Abstract
To segment infrared dim target, a combined marker based watershed algorithm is proposed, analysed, and verified.
Introduction and Algorithm Description
Marker based watershed segmentation [1] on infrared image is an important technique for target recognition. To efficiently segment infrared dim target, a new marker based watershed algorithm is proposed.
First, row demeaning which is derived from infrared image property [2] is used to enhance the target through reducing the grey intensity disparity of different row and smoothing noises in the same row. Second, non-parameter uniform kernel [2], which suppresses clutter background effectively while eliminating target less, efficiently enhances the target again. Third, linear extension further enhances the target. Fourth, some pixels of the target region are segmented as possible markers through the Otsu method. Fifth, the real foreground markers of target regions are obtained through filtering and connecting the possible markers by using COC filter. COC filter is the serialization and combination of morphological closing [1], opening [1] and closing, which makes up the shortcomings of separately using of the opening and the closing. Sixth, the Euclidean distance transformation is used to differentiate the background following foreground markers. Then, the watershed [1] is directly used to separate the background following different foreground markers. Finally, the final marker image, in which each foreground marker is enclosed by the corresponding separated background region, is formed through a logical OR operation on the foreground marker image and separated background image. So, after generating the gradient image by using the Sobel detector, the dim target is efficiently segmented by operating the watershed on the gradient image guided by the final marker image.
Infrared dim ship target images were used to test the algorithm. The results verified that the proposed algorithm was efficient. The combination of infrared image property and non-parameter kernel method, morphological operations and distance transformation produce an effective final marker image which makes the marker finding more efficiently and accurately. So, the watershed can efficiently segment the targets.