Accurate and Robust Iris Recognition Using Modified Classical Hough Transform

Circle Hough Transform (CHT) is a robust variant of the Hough Transform (HT) for the detection of circular features. Accurate iris recognition is one of the application areas of this technique. Robustness and accuracy are of utter significance but at the expense of high-computational time and space complexity as the method processes the entire image provided as an input. The present work formulates a computationally more efficient suggested solution as a modified Circle Hough Transform (MCHT) by fixating time–space complexity in terms of reducing the area of the image to cover (the number of pixels to process), and hence significantly decreasing computational time without compromising the accuracy of the method. The modified method is tested on a sample set of collected iris images. Each image is divided into three different sized skeleton grids of size 3\(\,\times \,\)3, 5\(\,\times \,\)5 and 7\(\,\times \,\)7 (pixels). The center of each grid type applied on the image gives the Region of Interest (ROI) of the image sufficient to detect circular parameters as center and radius of the iris using CHT. The experiment shows the comparison of computational time required to detect the iris from CHT applied to the whole image versus the computational time required to detect the iris from just the ROI of the image using the grids. Additionally, the results of the comparison of the expected time and observed time of detection of the iris over a large number of images is presented. There is a substantial reduction in computational time complexity up to 89% using the 3\(\,\times \,\)3 sized grids, and up to 96% using 5\(\,\times \,\)5 sized grids and up to 98% in 7\(\,\times \,\)7 sized grids with equally fair amount of reduction in space utilization. The experiment was performed to observe which grid size gave the most accurate center and radius values along with the most efficient performance. The results showed that the 3\(\,\times \,\)3 and 5\(\,\times \,\)5 sized grids provided better results as compared to the 7\(\,\times \,\)7 sized grids, the results of which lacked accuracy for some images. From the results of the experiments with varying grid sizes, the conclusion obtained is that the accuracy is compromised by grid sizes 7\(\,\times \,\)7 and higher, and grid sizes of 3\(\,\times \,\)3 or 5\(\,\times \,\)5 provide the most accurate and efficient iris detection.