Research Algorithms Of Infrared Small Target Detection Based On Visual Saliency

In recent years, saliency analysis has been successfully applied in object detection, recognition, image compression and other areas. In these areas, infrared small target detection is also included. It is a key technique in infrared homing guidance, search and tracking, early warning and so forth. Specially, infrared small target detection under the cluttered clouds,the sea-sky background,the complex ground background is still a hot and difficult topic in the current research. Based on the comprehensive summary of previous work, this paper deeply studies the feature extraction of infrared small target image and the construction of the visual saliency model of infrared image in complex background. The main work of this paper is showed as follows:Firstly, in this pap er the mathematical model of infrared small target image is introduced and the characteristics of image target, background and noise are analyzed. We also suppress the noise, and increase the contrast between background and target by image pre-processing.Secondly, under the guidance of the mathematical models of the infrared small target image we propose a new algorithm of saliency object detection based on spectral residual method. This algorithm combines the classical frequency domain method with features of one-order directional derivatives of image to improve the ability of anti- noise and the accuracy.Thirdly,aiming at the defect of spectral residual method,we propose a hypercomplex matrix representation of infrared image and we employ the amplitude spectrum to suppress repeat nonsaliency region. Our method combines the image second-order direction derivative features with gray feature to reconstruct infrared image. The convolution of the image amplitude spectrum with a low-pass Gaussian kernel of an appropriate scale is equivalent to an image saliency detector. Experimental results show that the proposed algorithm has greater ability of target enhancement and background suppression.Finally, we put forward two ideas of saliency image segmentation. O ne of them is a traditional method based on threshold analysis, another is segmentation method based on fuzzy metric. The experimental results show that the former algorithm of the saliency image segmentation more efficient and the ability of accuracy is guaranteed.