Research On Infrared Small Target Detection And Segmentation Method Based On Complex Background

The small target detection is a crucial technique of infrared image system and widely used in surveillance,reconnaissance,early warning,and etc.However,due to the optics point spread function of the thermal imaging system and the long distance imagery,the targets always appear as small and dim shape with little texture feature.Besides,the heavy clutters enhance the complexity of background and targets always suffer from low signal-to-clutter ratio(SCRG).As a consequence,small target detection under poor quality images still remains an open problem.In this thesis,our work mainly focuses on the detection and segmentation of weak and small infrared targets in complex backgrounds.The main tasks are as follows:(1)In past decades,many previous methods concentrated on traditional top-hat transformation which relyies on the hand-crafted shape and value of structural elements.However,these methods are inevitably challenged by two aspects: 1)The structural elements can not suppress the heavy clutters because the construction of it is always according to the prior informmation of target and unable to consider the feature of clutters.2)The simple structural element is hardly to extract sufficient local feature information adaptively for background suppression.Aim to these to two aspects,we propose corresponded algorithm in this thesis.It is noticeable that small target usually emerges as a small bright region in an infrared image.And the adjacent pixels of small targets are usually dependent on each other.Moreover,the feature of the background is always stable.Based on these situations,we propose adaptive top-hat transformation based on M-filter and guided filter to enhance target and suppress background respectively in this thesis.(2)The small targets usually have the salient feature when compared to background,but the background is usually relatively stable.And the target and the surrounding background can show a large local difference.Therefore,in order to make full use of the local features to segment the target,we propose a new graph cut segmentation method based on the tensor field.First,we reconstruct the traditional graph cut problem into a special non-convex optimization problem through the maximum flow model and extend it to hypergraph model.Then,we combine the problem with the tensor field based on directional bilateral filter thereby constructed a segmentation model for small targets.Finally,we propose a proximal minimization algorithm to solve this non-convex optimization problem effectively and achieve global convergence.Subsequent experiments and result of test further illustrate the superiority of the method.