Authors: Ping Chen; Peizhen Wang; Yanxiang Han; Zhisheng Zhang
Addresses: School of Mechanical Engineering, Southeast University, Jiangning District, Nanjing, Jiangsu, 21189, China ' School of Electrical and Information, Anhui University of Technology, Huashan District, Ma'anshan, Anhui, 243002, China ' School of Mechanical Engineering, Southeast University, Jiangning District, Nanjing, Jiangsu, 21189, China ' School of Mechanical Engineering, Southeast University, Jiangning District, Nanjing, Jiangsu, 21189, China
Abstract: Automatically recognising the coke microstructures from captured coke microscopic image is a crucial task for analysing their macroscopic properties and guiding their production. This paper aims to predict the coke macroscopic properties using limited microstructure statistical information. The proposed algorithm mainly consists of two stages: image segmentation and microstructure recognition. Firstly, I1I2I3 colour space is divided into grid cells for representing them as colour indexes and coke microscopic image is segmented into several clusters with grid clustering. Furthermore, isotropic and anisotropic textures of coke microstructures are described with LBP texture features and coke microstructures are predicted through matching the planar image statistics using support vector machines. Experimental results show that the proposed algorithm is effective for automatically segmenting and recognising the coke microstructures.
Keywords: coke microstructures; microstructure recognition; image segmentation; grid clustering; SVM; support vector machines; image recognition; colour indexes.
International Journal of Computer Applications in Technology, 2014 Vol.50 No.1/2, pp.51 - 60
Received: 08 May 2021
Accepted: 12 May 2021
Published online: 25 Jul 2014 *