International Journal of Applied Mathematics and Machine LearningAuthor's: Jie Zhao, Guangmei He, Yu Zhang, Xin Wen and Peipei Wang
Pages: [65] - [80]
Received Date: December 23, 2015; Revised January 18, 2016
Submitted by: Yang Li.
DOI: http://dx.doi.org/10.18642/ijamml_7100121606
Aiming to solve the problems such as the weak edge contour of the target, the low gray deviation between targets and background as well as the weak boundaries in infrared images and the low noise-proof ability which are hard to be solved by the Distance Regularized Level Set Evolution (DRLSE), this paper purposed the Distance Regularized Level Set Evolution Mode Based on Phase and Gradient Information (PGDRLSE), and this model reconstructed indicator function of boundary by combining the gradient and phase information, and then came to the new energy function. During the process of detecting and tracking the moving target, this article detected the target by using background subtraction method together with frame difference method and obtained the minimum circumscribed rectangular frames of moving target in every infrared video frame with the mean-shift method, and utilized the PGDRLSE to obtain moving human target outline eventually. Experiment results showed that this method could realize the accurate extraction of weak boundary of moving target in infrared images, and accelerate the evolution speed. At the same time, it has strong noise proof ability.
contour extraction, DRLSE model, variational level set, gradient and phase information, mean-shift.
