A Study Of Automatic Eyebrow Recognition Method Based On Comparison Of Feature Strings

Biometrics offers greater security and convenience than traditional methods of personal recognition, it can not be lost,impersonated or cracked, so it has received growing interest and became a research hotspot and strategic technology at all countries. Biometrics have been widely used in government,military,banking,human-computer interaction, e-commerce, e-government, as well as in areas of security and defense. Existing biometrics such as fingerprint recognition, iris recognition and face recognition etc., have achieved good result, but there's still need for a new kind of biometrics.Eyebrow follows the principle of universality,uniqueness,stability and collectable, so it can be used in identity authentication, that is, eyebrow recognition. Eyebrow recognition is a new biometrics rising recent year, it has got some evidence of cognitive psychology. This paper did a brief introduction to existing eyebrow recognition methods, and focused on how to recognize eyebrow automatically, mainly discussed how to extract eyebrow form original eyebrow images automatically and how to select initial center for Vector Quantization Classifier automatically.This thesis mainly concerns the following aspects:(1) The automatic extraction of eyebrow: At first, we made gray-scale transformation to 24-bit truecolor eyebrow images at our eyebrow databese and got gray-scale eyebrow images; After that, we compared several image enhancement methods and adopt the threshold expansion method to enhance gray-scale eyebrow image; Later we calculated thresholding automatically by using Otsu's method in image binarization and cleared black pixels near the edge of eyebrow image to erased hair and eyes; At last, we calculated eyebrow region by projection analysis and got pure eyebrow images.(2) The extracting of feature vector: At first we make each column of pure eyebrow images in training set to frequency domain by using fourier transform. Then we construct feature vector by using 32-D lowest fourier coefficient and these feature vectors constitute a training vector set.(3) The construction of feature string: At the beginning, we generated a group of clusters by using maximalθ-distant subtrees based clustering algorithm, then we calculated the centers of each clusters, these centers were initial centers of K-Means cluster algorithm. After that, using K-Means cluster algorithm to generate VQ classifier. At last get feature string from pure eyebrow image by using VQ classifier.(4) The recognition of eyebrow: We matched the test feature string generated by the above-mentioned method and template feature string by edit distance, to recognize the identity of the eyebrow image.(5) Experiment: Four groups of experimental based on the above-mentioned method were done and achieve a highest accuracy of 91%. The results have shown the effectiveness and feasibility of eyebrow recognition.At last, the existed problems in eyebrow recognition method and the aspect of following research is indicated.