My paper titled “Robust Edge-Stop Functions (ESF) for Edge-Based Active Contour Models in Medical Image Segmentation” was just accepted in the IEEE Signal Processing Letters (SPL). The paper is one of my Ph.D. works and going to be a chapter for my thesis. The IEEE SPL ranks among the Q1 journals based on The SCImago Journal & Country Rank (SJR). Information about the journal rank can be seen here. I have chosen an open access (OA) option for the manuscript so you can find and download it freely from its source, IEEE Xplore.
The idea is quite tricky. We utilize pixel classification scores to enhance the traditional edge-stop function. Pixels with score 0 or 1 are not ambiguous. They are clear to be classified as either background or foreground and we give them a value of 1 for their clarity. In contrast, pixels with score 0.5 are very ambiguous and the boundary may lie in the vicinity of those pixels. We give them a value of 0 for the ambiguity. Then, we define a function ρ(s) for mapping scores ∈ [0,1] to values ∈ [0,1]. Through energy minimization using the level set method, we can find the desired boundary. The comparison of edge-stop function with and without enhancement is described in the following animation. The traditional ESF is used in the left image while the robust ESF in the right. Click on the image to enlarge.
You may download the source code for either academic or research purpose by giving an appropriate credit for the following work:
A. Pratondo, C.-K. Chui, and S.-H. Ong, “Robust Edge-Stop Functions for Edge-Based Active Contour Models in Medical Image Segmentation,” IEEE Signal Processing Letters, vol. 23, no. 2, pp. 222 – 226, 2016
Please note that I used a function from Li et al. for the distance regularized level set evolution (DRLSE) implementation. The detail about DRLSE can be found in the following publication.
C. Li, C. Xu, C. Gui, M. D. Fox, “Distance Regularized Level Set Evolution and Its Application to Image Segmentation”, IEEE Trans. Image Processing, vol. 19 (12), pp.3243-3254, 2010.
The source code is available here. Should you have any questions, please feel free to contact me.
2 responses to “Robust Edge-Stop Functions for Edge-Based Active Contour Models in Medical Image Segmentation”
I read your article and used your source code, but your code is incorrect. create_rho has a little error. this function uses channel one (red) and channel three (blue) Respectively for foreground and background. so assume my photo has a white region that every channel (rgb) is 255. this function that pixels are white considers for both background and foreground.
Can you help me?
Hi,
You can create more conditions to identify a pixel blue or red; e.g, red for channel 1 = 255, channel 2 = 0, channel 3 = 0.