Study Of Semi-Supervised Learning-Based And Support Vector Machine-Based Eyebrow Recognition

With the development of world-wide Electronic Commerce, it is required to get higher accuracy, security and practicability of personal identification. Traditional identification methods including IC card and password have many problems such as with inconvenience to carry and with ease to loss, while biometric technology can overcome some of these disadvantages and thus has been paid more and more attention. At present, biometric features mainly include face, iris, fingerprint, palm-print, voice, gait, human handwriting, etc. It has been shown that eyebrow is so important a feature that it may play at least the same role as eye in face recognition.However, rare research could be found about using eyebrow as an independent biometric. This paper further investigates the probability and feasibility of eyebrow recognition using semi-supervised learning and support vector machine, with main work described as follows:1) Eyebrow images segmentation method based on semi-supervised learning. This method only need labels a number of eyebrow and non-eyebrow points by manually drawing several simple lines on eyebrow image to finish eyebrow segmentation. Segmentation experiments of this paper show that it can achieve a very good segmentation result and can be used to preprocess eyebrow image before recognition. The fault of this method is the speed of segmentation is very slow. The average time for segmentation is up 1-2 minutes when the images is 24-bit colorized images with size 768*576 and the pixels block size is 7×7.2) Eyebrow images segmentation method based on semi-supervised learning and locally sensitive hashing method. The method uses locally sensitive hashing method to improve the speed of segmentation, and the segmentation experiments of this paper show the average segmentation time reduces to below twenty seconds by this method, when the images is 24-bit colorized images with size 768*576 and the pixels block size is 7×7.3) Eyebrow feature extraction method based on Fourier transform and Gabor filter. At first, this method does Fourier Transform or Gabor filter on eyebrow images to extract the feature vectors; and then get proper feature of eyebrow by using Principal Component Analysis to reduce the dimensions of high-dimensional eyebrow feature vectors.4) Eyebrow recognition based on Support Vector Machine. Study how to use Support vector Machine for eyebrow recognition and the possible influence to recognition rate of different kernels and feature extraction methods. Experiments on databases which include 400 eyebrow images from 40 persons,1000 eyebrow images from 100 persons and 1140 eyebrow images from 114 persons show that the method can reach the accuracy of 90.5%,82.1053% and 77.7193% respectively.