Iris Image Recognition Using Gradient Features Based Extraction

Nowadays a number of techniques have been proposed for iris recognition based on several features and classifiers. In this thesis, we propose an iris recognition technique for distantly acquired face images using image gradient-based feature extraction with different distance classifiers. We have used Sobel, Robert, one-dimensional derivatives [1, 0, 1], and Prewitt operators to extract the image gradient features from the iris images. L2-norm and L2-Hys are implemented to normalize the whole gradient vector. In the classification stage, we adopt several distance functions like Euclidean distance, Standardized Euclidean distance, Minkowski distance, City Block distance, Chebychev distance, Cosine distance, Correlation distance, Hamming distance, and Spearman distance to build the classification models. For the experimental evaluation, we have used the CASIA-v4 image database including regular and complicated iris images in the same subject. We statistically evaluate the performance of the classifier by utilizing the confusion matrix and analyze the results graphically by the Receiver Operating Characteristic curves for image gradient features. It can be observed from the experimental results that the above distance functions influence the recognition accuracy of the classifier and the performance of the distance metric depends on the properties of the data. Experimental results for iris image recognition demonstrate the effectiveness of GLAC compared with other methods, such as HOG. It enables the extraction of richer information i.e. 2nd order statistics of gradients which means auto-correlations from images and obtaining more discriminative powerful features than other standard histogram-based methods.

• Supervision