A novel image segmentation method based on fast density clustering algorithm. (August 2018)

• Record Type: Journal Article
• Authors: Chen, Jinyin; Zheng, Haibin; Lin, Xiang; Wu, Yangyang; Su, Mengmeng
• Is Part Of: Engineering applications of artificial intelligence. Volume 73(2018)
• Page Start: 92
• Page End: 110
• Publication Date: 2018-08
• Subjects: Image segmentation -- Fast density clustering -- Parallel computing -- Automatic cluster center determination -- Parameter self-adaptive -- Cuckoo search algorithm (CSA); Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2018.04.023
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Image segmentation is one of the key technologies for image processing. Most image segmentation methods based on clustering algorithms encountered with challenges including cluster center sensitivity, parameter dependence, low self-adaptability and cluster center determination difficulty. Accordingly, a novel image segmentation method based on fast density clustering algorithm (IS-FDC) is proposed in this paper. Pixel similarity is calculated on basis of both pixel value and its position information. IS-FDC is based on fast density clustering algorithm (FDC), in which cluster centers could be determined automatically by multiple linear regression analysis, and the only sensible parameter confidence interval is self-adaptive based on cuckoo search algorithm (CS). Currently density based clustering algorithms applied to image segmentation have difficulties, especially for its high time complexity and memory complexity. Parallel partition and scaling strategies are put forward to speed up clustering process. Multiple images from Berkeley dataset are adopted for simulations and analysis. IS-FDC is compared with several outstanding algorithms based on several evaluation indexes including both supervised and unsupervised algorithm indexes to testify that the proposed IS-FDC is outperformed. Abundant experimental results proved that IS-FDC is robust to parameters, which can automatically determine the number of segmentation and improve the accuracy of segmentation effectively.