Gaussian Regularization Based Active Contour Model for Extraction of Shape Boundaries in Vector Valued Images

Abstract

—Extraction of shape boundaries also known as image segmentation that tries to divide a digital image into several segments for further processing is an important step in digital image analysis. The segmented images are useful in many applications such as object detection, object classification, biometric authentication, and disease diagnosis. This task can be accomplished effectively using active contour model. The two types of active contour models are global segmentation and selective segmentation. A global segmentation model is a way for segmenting an image's entire objects. Global models, unfortunately, are unable to segment a single object’s shape boundary that must be extracted. A selective segmentation model, which seeks to segment selected object’s shape boundary in an image, can overcome this restriction. The recent selective segmentation model for vector valued images has a significant computational cost due to the requirement of solving the curvature term, that result from the usage of traditional regularization term in the formulation. Hence, we proposed a new selective segmentation model for vector valued image by replacing the traditional regularization term with Gaussian function which is easier and faster to solve. The proposed model was tested using MATLAB software in segmenting synthetic, natural and medical images. Numerical experiments shows that the proposed model was about 247 times faster than the existing model with a comparable accuracy.