Interactive Growcut Multi-Label N-D Segmentation Based Algorithm for SCLERA Segmentation

Abstract

This paper presents a computationally efficient sclera detection, segmentation and enhancement approach. The proposed sclera detection and segmentation approach is developed based on the cellular automaton which evolves using the Grow-Cut algorithm and enhancement approach based on Multiscale Retinex. The major advantage of the developed approach is its computational simplicity, speed and accuracy as compared to the prior detection, segmentation and enhancement approaches developed for sclera segmentation images. The new used approach was tested in Intel i5 4570 with 3.2 GHZ processor and 16GB RAM and Intel Dual Core with 2.48GHZ processor and with 2GB RAM. Using these two different units, the result shows that segmentation under Intel i5 segment sclera for three 3 hours with an image size of 640x480 pixels while in Intel Dual Core 2GB RAM having the same image size segments the sclera for fourteen (14) hours and eighteen (18) minutes. The experimental results obtained from ND-IRIS 0405 database and publicly available images and respectively achieved better performance in terms of computational simplicity, speed and accuracy as compared to the previous proposed method. Multiscale Retinex enhanced image very fast with a highest minimum execution time of seventeen seconds (17secs) for an image size of 640 x 480 pixels for both colour and gray scale. The proposed approach for enhancement shows better performance in enhancing low contrast, dark with poor illumination images but colour rendition is quite poor for images having good illumination. The experimental results presented in this paper clearly demonstrate the applicability of the proposed sclera detection, segmentation and enhancement approach, i.e., significant reduction in computational complexity in detection segmentation and enhancement while providing comparable segmentation and enhancement performance.