Research On The Fusion Algorithm Of Infrared Image And Visible Image Based On Multi-scale Transform

The fusion of infrared image and visible image is an important branch of heterogeneous multi-sensor image fusion technology.Because of the redundancy and complementarity in infrared and visible image,image fusion can accurately describe the object by effectively combining the target characteristics in infrared image and the details of the scene in visible image,there are important application values in scene monitoring,battlefield reconnaissance and resources detection,etc.It has been widely applied in many aspects of environmental monitoring in recent years.In this paper,the fusion technology of infrared image and visible image based on multi-scale transform have been discussed,the common multi-scale transform algorithm,including the pyramid transform,wavelet transform,non-subsampled contourlet transform and non-subsampled shearlet transform have also been analyzed.On this basis,according to the imaging characteristics of thermal plume's infrared image and visible image,this paper deeply discusses the multi-scale decomposition and reconstruction of image in fusion process,and the evaluation,selection and improvement of the high frequency subband image's fusion rules,the concrete work is shown as follows:(1)A rapid image fusion method based on non-subsampled contourlet transform is presented to solve the problem that the recursive filtering impacts the efficiency of the multi-scale decomposition and reconstruction seriously.In the improved method,all filters in the non-subsampled pyramid filter bank and non-subsampled directional filter bank are convoluted based on the scales and direction levels of the decomposition,source image is filtered with convoluted result to acquire the sub-band images directly.The low frequency coefficients are fused by using the rule based on the regional energy;the high frequency coefficients are fused based on gradient feature,which could balance the weight of high frequency coefficient adaptively by the high-frequency subband image's gradient map.(2)A rapid image decomposition and reconstruction method based on non-subsampled shearlet transform is presented to further improve the efficiency of the multi-scale decomposition and reconstruction,source image is filtered with convoluted result to acquire the sub-band images directly.On this basis,according to the infrared image and visible image of thermal plume,the fusion rules are presented based on the edge properties.The low frequency coefficients are fused by using the rule based on the standard deviation of area feature measure.The highest frequency coefficients are fused by using the maximum energy selection rule;the other high frequency coefficients are fused by combining sum modified-laplacian(SML)and regional standard deviation with gradient features,which emphasizes the edge information of the target plume.(3)In order to highlight the temperature information of the target plume in the experiment,combining with the target extraction,the color fusion method of visible image and infrared image is realized based on the improved fusion algorithm,the infrared image is segmented by setting the threshold,and the color information of visible image is passed to the background region of color fused image,and the target region of color fused image obtain color information though pseudo color transform.The results can be used to obtain thermal plume's space distribution information and the atmospheric transfer,diffusion and deposition process of pollutants.It also can improve the recognition ability for the thermal plume and interpretation ability for the environment,and provide certain scientific basis for comprehensive control of environment.