Distribution Information Based Intuitionistic Fuzzy Clustering for Infrared Ship Segmentation

This paper presents a distribution information based intuitionistic fuzzy clustering method for infrared ship segmentation. The algorithm could effectively suppress the influences of nontarget objects with high intensity and intensity inhomogeneity in the infrared ship images. There are mainly two improvements in this paper. First, it proposes a fuzzy clustering algorithm incorporating global distribution information of ship targets in the form of the Gaussian model. The spatial information, along with intensity, is used to exert different effects on different classes. Second, an intuitionistic fuzzy clustering way is incorporated into the process of ship segmentation, which combines the intensity distribution information of the local region. The intuitionistic fuzzy distance and local intensity distribution information would help in solving the problem of intensity inhomogeneity and blurring edges. Experiment results on the dataset containing 200 infrared ship images indicate the superiority of the proposed method compared with other state-of-the-art methods.

I. Introduction

Infrared (IR) images have a wide range of applications in military and civilian domains [1]. Out of the imaging principle, which is based on temperature characteristics of objects, the imaging environment is not limited by weather or illumination conditions. In recent years, IR imaging has been utilized in maritime surveillance including ship detection, segmentation, and recognition [2], [3]. This makes the applications of an IR imaging technique have great military significance. IR ship segmentation is a key step of target recognition and analysis, whose results would exert a remarkable effect on the accuracy of downstream processes. Generally, a precise segmentation result of the ship target would contribute a lot to improving the accuracy rate of identification. However, it is difficult to obtain an accurate segmentation result due to the restriction of properties of IR images. Uneven heat dissipation leads to an uneven intensity distribution of targets [4]. Blurring edge is also a common phenomenon in IR images, which results in a low resolution along with losing texture details. Besides, there would be a complex background containing mist, surface reflection, and nonship objects with high intensity [5]. For these reasons, IR ship segmentation is still a challenging task.