基于特征融合与稀疏表示的人耳识别

Ear Recognition Based on Feature Fusion and Sparse Representation

人耳识别是一种新兴的生物识别技术,具有较高的理论研究价值和市场应用前景,并随着图像处理、模式识别等领域的发展而逐步发展。在人耳识别中特征提取是生物特征识别技术的关键环节,对最终分类结果的准确性起着决定性作用。因此为了提高人耳识别技术中分类结果的正确率,提出了一种基于特征融合和稀疏表示的人耳识别方法。该方法采用四个方向上的Sobel算子检测边缘,并在每个边缘图上提取边缘特征;同时利用灰度共生矩阵提取四个方向上人耳图像的纹理特征,结合边缘特征和纹理特征,最后通过稀疏表示模型对人耳进行分类识别。实验结果表明,采用边缘特征和纹理特征相融合的方法能较大提升人耳识别的准确率,从而验证了该方法在人耳识别技术中的有效性能。

Ear recognition is an emerging biometric recognition technology, with high theoretical research value and market prospect,and develops gradually with the development of image processing, pattern recognition and other fields. Feature extraction is the key to this technology which plays a decisive role in the accuracy of the final classification result. Therefore,in order to improve the accuracy of clas- sification result in the technology of ear recognition, a method of ear recognition based on feature fusion and sparse representation is pres- ented. In this method, the Sobel operator from four direction is adopted to detect the edges and extract their feature. At the same time the GLCM （ Gray Level Co-occurrence Matrix） is used to extract texture feature of ear images. Finally sparse representation model is utilized to conduct classification recognition of ear in combination of edge and texture features. The experiment shows that the proposed method can improve ear recognition accuracy greatly,thus confirming its effectiveness in the survey of ear recognition.