UAV Infrared And Visible Image Fusion Algorithm Research

UAV plays an important role in military and civilian fields. With carryingdifferent imaging sensors, UAV does not only adapt to the change of environmentalconditions, but also obtain more plentiful spatial information. Because of differentimaging principle, the acquired multi-sensor images have different characteristics,and can’t get the full scene information. UAV multi-sensor image fusion uses differentimages about the same scene, which obtained by many different imaging sensors, andintegrates various complementary information to get fused image with morecomprehensive, more intuitive, more reliable information，and lays the foundation forfurther target identification and tracking.Before image fusion of two or more multi-sensor images based on the samescene, UAV imaging sensors are restricted by imaging time, shooting angle and so on,so the acquired images may not reach alignment in spatial position, thus imageregistration is carried out before multi-sensor image fusion.The paper mainly makes research on image registration and fusion of infraredimage and visible image, and combines complementary information to obtain perfectfused image. The main contributions are summarized as follows:1) According to the characteristics of the UAV infrared and visible images on thechange of spatial location, the paper analyses image registration algorithm basedon mutual information theory, and proposes new registration method based ongradient mutual information and uniform designed PSO. The method usesuniform design to improve the space distribution of initial particles, so that theinitial particles can distribute uniformly in the search space, and reduce thepossibility of trapping local optimum value, and improve the registrationprecision and speed, and can obtain better image registration parameters.2) For the image fusion of UAV infrared image and visible image, a new algorithmcombined Nonsubsample Contourlet transform (NSCT) and IndependentComponent Analysis (ICA) has been proposed. The algorithm carries out independent component transformation on the high-frequency subbandcoefficient, which achieved by NSCT, so as to eliminate the correlation betweeneach subband coefficient. While an adaptive weighted fusion rule is designedfor the transformed independent component coefficients, the experimental resultdenotes that the fused image retains the structural features of the infrared image,and contains rich detail information of visible image. Through the analysis ofsubjective and objective evaluation standard, the fused image has the expectedeffect.