In this study, a new method for image retrieval based on shape features is proposed that finds feature points by curvature function for each shape. The initial features are generated at these feature points. Then a supervised feature extraction method based on PCA and multilayer perceptron creates a smaller set of features from nonlinear combination of the original features. The proposed algorithm is invariant to several kinds of transformations including some articulations and modest occlusions. The retrieval performance of the approach is illustrated using the MPEG-7 shape database, which is one of the most complete shape databases currently available. Present experiments indicate that the proposed representation is well suited for object indexing and retrieval in large databases. Furthermore, the representation can be used as a staring point to obtain more compact descriptors.
]]>