Bilodeau and Bergevin propose a qualitative, volumetric part-based model meant to improve the categorical invariance and viewpoint invariance in content-based image retrieval. In addition, the paper presents a novel two-step part-categorization method. This algorithm is Bilodeau and Bergevin’s main contribution.

The method has two steps: transforming parts extracted from a segmented contour to a primitive map, and then categorizing the transformed parts using interpretation rules. The first step is based on a constant-curvature contour primitive (CCP) map obtained by processing a grey-level image with a generic contour extraction and a segmentation algorithm based on a standard Canny edge detector, a contour-extraction method, and an original contour segmentation and approximation method. A rule-based classifier is developed to categorize the simplified parts. The classifier rules are based on six attributes obtained from the CCPs of the part, including their geometric relationships.

The advantages of this method are proven by experiments using real images of complex multi-partobjects.