FLIC: Fast linear iterative clustering with active search

• DOI: 10.1007/s41095-018-0123-y
• Publication Date: 2018-10-27T03:58:38Z
• Authors: Jiaxing Zhao; Ren Bo; Qibin Hou; Ming-Ming Cheng; Paul Rosin
• Keywords: Benchmark (surveying)

In this paper, we reconsider the clustering problem for image over-segmentation from a new perspective. We propose novel search algorithm called "active search" which explicitly considers neighbor continuity. Based on method, design back-and-forth traversal strategy and joint assignment update step to speed up algorithm. Compared earlier methods, such as simple linear iterative (SLIC) its variants, use fixed regions perform steps separately, our scheme reduces number of iterations required convergence, also provides better boundaries in results. Extensive evaluation using Berkeley segmentation benchmark verifies that method outperforms competing methods under various metrics. particular, is fastest, achieving approximately 30 fps 481 × 321 single CPU core. To facilitate further research, code made publicly available.