Abstract:

We propose a novel ear recognition algorithm based on Gabor features and improved LDA to deal with the inevitable problem of small sample size.We firstly extract ear features by the local Gabor filter,and redefine the new Fisher criteria and the intra class scatter matrix.Then we seek the optimal projection direction by mapping from a higherdimensional space to a lowerdimensional space,Finally we make a comparison of the Euclidean distance of projecting feature vectors  between the training samples and the testing samples,and classify them accordingly. Experimental results show that, compared with the traditional methods, the proposed algorithm can effectively solve the small sample size problem in ear recognition with a higher recognition accuracy.

Key words: local gabor feature；improved LDA；euclidean distance；small sample size problem；ear recognition