Infrared Dim Target Detection Based on Multi-scale Space and Multidirectional Gradient Search

摘要

According to the difference between gradients of targets and background in an infrared image, a multi-scale space and multidirectional gradients of search method for detecting infrared weak targets is presented. Before detecting the target, top-hat algorithm is used to preprocess the image to highlight weak targets and suppress large background areas. Then, using the normalized LOG (Laplacian Of Gaussion) scale space, the center coordinates and sizes of proposed objects in the image are obtained according to the different filter responses of the image pixels. An adaptive window is defined for each proposed target. Adaptive threshold of gradient is calculated based on the response image. Multidirectional gradients of search are performed based on the threshold. Next, the response image is binarized and the objects is detected after blob analysis. Theories and experiments indicate that our method can detect targets of different sizes under noisy infrared images.