Abstract: Line segment detection in low-SNR infrared image is difficult as the gradient field calculation is badly affected by noise. In this paper, a new approach was presented to distinguish line elements and non-line ones in an image according to the fact that gradient field phase distribution of a line area distinctly appears non-isotropic. Local non-isotropy was defined and a calculating operator was presented. By combing phase-grouping algorithm and contrario detection framework, a new line segments detection algorithm was proposed. Experimental results indicate that the algorithm can effectively detect line segments in low-SNR images with a low false alarm rate.