Interior Distance Ratio to a Regular Shape for Fast Shape Recognition

Abstract

A fast shape recognition method based on regular graphic is proposed in this paper. It is the Interior Distance Ratio to a regular Shape (minimum bounding rectangle (MBR) or minimum circumscribed circle (MCC)) (SIDR). Regular shapes themselves have either axisymmetric or origin symmetry, which gives them regularity. Shape, as a feature of an object, plays a significant role in computer vision and image analysis. The shape descriptor is widely used to compute remarkable features of the visual image, especially in image understanding and analysis. SIDR is a new remarkable feature of the shape, which is the distribution of the interior distance between the shape contour points and its minimum bounding rectangle or minimum circumscribed circle. It can provide more effective performance support for practical application fields of computer vision, such as object detection and recognition. The minimum bounding rectangle or minimum circumscribed circle can change according to the change in a shape’s position, scale and direction, which is extremely suitable for describing a shape that has deformation. In addition, the rectangularity and circularity derived from them also have the potential peculiarity to describe the shape feature. Therefore, this paper uses the interior distance ratio of the shape to represent the shape feature. First, the minimum bounding rectangle or minimum circumscribed circle of the shape is selected according to the rectangularity and circularity of the shape. Then, the interior distance proportional distribution from the shape contour point to the minimum bounding rectangle or minimum circumscribed circle is obtained. Finally, a histogram is used to represent the distribution feature, and shape matching and recognition are carried out. A self-built dataset and three international generic datasets are used to verify the validity of the method. The performance exhibits the sophisticated property (accuracy and matching speed) of the proposed method. It is worth mentioning that this simple method has a recognition rate of close to 100% on the self-built dataset and has achieved excellent results for other datasets compared with some international state-of-the-art methods.