Master Thesis A Fusion of Features Exactions to Recognize Human Ear Biometrics

Abstract

The demand for more secure authentication has increased on several occasions. Exploiting biometrics in various forms such as face, voice, handwriting, and gait recognition is a reliable method for authentication. Recently, the analysis of ear images as a biometric method has become a robust identification method. A number of researchers have shown that ear recognition is a viable alternative to more common biometrics such as fingerprint, face, and iris recognition, because the ear is relatively stable over time, non-invasive to capture, expressionless, and both the geometry and shape of the ear have significant variation among individuals. Researchers have tried a variety of methods to improve ear recognition. Some researchers have enhanced existing algorithms to assist in recognizing individuals by their ears. Others have taken algorithms that have been tried and tested for another purpose, such as face recognition, and applied them to ear recognition. These approaches have resulted in a number of state-of-the-art effective methods for identifying individuals by ear. Many of the challenges occur due to errors in the method of capturing images, poor illumination, image dimensions, off-angle ears, etc. The various methods have been adopted by researchers in order to enhance and increase the performance of ear recognition. Most of the ear recognition systems incorporate processes before the feature extraction stage; first, the pre-processing stage, which is done to enhance only the region of interest. This stage includes segmentation and normalization. Subsequently, to enhance the normalized ear image. In this research, the Histogram Equalization (HE) technique has been implemented to facilitate the application of the feature extraction step. Then we presented an approach based on a fusion of two different techniques of feature extraction: Histograms of Oriented Gradients (HOG) and Local Binary Patterns (LBP) to extract the desired features. whereas Principal Component Analysis (PCA) is used to reduce the space of the feature dimensionality. For classification, Linear Discriminant Analysis (LDA) is used. The proposed technique is applied to the images of the IITD I database. The proposed method has yielded significant achievements compared with other studies.