Study On Infrared And Visible Image Processing Based On Multi-scale NSCT Transform

Infrared technology has the advantages of strong anti-interference ability,strong environmental adaptability,good concealment and strong recognition ability.It is widely used in military investigation,medical diagnosis,industrial detection and resource exploration.Due to the complicated imaging environment and the limitation of the inherent characteristics of detectors,the infrared and visible images collected by the equipment have some defects,such as low contrast,blurred edge and low signal-to-noise ratio(SNR),which leads to the difficulty of the image details being detected by human eyes,which is bound to be very unfavorable to the subsequent recognition,tracking and feature extraction,and are the bottleneck of the further development and application of the related technology.Therefore,the necessary processing is needed to improve the overall contrast of the image,suppress the noise of the image and improve the signal-to-noise ratio.In this paper,the Multiscale geometric analysis tool is used to non-subsampled Contourlet transform(NSCT),and the image enhancement,denoising and fusion processing are studied deeply.The main research topics include:1.Multi-scale geometric analysis.The development of multiscale geometric analysis is studied,the wavelet transform,Contourlet transform and NSCT transform are studied,and the coefficients of Contourlet transform and NSCT transform are analyzed and simulated.The NSCT transformation with anisotropy is selected as the analytical tool of this paper,which lays a theoretical foundation for the processing of enhancement,denoising and fusion.2.Infrared and visible image enhancement.Aiming at the problems of low contrast and edge blurring in infrared and visible images,this paper presents an improved Retinex and fractional order differential enhancement algorithm.The NSCT transformation with good characteristics is introduced,because the Retinex algorithm can enhance the whole contour of the image well,the fractional order differential algorithm can enhance the detail of the edge and texture of the image,so it combines the advantages of the two and improves them,the low frequency is enhanced by improved Retinex theory,the high frequency is enhanced by the adaptive order fraction differential algorithm after the noise is isolated by the Bayes threshold.the algorithm proposed in this paper is compared and analyzed with other enhancement algorithms subjectively and objectively.3.Infrared image denoising.Aiming at the problems of low signal-to-noise ratio and poor visual effect caused by noise interference in infrared image acquisition,this paper analyzes the common threshold denoising algorithms,and proposes an adaptive threshold denoising algorithm based on NSCT transformation.Under the background of NSCT transformation,the selection of threshold is optimized,and a new threshold processing function is constructed,and the influence factor is automatically updated by the subband coefficient,thus the denoising image is adaptively processed.After processing,the peak signal-to-noise ratio(PSNR)of the image is significantly improved,which is superior to non-local means(NLM)and block matching 3 dimension(BM3D)algorithms.4.Infrared and visible image fusion.In view of the limitations and differences of infrared and visible images,this paper presents an infrared and visible image fusion based on fuzzy set theory and neighborhood characteristics,combining with the excellent characteristics of NSCT.The modified fuzzy membership function is matched by k-means,and the enhanced preprocessing is carried out,then the combination of gradient and gray weights,the improved region energy weighted average fusion scheme is adopted at high and low frequency respectively,and the fused image is generated by inverse transformation.The fusion effect of this algorithm and other algorithms is analyzed by subjective human eye,and with the help of information entropy(IE),average gradient(AG),space frequency(SF)and structural similarity(SSIM)for objective analysis,which further reflects the advanced nature and effectiveness of this algorithm.