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

Abstract

The study was focus on the improved accuracy of Growcut algorithm for sclera segmentation and Multiscale Retinex for image quality enhancement. The primary advantage of the proposed method was the improved accuracy and the image quality enhancement compared to the prior segmentation and enhancement approach developed for sclera segmentation and image enhancement process. The Python Interpreter, the ND-IRIS-0405 Database and the publicly available images were used to evaluate the proposed segmentation method. The interactive Growcut algorithm used a semi-automated technique that provided a reliable method for segmenting the sclera, both dimensional and non-dimensional images that can segment both RGB and grayscale images with greater interactivity and user control of the segmentation process. The Growcut algorithm demonstrated a minimum accuracy of 93% for sclera segmentation. It also displayed more improved and more accurate segmentation result of sclera region as compared to K-means and Thresholding algorithm. Segmentation efficiency improved from 72% to 97%. In the same way, Multiscale Retinex algorithm showed excellent performance concerning image quality enhancement in improving low contrast, dark and poor lighting condition images with an average of 1.59. Thus, Growcut and Multiscale Retinex algorithm lessen the problem of K-means and Thresholding regarding sclera segmentation and image enhancement that demonstrates its accuracy, quality image enhancement with user interactivity and the cellular automaton.