Ear Recognition Based On Frequency Domain Characteristics

In recent years, biometric identification technology has been paid more andmore attention. Speech recognition, fingerprint recognition and iris recognition havebeen widely used in large enterprises and government confidential departments.Meanwhile, ear recognition is in a stage of rapid development. The human ear imageis not only single, unique, stable and ease to collect, but also not susceptible tomakeup, facial expression and other factors. This paper conducts a detailed researchon the ear recognition based on frequency domain characteristics. The specific stepsare as follows:(1) Ear image pre-processing. Color ear images are transformed into gray images bythe weighting method, and then the frequency domain filter is usedfor image filtering, and finally histogram correction is conducted on ear images.(2) Propose the ear recognition method based on weighted wavelet transform andDCT. Firstly, two dimensional discrete wavelet transform is conducted on thehuman ear images, and then block discrete cosine transform is implemented onthe low frequency components of wavelet transform and weightedhigh-frequency components in order to extract DCT coefficients of the imageand construct the feature vectors. Finally, this paper uses the nearest neighborclassifier combined with weighted distance for classification and recognition.(3) Propose the ear recognition method based on EPSO-BP classifier. The improvedPSO algorithm combines multiple local minimum points, effectively avoidingthe local optimal problem in PSO algorithm. BP learning algorithm is a kind oflearning algorithm, whose learning ability and generalization ability are strong,but the algorithm convergence speed is slow, and easy to fall into local minimumarea. The paper combines the EPSO algorithm and BP algorithm, finding out theapproximate range of weights of BP network using the EPSO algorithm.(4) Propose the ear recognition method based on Gabor transform and block LBP.This paper conducts Gabor transform on the ear images, and then multiplies theGabor wavelet amplitude and the ear image, getting the Gabor wavelet amplitudeimage. Finally, the block LBP algorithm is applied to the Gabor waveletamplitude image.