In this paper

a contour based image retrieval algorithm is proposed for improving the algorithm proposed by Choi Wai pak et al

in which the shape of an object is represented based on the normalized maximal disks. In order to generate a simpler and smoother contour

the one dimensional Gaussian functions of two different scales are employed respectively for the concave and convex part of the contour in the proposed algorithm. Additionally

the skeleton of the contour is extracted by a skeletonization algorithm. Finally

the histogram of the distances between the evolved contour and skeleton is used to describe the shape for the retrieval purpose. As compared with the original algorithm that uses only the skeleton of an object

the algorithm proposed uses not only the contour that represents the shape of an object from outer but also the skeleton that preserves the original objects topology from inner. Experimental results show that the new algorithm proposed here outperforms that proposed by Choi Wai pak et al. in the robustness to the scaling

rotation and noise corruptions.