural recognition.