Research On Infrared Weak Target Detection Algorithm Based On Scale Space

Target detection and tracking is an important research issue in the field of computer vision (CV). Extracting the moving target accurately from the image sequence is a key part in the application of intelligent transportation system, intelligent monitoring system and intelligent navigation.Considering about the increasing development of network and information technology, infrared target detection has been the hot topic in varieties of image-correlated industry recently, especially in the military field and industry application. It relies on computer technology and artificial intelligence, not only can complete the extract the small target effective, but also can enhance the quality of the image. It laid the good foundation of target tracking. However, the complex background noise and the lack of detail feature on account of the target shape change may cause many difficulties to achieve infrared target detection. In the above factors, the target motion posture and size changes will make the problem more complicated. The scholars all over the world have figured out a lot of improved algorithms to overcome these problems. On the basis of predecessors'research, this paper proposes three kinds of effective multi-scale detection algorithm through deep analysis and exploration, the main contents are as follows:(1) An infrared small target method based on the saliency and scale-space theory is presented for complex sky background. Firstly, a spectral residual is used to deal with the original image in order to reduce identification area of infrared small target. Secondly, the Difference of Gaussian (DoG) operator is used to obtain scale space of the image after preprocessing, and feature points are detected, which gets an optimal scale image. Then, the feature images are weighted and fused. Finally infrared small target detection is achieved by the segmentation of information entropy. Simulation results show that compared with other the reference algorithm, the proposed method can detect the infrared small target more effectively and enhance the SCR of target image. At the same time, the algorithm can effectively detect dim targets in different complex scenes and lay the foundation for an infrared small target tracking application.(2) We propose an efficient small target detection algorithm that is mainly based on the dual multi-scale filters which work sequentially. The algorithm consists of two stages:at the first stage, Spectrum Scale-Space (SSS) is used as the pre-process procedure to obtain the multi-scale saliency maps, which can suppress the low frequency background noise and make the target region prominently at different scale levels, the more detail information and feature information can be exhibited in the different decomposition image level, and then the least information entropy is used as the criterion to select the optimal salient map out; At the second stage, the Gabor wavelets (GW) algorithm is utilized to suppress the high frequency noise remained in the optimal salient map and match the feature of size and direction of small target at different scales and angles, and next, to ensure the robustness of the target detection, Non-negative Matrix Factorization (NMF) is applied to fuse all the GW multi-scale images into one optimal target image, which is the final output of the presented method. Experimental results show that, compared with the contrast method, the proposed algorithm has high SCRG and high correct target detection rate, and works well in different types of complex backgrounds.(3) An effective method of Phase Spectrum Scale-Space based on Lo gradient minimization theory to infrared small target detection is presented. In this method, a multi-layers Phase Spectrum Scale-Space (PSSS) is used first time to generate a series of saliency maps with different scales. By this means, target with different size can be highlighted and background clutter can be suppressed effectively. Secondly, a criterion to select the optimal salient map is defined by the theory of Least Information Entropy (LIE). With this criterion, the potential target regions could most likely be selected as the optimal salient map. Meanwhile, by the introduction of the adjustment coefficient which is equal to the reciprocal of LIE, the details of the target is effectively enhanced. Finally, the Lo gradient minimization method to image smoothing is analyzed, and it is applied as a final detect procedure to extract the target out. Experimental results indicate that, compared with the contrast algorithms, the presented method is superior in detection rate and false alarm rate. So, it is an efficient method to detect IR small target in complex background.