Effect of supervised region of interest against edge detection method for iris localisation

Abstract

With the recent developments in information technology, health diagnosis based on iris analysis and biometrics has received considerable attention. For iris recognition, iris localisation, which is not an easy task, is an important phase. Moreover, for iris localisation, dealing with nonideal iris images could cause an incorrect location. Conventional methods for iris location involve multiple searches, which can be noisy and outdated. Such techniques could be inaccurate while describing pupillary boundaries and could lead to multiple errors while performing feature recognition and extraction. Hence, to address such issues, we propose a method for iris localisation of both ideal and nonideal iris images. In this research, the algorithm operates by determining all regions of interest (ROI) classifications through the use of a support vector machine (SVM) as well as the application of histograms that use grey levels as descriptors in all regions from those exhibiting growth. The valid region of interest (ROI) obtained from the probabilities graph of an SVM was obtained by examining the global minimum conditions determined using a second derivative model of the graph of functions. Moreover, this helped to eliminate the sensitive noises and decrease the calculations while reserving relevant information as far as possible. During the experiment, the comparison edge detection method was used with Canny and a multi-resolution local approach. The results demonstrated that the proposed ROI provided better results compared with those obtained without ROI.