Segmentation algorithm for small targets based on improved data field and fuzzy c-means clustering

Abstract

Fuzzy c-means (FCM) method has been widely used in image segmentation and applications of fuzzy techniques for unsupervised classification. However, given its sensitivity to initial cluster centers and poor resistance to noise, the performance of FCM method may be reduced. Thus, a novel segmentation method is proposed based on the improved data field and FCM clustering. First, an image data field is reconstructed using the improved data field method to balance the distribution of background and to improve its robustness to noise. Second, the potential hearts φ1, φ2, …, φc of the image data field are selected as the initial cluster centers. Furthermore, the potential value ϕ of the data field (space information) is conjugated with the variance σ of the grayscale image (grayscale information) to modify the degree of belonging functions of the standard FCM algorithm, which is utilized to constantly updated to identify the desired cluster centers in the image data field. Finally, the obtained segment results are quantified by the misclassification error (ME) and the mean structure similarity (MSSIM). Moreover, the robustness to noise performance of the abovementioned algorithms is tested. Comparing with the traditional FCM algorithm, the experimental results show that the presented method for the images with complicated background, dim targets, and low-contrast grayscale between targets and background performs better and has better rationality and robustness to noise.