Automatic iterative algorithm for image segmentation using a modified pulse-coupled neural network

• Published in: Neurocomputing (Amsterdam) Vol. 119; pp. 332 - 338
• Main Authors: Gao, Chao, Zhou, Dongguo, Guo, Yongcai
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 07.11.2013; Elsevier
• Subjects: Applied sciences; Artificial intelligence; Bayes clustering method; Computer science; control theory; systems; Connectionism. Neural networks; Data processing. List processing. Character string processing; Exact sciences and technology; Extraction; Image segmentation; Infrared imagery; Iterative algorithms; Mathematical models; Memory organisation. Data processing; Neural networks; Neural threshold; Pattern recognition. Digital image processing. Computational geometry; Pulse-coupled neural network; Segmentation; Software; Thresholds; Image segmentation; Neural threshold; Pulse-coupled neural network; Bayes clustering method; Infrared thermography; Computer vision; Data analysis; Image processing; Brightness; Cluster; Iterative method; Neural network; Modeling; Efficiency; Feasibility
• ISSN: 0925-2312; 1872-8286
• DOI: 10.1016/j.neucom.2013.03.025

The pulse-coupled neural network (PCNN) is widely used in image segmentation. However, the determination of parameter values in the PCNN framework is an unavoidable and trivial task that may cause neurons to behave unexpectedly, thus affecting segmentation performance. Therefore, this paper presents an efficient iterative algorithm using a modified PCNN for automatic image segmentation. In contrast to existing PCNN models, a new neural threshold was first established for the modified PCNN instead of a general dynamic threshold, allowing for greater efficiency in controlling the pulse output. Besides, a varying linking coefficient value was constructed for efficiently adjusting the neural behavior. By incorporating the Bayes clustering method, it thereby extends the feasibility of the model for the extraction of targets with inhomogeneous brightness, thus resulting in a simpler iterative algorithm for segmentation. Experiments on real-world infrared images demonstrate the efficiency of our proposed model. Moreover, compared with simplified PCNN models and classic segmentation methods, the proposed model shows fewer misclassification errors and higher segmentation performance. •We present a novel iterative algorithm for segmentation using a modified PCNN.•The model does not require the parameters setting by trial-and-error•A varying linking coefficient value can be adaptively adjusted during iteration.•Bayes clustering method is employed to promote the performance.