Affiliations: [a] School of Mechanical and Electrical Engineering, Wenzhou University, Wenzhou, China
| [b] College of Mechanical Engineering, Zhejiang University, Hangzhou, China
Correspondence:
[*]
Corresponding author. Sheng Luo, School of Mechanical and Electrical Engineering, Wenzhou University, Wenzhou 325000, China. Tel.: +86 13758711910; Fax: +86 05778868 9166; E-mail: [email protected].
Abstract: In spite of a large volume of segmentation algorithms, it is difficult to coherently segment a series of images from the same scene. Semantic segmentation might be a solution, but it is sophisticated with heavy computing cost. Many contour detectors could locate objects accurately, but the contours are broken, and Watershed algorithm could find the subtle ridges that could connect the broken contours, but often fails for over-segmentation. We try to combine these two methods to decompose images into meaningful regions, and propose an index S-measure to measure the segmentation consistency between images. Two steps are involved: detect edges by contour detector, and fuse the broken edges by Watershed algorithm. To fuse the broken edges, two ways are proposed, dilating edges with same template and with adaptive width templates. Experiments illustrate that the decomposing is fast, effectively, meaningful, robustly, and coherently, and the segmentation is consistent between images even with some transformations.
