Adaptive Threshold and Piecewise Fitting for Iris Localisation

Abstract

Machine vision needs detectors to get features and properties of objects in each image to be used for ensuring the effectiveness of each study conducted. The most recognisable Canny edge detection method aims to lessen noise and eliminate unwanted edge by employing threshold values and hysteresis for localisation advantages. Canny method uses a high threshold and a low threshold to decrease of false positive pixel edges and to describe the contours in the image crucially as compared to use one fixed threshold value for the maximum gradient is not the best option. Nevertheless, using two unchanged threshold values still do not guarantee to provide the best results because the image contains huge variations. Previously, adaptive thresholds have been introduced for specific domain only. Here, a technique to determine the threshold values from the foreground and background image pixels in the global and local image will be used for further analysis. This approach involves partitioning an image into several similar size blocks at multiple resolution levels. Then, a sampling procedure uses on global and local images to acquire the best threshold value by selecting the highest between the class variance values. Finally, piecewise cubic spline will completely segment the boundary for iris and pupil region. Experiments have been done on CASIA V2 datasets. The results show that edge image obtained outperform the Canny method and previous work. Iris localization image obtained also more accurate compared to Hough method.