Shape is an important feature of objects in image and shape-based image retrieval has obtained more and more attentions in recent research on content-based management and utilization of image database system. Although several systems have been developed
two main shortcomings are still existed. The first is that the performance is not stable. The second is that the variance with respect to translation
scaling
and rotation. To cure the above problems
this paper presents a novel shape-based image retrieval algorithm. The algorithm first transforms the luminance image with wavelet modulus maximum to get multi-scale edge images
then employs a set of seven invariant moments to extract the features of image. Consequently
each image is characterized by a multi-scale moment vector in feature space. Similarity is given by the Euclidean distance between two images' normalized moment vectors. Experimental results on clothes image database show that this algorithm can well capture the shape and spatial information of image and it is invariant with respect to translation
scaling and rotation of objects. In addition
the algorithm is also tested with more complicated flower images in a database; the experimental results further verify the effectiveness of the algorithm.