Edge detection method for terahertz image based on principal component analysis and active contour model

Affiliation of Author(s):电子信息工程学院

Journal:ACM Int. Conf. Proc. Ser.

Abstract:Terahertz images are usually blurry and inhomogeneous, thus it has poor performance to use traditional active contour models to detect the edges in terahertz image. In order to improve the robustness and efficiency of edge detection for terahertz images, an edge detection method based on principal component analysis (PCA) and active contour model is proposed. Firstly, the terahertz image is divided into a number of patches, then PCA method is used to analyze the patch vectors and they are projected onto a low-dimensional subspace. Then a fitting energy functional is constructed by utilizing the obtained projected vectors. In order to minimize the energy functional, a level set evolution method is used and the final location of zero level set is the edge in the image. Finally, a number of experiments on several kinds of terahertz images are made to compare with other edge detection methods based on Chan-Vese (CV) model, local binary fitting (LBF) model, active contours with selective binary and Gaussian filtering regularized level set (SBGFRLS) model, and robust and fast active contour models (RFACM). The experiment results show that the proposed method is more robust and efficient than the above four methods for terahertz image edge detection, which proves that the proposed method is a fast and robust edge detection for method terahertz image. © 2018 Association for Computing Machinery.

Translation or Not:no

Date of Publication:2018-03-16

Co-author:田华臣,庚嵩

Correspondence Author:wyq

Date of Publication:2018-03-16