A Shape Recognition Method Based on Syntax-Automaton

A good basis for invariant recognition is the shape or the contour of an object, which is usually stable and persistent. There are some challenges to be solved. The first is that the object has different shapes from different views. The second is to organize the information into a structured data, so that they can be manipulated easily. We represent the image as a set of lines by a detector, and all lines are in the form of vectors instead of pixels. In order to get a better performance on the real-world image with multi-view object, we generate some template data by rotating 3D model. We define the sequential structure and the template is described by the new descriptor. The feature is based on the relationship between two adjacent contour lines, which remains unchanged as scale and rotation varies. Then we organize the phrase of each template into an automaton and execute the syntax-simulation by the state-transition in the automaton to complete object recognition. Finally, we do experiments on F117-282 and B2-282 data set. The results of experiments confirm that the present model of object recognition has a good performance on the image with complex background. The method in this paper is not only the object detection, but also the precise verification for each component of the object.