Relative Performance Analysis of Edge Detection Techniques in Iris Recognition System

Nowadays biometric authentication is mostly used to protect access to highly confidential assets. Identity management is more important than ever as to strengthen global security, make transportation safer and protect a vital commercial entrance. Iris recognition recognizes people very accurately and reliably based on the random texture that visible on the iris of the eye while also being one of the least invasive. To thrive an iris pattern identification algorithm with approximately zero false acceptance rate, this paper analyses various edge area detection techniques applying in different iris images. Every algorithm should pass through some basic image pre-processing steps due to iris image quality, including nonlinearly distorted iris images, iris images at a distance or moving condition, and falsified iris images all are open difficulties in this recognition system. A basic work to solve the problems is to study, design and develop various algorithms for all these variations of images. Although current literature has a variety of edge detection techniques like canny, Sobel, Prewitt, this paper does not always lead to acceptable results. But the experimental result demonstrates that the Canny's algorithm performs comparatively better to detect edge points in a digital image even at a slow rate change of gray level. We have examined noisy iris images applying salt and pepper noise as well as Gaussian noise. Different filtering techniques can be applied to eradicate the undesirable noise. The effects of edge detection techniques of the mean, median and Gaussian filtered images have been observed in the paper. Gaussian filter is powerful for noise eradication. When the Gaussian filter in the vertical orientation is implemented to normalize iris image, the time complexity of this approach is reduced considerably. Experimental results show the validity of this methodology.