Research On The Algorithm Of Human Ear Image Processing

As information security has been great concerned by people, along with the rise ofbiometric identification technology, and as a new member in the field of biometricidentification-ear recognition cause great concern of experts and scholars when itappeared. Ear recognition is passive identification and it has the advantage of non-contactand concealment and the small size of image etc, and it become popular research directionin the biometric identification technology. Currently rely mainly on the feature extraction ofthe outer ear of ear for identification, but there are low rate recognition of ear images, theear images have the robustness poor ear defects by attitude angle change of image, in orderto overcoming these existing shortcomings of ear recognition, the main work of this paper isdepend on the existing feature extraction algorithms for improvement and innovation toimprove the recognition rate of human ear images.First, the human ear images need preprocessing, including the adjustment of imagecontrast, filtering and histogram equalization for three areas of work, in order to have thepurpose of eliminating light, noise and shooting angle changes on the impact of the imagesand enhancing the quality of images, creating the ideal experimental conditions forvalidating the algorithm in the follow-up.The feature extraction of ear is the key step in the system of ear recognition, determinesthe classification results of high accuracy or low accuracy. The experiment show that asingle feature extraction method requires specific conditions to achieve a higher recognitionrate, but adopting double feature extraction can overcome single feature extraction of thislimitation. In order to improve accuracy results of recognition, this paper presents a new andimproved algorithm of feature extraction-complementary double feature extractionmethod which is based on principal component analysis (PCA) and fisherface. And thismethod is applied to the images recognition of human ear. The experimental results from thehuman ear image library which is provided by Beijing University of Science andTechnology shows that the human ear recognition rate of this method is obviously higherthan the one of single feature extraction of PCA、fisherface and ICA.There is a problem-high feature dimensions in the application of2KDPCA in humanear recognition image feature extraction, resulting in long recognition time, and requiringlarge data storage space. This thesis proposes an improved method. It gets rid of correlation problem from the rows and columns of human ear images at the same time, then we canachieve the goal of lowering the dimensions of extracted human ear feature. and finally weuse the BP neural network classifier for classification. Experimental results show thatimproving the human ear recognition speed, requiring less data storage space, whilemaintaining a high recognition rate.The low recognition rate caused by the attitude angle changing which has been thedifficulty of the study, because some of the existing feature extraction methods have thelower recognition for attitude change of ear image, in this paper, complementary featurefusion algorithm which is based on principal component analysis (PCA) and independentcomponent analysis (ICA) apply to the attitude angle changing of ear image recognition.The algorithm for the recognition of attitude angle changing of ear image recognition hasimproved significantly, and make up the shortfall of the principal component analysis (PCA),independent component analysis (ICA) and other single-image feature extraction for attitudeangle changing of ear image. According to the analysis of experimental data shows that thehuman ear recognition rate of this method is obviously higher than the one of single featureextraction of PCA and ICA for attitude angle changing of ear image, which has a highpractical value.