Selective  Segmentation Method Based on Exponential Weighted Geodesic Distance Driven Model and Thresholding Method

Abstract

Selective segmentation represents a pivotal image processing technique within the domain of computer vision. Its objective is to facilitate the precise identification and extraction of a  region of interest (ROI) within an image, while excluding background regions that are irrelevant to the task at hand.  Nevertheless, the challenge of segmentation for images with noise, particularly those with grayscale inhomogeneity, still exists in many state-of-the-art segmentation models. In order to address this challenge, this paper proposes a novel weighted selective segmentation method based on an exponential weighted geodesic distance-driven scheme and a thresholding method inspired by the K-means method. In particular, the proposed method employs a two-stage approach. Initially, the property of the geodesic distance based on the exponential transformation, whereby the distance is smaller within the ROI and larger in the background, is leveraged to enhance the contrast between the ROI and the background. Subsequently, the segmentation of the ROI is obtained through thresholding on the enhanced image generated in the initial stage. The experiments demonstrate that our proposed method is capable of suppressing the influence of noise and grayscale inhomogeneity to a certain extent, and that it yields higher segmentation accuracy than several existing state-of-the-art variational selective segmentation models.

Keywords: Selective segmentation, Exponential weighted geodesic distance (EWGD), Scaled alternatingdirection method of multiplier (Scaled-ADMM), Thresholding method