Research On Infrared Image Quality Enhancement And Small Target Detection Algorithms

The infrared(IR)imaging system can be widely applied in military and civilian field,and has become one of most focus area of national researchers.The inherent imaging mechanism makes the infrared image have a poor imaging effect with low local contrast and blur edges.Owing to the limited production capacity at modern stage and the weight and volume requirements for certain application occasion,the infrared imaging system suffers kinds of noises which cause significant degradation of image quality.It makes the IR system difficult to monitor the environment and the target.The current denoising and enhancement method display a poor capacity for the people's demand for high quality images.Therefore,it exhibits important scientific significance to study the infrared imaging mechanism and improve the quality of the infrared image.The infrared small target detection has been a hot topic of IR system for long time.The most small target detection algorithms are based on the contrast between the target and the surrounding local area.However,the real world are composed of kinds of substance,the thermal radiation image contains many edge clutter which also produce local contrast,causing much false alarms in the detection result and degrading the performance of IR system.Therefore,how to distinguish the contrast are generated by the edge or the small target will bring great significance to IR small target detection.In this dissertation,we analyze the characteristics of the IR image and study some key issues of IR image preprocessing.We also study the small target detection problem when the high quality image is obtained.The main contributions of this dissertation are as follows:A novel nonuniformity correction(NUC)framework based on a guided strategy is proposed.The proposed method utilizes a noise model image to guide the NUC procedure.As the weight of the noise model image varies as time passing,we propose three kinds of guide total variation NUC methods.A motion detection factor is introduced to stop the noise parameters' updating when the motion stops.The experimental results demonstrate that the proposed framework can fast converge with less artifacts introduced.In order to reduce the stripe noise in the IR remote sensing image,a destriping method based on region separation and double sparse regulation unidirectional variation(UV)model is proposed.By studing the properties of the stripe noise,we found that the stripe noise has sparse property in both the special domain and the gradient domain.Therefore,we combine the UV model with the two sparsity regulations.In addition,as the various noise properties exist in distinct area for an image,a divide-and-conquer strategy is proposed.We present an adaptive region segment method,and set various weights for different areas based on the segment result.Plenty of experimental results on both synthetic and real remote sensing images demonstrate that the proposed model has perfect destriping performance in terms of preservation of small(stripe)details,artifacts' reduction and improve the ability of stripe noise estimation.In order to enhance the detail of the 14 bit or 16 bit high dynamic range(HDR)IR image,this paper proposes a detail enhancement method based on local edge preservation filter method(LEPF).The LEPF can adaptively estimate the local area's edge.The proposed method adjusts the image's contrast in both global and local view,and stretching the details based on human view system(HVS)principle.The experimental results on many IR dataset demonstrate that the proposed enhancement method can provide a significant performance for HDR IR image with less artifacts and a natural look.For the long-range infrared small target detection with complex backgrounds,the scene's edge can seriously influence the detection performance.In order to reduce this phenomenon,this paper proposes a patch similarity propagated based background estimation model.This background estimation model propagates the patch similarity from near to far,and can measure the geodesic distance and the photometric distance between pixels along the path.The model can aggregates the similar pixel together better,and preserves more edge structure in the estimated background image.Thus the target image will contain less edge clutter.The experimental results demonstrate that the proposed method can effectively suppress the interference of the edge,reduce the false alarm rate of detection and improve the detection rate of infrared small target detection.