This paper proposes a novel junction detection algorithm which combines image intensity variation and edge contour information. Firstly, use the Canny edge contour detector to extract edge map. Secondly, imaginary parts of multi-scale Gabor filters (IPGFs) are used to smooth the input image, and then acquire the normalized information entropies at various scales. Finally, the multiplication of different normalized information entropies is served as a new interest point measure, which is used to detect interest point. The proposed method has two advantages: on one hand, the proposed method not only utilizes contour shape and local maximum curvature to extract junctions but also considers the gray variation information of edge pixels and their surrounding pixels; on the other hand, the multi-scale product is served as junction measure which can inhabit the non-junction. Compared with four other state-of-art methods, the proposed method showed excellent performance in terms of the average repeatability and the localization accuracy.