Fusion of Phase Congruency and Harris Algorithm for Extraction of Iris Corner Points Gugulethu Mabuza-Hocquet CSIR: Modelling and Digital Science UJ: Engineering and the Built Environment Pretoria, South Africa GMabuza@csir.co.za Fulufhelo Nelwamondo CSIR: Modelling and Digital Science UJ: Engineering and the Built Environment Pretoria, South Africa FNelwamondo@csir.co.za Abstract—Iris recognition uses automated techniques to extract iris features which are stored in a database as a feature tem- plate to be later used for individual identification and authen- tication. Strict image quality control is a basic requirement for most iris identification systems. Low cost devices used under uncontrolled environments acquire poor iris images with in- consistent illumination and specular reflections. These factors inflict challenges towards the accurate identification and ex- traction of reliable iris features. This work proposes a fusion of Phase congruency and Harris algorithm to detect corner fea- tures found within the arrangement of iris patterns. This fu- sion produces a feature vector with the exact location of corner features that are not only congruent in phase but are also in- variant to illumination and rotation. Results of the proposed approach are tested on two non-ideal databases and obtain an accurate match rate of 99.9% while producing a feature tem- plate of 512 bits that requires low storage space. Keywords- phase congruency; harris corner detector; iris segmentation; Chan-Vese algorithm ; feature extraction. I. INTRODUCTION Research in iris biometrics has gained more attention over the years due to the stability, uniqueness and reliability presented by iris patterns [1, 2, 3]. Fig.1 shows the stages of a classical iris recognition system that prevail after image acquisition. The stages are; (1) segmentation, (2) normaliza- tion, (3) feature extraction and (4) matching. Figure 1. Traditional iris recognitin system Each stage of the iris recognition system has automated and traditional algorithms that have been successfully used to extract and store iris features as a code in a template. The traditional methods include the use of the integro-differential equation for iris segmentation [1]. This algorithm assumes that the pupil and iris boundaries are concentric circles, i.e share the same center, which is not generally the case. Also, when the acquired images are captured under uncontrolled conditions such that an image is not focused around the cam- era’s centered gaze, due to factors such as camera lens, off- angle images and head tilt, the segmentation process fails, as shown in Fig 2. The normalization stage uses the rubber- sheet model to transform the segmented iris from a Cartesian plane to polar coordinates. This method changes the geomet- rical structure and arrangement of the iris patterns and there- fore, the extracted features cannot be traced back to the original image. The 2D Gabor wavelets used for feature extraction to generate an iris code of 1024 bits, do not dis- criminate the various feature types found within the iris. Lastly, the Hamming distance, which is a dissimilarity measure, is used to quantify a match between a reference template and its query. The similarity in the traditional algo- rithms is that they are either intensity based methods or re- gion based methods. Therefore, (i) based on the device used during image acquisition, (ii) the exposure of the capturing device to natural scenes or uncontrolled environments; which leads to inconsistent illumination causing specular reflections within the iris, there is often a compromise in the nature and quality of the features to be extracted. These external factors pose a negative impact not only on the captured images, but also the detection, quality and reliability of the features that have to be extracted. Since the goal of any recognition sys- tem is to produce a perfect match between a reference image and its query image, the identified and extracted features types should be such that they are not sensitive to these con- ditions [4], but instead can conform to any environment. The biological formation and complexity of iris patterns requires a robust feature detection and extraction algorithm in order to achieve reliable feature points that are invariant under any illumination conditions. Figure 2. Segmentation failure with traditional algorithm. A corner and edge detection method that measures the significance of features in computer images is known as Phase Congruency (PC), [5, 6]. Its invariance to image illu- mination and contrast makes it a robust and reliable feature 2015 Third International Conference on Artificial Intelligence, Modelling and Simulation 978-1-4673-8675-3/15 $31.00 © 2015 IEEE DOI 10.1109/AIMS.2015.57 315 2015 Third International Conference on Artificial Intelligence, Modelling and Simulation 978-1-4673-8675-3/15 $31.00 © 2015 IEEE DOI 10.1109/AIMS.2015.57 315 2015 Third International Conference on Artificial Intelligence, Modelling and Simulation 978-1-4673-8675-3/15 $31.00 © 2015 IEEE DOI 10.1109/AIMS.2015.57 315 2015 Third International Conference on Artificial Intelligence, Modelling and Simulation 978-1-4673-8675-3/15 $31.00 © 2015 IEEE DOI 10.1109/AIMS.2015.57 315