A New Method for Image Retrieval Based on Shape Decomposition

In our proposed scheme, two shape features are extracted by our shape decomposition method. The first feature is the proportion of the number of contour points in each of the four quadrants, and the second feature is the number of contour points intersecting with two quadrant-segmentation lines at the first level. When a query shape is submitted to the system, the second feature is used to generate the lower and upper bounds for the search space. Then, the database shapes within these two bounds are compared with the query shape using the first feature. Experimental results show that our scheme can handle shapes with holes very well, and has high performance in three shape databases.