Unsupervised Segmentation of Image Using Novel Curve Evolution Method

Abstract

Here, a novel algorithm for unsupervised field-depth (DOF) image segmentation is defined. To detect the object of interest (OOI) in the saliency space, a multi-scale re-blurring technique is used first. Firstly blurring is carried out to remove artifacts, later reblurring procedure. Thereafter, an active contour model based upon hybrid energy system is suggested to evaluate the OOI boundary. A global energy element relevant to the saliency map is implemented in this model to find the globally minimum, and a local energy term about the low DOF picture is used to increase the precision of segmentation. Additionally, this model is equipped with an elastic parameter to offset the weight of global and local resources. In addition, an unsupervised approach for initializing curves is intended to reduce the amount of iterations for evolution. More the iterations, the complexity and computation time to obtain the results may hike up leading to slow up the process of acquiring precise contours. Lastly, we perform experiments on different low DOF pictures, and the resultant demonstrates the high precision and robustness of the proposed method.