Parameter-free Image Segmentation Based on Extreme Learning Machine
Hongwei Zhang, Liuai Wu, Yanchun Yang · 2019
For the problem of spending much time on adapting parameters, a parameter-free image segmentation method based on extreme learning machine (ELM) is proposed. Firstly, each image is segmented as superpixels by simple linear iterative clustering (SLIC) with different parameters. Secondly, each superpixel segmentation result is combined with some rules, and initial segmentation results are obtained. Each initial segmentation result is evaluated, and the parameter with the best performance is selected as its class. Thirdly, in order to construct the training sets of ELM, the cooccurrence of each image is constructed, and some of its attributes are calculated as its features, and a parameter-free framework is learned by ELM. The experimental results show that the proposed method in this paper gets better segmentation results, which is closer to human annotation than other methods.