Premium
Novel local information kernelized fuzzy C ‐means algorithm for image segmentation
Author(s) -
Li Songcheng,
Lu Junyong,
Cheng Long,
Li Xiangping
Publication year - 2021
Publication title -
international journal of imaging systems and technology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.359
H-Index - 47
eISSN - 1098-1098
pISSN - 0899-9457
DOI - 10.1002/ima.22491
Subject(s) - artificial intelligence , pattern recognition (psychology) , computer science , outlier , segmentation , image segmentation , robustness (evolution) , initialization , algorithm , scale space segmentation , fuzzy logic , noise (video) , image (mathematics) , biochemistry , chemistry , gene , programming language
In MRI, the image with poor quality, especially the image with noise interference or low contrast, may provide insufficient data for the visual interpretation of the affected part. Image segmentation provides an effective method to facilitate early detection and further diagnosis. By introducing a Particle Swarm Optimization (PSO) initialization step and a novel dissimilarity measure metric, we present a local information kernelized fuzzy C‐means (LIKFCM) algorithm for image segmentation. The dissimilarity measure metric, considering an adaptive tradeoff weighted factor, incorporates the Mahalanobis distance and outliers‐rejection‐based spatial term which eliminates unreliable neighboring information. By using this dissimilarity measure metric, the new algorithm could take reliable contextual information into account and achieve better segmentation results on images with complexed boundaries. Furthermore, the adaptive tradeoff factor depends on a fast noise estimation algorithm. This factor avoids subjective adjustment and makes the LIKFCM algorithm more universal. To evaluate the performance of the proposed algorithm both quantitatively and qualitatively, experiments are conducted both on synthetic images and real‐world images with different kinds of noise. Segmentation Accuracy (SA) and Comparison scores are used to evaluate the performance of both proposed algorithm and other methods. Experimental results illustrate that the proposed algorithm has better performance on denoising and reserving useful edges. The LIKFCM algorithm not only shows more robustness to noise but also preserves the texture details of the images.