With increasing emphasis on automated manufacturing there has been a growing interest in interpreting images containing two dimensional (2-D) objects. Most of the current 2-D objects recognition system are model based. In such systems, each representation of a known set of objects is precompiled and stored in a database of models. Then the database can be used to recognize unknown objects in the image.
This paper presents an efficient 2-D objects recognition system which is invariant with rotation, translation and scale. Most important parts of this system are model building and index design. In this study, Objects model is built by a corner detection algorithm. It has been widely used as the first step in many applications such as shape description and objects classification. This paper evaluates a number of conventional corner detection algorithm and addresses some problems such as noise sensitivity. To enhance the corner detectors, this paper proposes a new corner detector. After model was built, a decision tree is automatically constructed by using efficiency measure which estimates the difference among features.
The proposed recognition system has been demonstrated for several test patterns. The results show that the proposed system accomplishes fast and efficient recognition.