Ear Recognition Based On Curvelet And Improved Isomap Algorithm

As a new biometric identification method, ear recognition is getting more and moreattention due to its unique physical characteristics and observation angle. It can be usedalone as a biometric identification technology, and also can be combined with otherbiometric identification methods such as face, gait, iris and so on, forming a multimodalrecognition system. In all, ear recognition has far-reaching research value andapplication space.In this article, a new ear recognition system is established using Curvelet transformand improved Isomap algorithms. It consists of five modules. They are imageprocessing module, feature extraction module, space transforming module, featuredimensionality reduction module and feature classification module. In the imageprocessing module, Curvelet transformation is introduced to denoising the ear images,which can preserve the curve information of the ear perfectly. In the feature extractionmodule, the Curvelet coefficients are used to reconstruct the original ear images. Whilein the space transforming module, the similarity is described by Image Euclideandistance, which can be embedded into Isomap algorithm easily and overcome thesensitivity of Isomap algorithm to noise and distortion. In dimensionality reductionmodule, an improved Isomap algorithm based on manifold reconstruction is proposed toenhance the generalization of Isomap for a new sample. Nearest neighbor classifier isemployed to classify the features with low dimensionality in the feature classificationmodule.Experiments on USTB3ear database show that:（1）Curvelet transform denoisingmethod can improve the SNR of images to a great degree;（2） The imagesreconstructed by Curvelet coefficients have better quality and great robustness to noise,the more noise the images contains, the more obvious the improvement is;（3）TheImages Euclidean distance has a smoothing effect on the images, which can reduce theimpact of noise and distortions;（4）The improved Isomap algorithm based on manifoldreconstruction preserves the local linear relation of new samples to construct the lowdimensionality feature, which is fast and accurate;（5）Ear recognition system combingCurvelet and improved Isomap algorithms can improve the robustness to noise ofimproved Isomap algorithm by introducing Curvelet transform. So the system can makefull use of the generalization property of improved Isomap algorithm, and becomes a good ear recognition system with high right recognition rate and good robustness tonoise.In all, the ear recognition system can not only be robust to the noise and distortionsin the images, but also obtain the low dimensionality feature of new samples fast andprecisely, resulting to higher efficiency and more robustness. So this article is expectedto provide some new and valuable ideas for ear recognition technology in the futureresearch.