Researching For Ear Recognition Under Non-controlled Conditions

The basic principle of biometric identification technology is using aunique feature or characteristic possessed by the individual for identity authentication.Biometric identification technology is better the security and ease of use compared withthe traditional identity authentication technology; it is gradually replacing the traditionalidentity authentication technology. Ear recognition technology is a new branch of thebiometric identification technology becoming more and more extensive attention ofscholars at home and abroad. This paper conducted a detailed research on the earrecognition under non-controlled conditions. The following work:1. Image preprocessing for ear image database. First, the image library gradationconversion and the use of the median filter for processing, and then the image of thecutting and using bilinear interpolation for size normalized, and finally do the gray levelhistogram equalization processing.2. Research of ear recognition based on SIFT, and carries out the experiment. EarRecognition low recognition rate of the reasons is that the SIFT-based angle changes SIFTfeatures value changed.3. In order to overcome the SIFT non-fully affine invariant, proposed ear recognitionbased on the Gabor wavelet ASIFT feature points. Verified by experiment that recognitionaccuracy is greatly improved, the time complexity of the growth within the acceptablerange.4. Proposed ear recognition algorithm based on the Harris-SIFT. Harris corner detectionalgorithm find feature points using SIFT feature points described, complement to thecollection SIFT feature points, increasing number points of the match is successful.Experiment average recognition rate95.2%.5. By analyzing the experimental data and with reference to the experimental results otheralgorithms, manual analysis of the human ear image library identifiable. Eventuallyproposed when the depth of the rotation angle the human ear more30degrees, the use oftwo-dimensional gray-scale image for the human ear recognition unable obtain reliableresults.