Higher-order spectra (HOS) invariants for shape recognition

Abstract This paper describes a shape feature-based invariant object recognition method. First, a set of features invariant to rotation, translation, and scaling (RTS) is generated using the Radon transform and bispectral analysis. In order to improve the noise resistance of the invariants, the ensemble averaging technique is introduced into the estimation of bispectra. The feature data are further reduced to a smaller set using thresholding and principal component analysis. The resultant feature invariants are proved to be more reliable and discriminable in the classification stage than the original ones. It is shown experimentally that the extracted higher-order spectra (HOS) invariants form compact and isolated clusters in the feature space, and that a simple minimum distance classifier yields high classification accuracy with low SNR inputs. The comparison study with Hu's moment invariants and Fourier descriptors also shows that the performance of the proposed method is better than these two methods especially in the presence of background noise. The HOS invariants algorithm is also applied to shape-similarity-based image indexing. A new similarity matching technique based on Tanimoto measure is employed for fast image retrieval. The retrieval accuracy is high as shown in the experimental results.