Shape Matching Using Wavelet Transform Modulus Maxima

Object contour matching has been prevalently used in pattern recognition and content-based indexing and retrieval of digital images. In this paper, the contour descriptors are extracted from a WTMM (Wavelet Transform Modulus Maxima) image, which is produced by a one level biorthogonal wavelet transform and contains curvature points of the contour. The most important WTMMs are further utilized to find the correspondences between points on the two shapes. In order to solve the correspondence problem, the centroid of the most important WTMMs and the line across the centroid and the point furthest away from the centroid are used respectively as the reference point and the reference line. Angle descriptors are constructed between the reference line and the lines connecting each important WTMM point and the reference point. The modulus maxima magnitudes, the angle descriptors, and the normalized distances from each important WTMM point to the centroid are used to construct the feature vector. Both the correspondence matching percentage and the similarity score are derived from the feature vector. They are invariant to the translation, rotation, and scale changes. The experimental results have demonstrated that the proposed approach can accurately and efficiently match the object contour under these three changes.