Abstract: Extracting features which arc invariant and can discriminate targets with similar shape is one of key problems in shape-based target recognition. Multiscale geometric analysis(MUA) offers a high degree of directionality and anisotropy,which can express local features of objects more effectively. However, it is restricted greatly when used for object recognition because most of Multiscale geometric transforms are not invariant. A new feature descriptor which is invariant to the translation, scaling and rotation, was constructed in Contourlet field in this paper, which uses the idea of the image generalized moment. This method is specialized in extracting target local characters. Experimental results demonstrate the potential of Contourlet in feature extraction, and the features will not vary with translations, scaling and rotation. Furthermore, a study of the influence of using different decompose scale of Contourlet was carried out.

Key words: Feature extraction, MUA, Contourlet transform, Invariant feature