Research On Multi-focus Image Fusion Algorithm Based On NSST And SPCNN

Optical devices are often limited by the depth of field of the environment,resulting in blurred regions of the image.In order to obtain the global information of the scene and improve the information capacity of the image,multi-focus can fuse different focused images into a clear image of the whole scene so as to further image analysis and processing.In recent years,multifocus image fusion has been widely used in multi-source remote sensing fusion,computer vision,artificial intelligence and medical detection.In this paper,non-subsample Shearlet Transform(NSST)is used to efficiently capture the geometric features of the image,aiming at the problems of misjudgment of focusing region,unclear texture,discontinuity of contour and artifacts in multi-focus fusion.The Lab color space is used to segment the correlation between brightness and chromaticity to reduce the complexity of multi-dimensional image fusion algorithm and suppress the block effect.This paper proposes two fusion strategies.(1)In the high-frequency fusion process,there are two problems with the SPCNN fusion method.On the one hand,the focus region is misselected.On the other hand,the feature information is not ideal.An improved two-channel SPCNN fusion scheme based on NSST is proposed.Firstly,NSST is used to fuse and reconstruct the brightness channel.The improved SPCNN model is used to construct fusion rules in high frequency fusion.In the improved twochannel SPCNN model,the pulse-emitter part is removed and parameter setting is reduced.The high-frequency fusion weight is constructed by adopting the principle of maximizing the absolute value of two-channel.Then,the 、 parameters is set by introducing the region contrast sensitive function(CSF)in accordance with human visual characteristics to improve the extraction effect of significance information.Finally,the gamut channels(a and b)use energy-matching-fusion method.Experimental results show that the proposed algorithm performs well in definition,information richness,spatial information,contrast and visual fidelity.(2)Aiming at the problems of misjudgment of focusing region,unclear texture,discontinuous contour and artifacts in multi-focus fusion,an NSST fusion method combining superpixel and guided wave is proposed.Firstly,NSST is used to fuse and reconstruct the brightness channel.The high-frequency subband can correctly identify the focus region by large-scale filtering,and retain the texture information of edge by small-scale filtering.The high-frequency fusion decision graph is constructed by multiple guided filtering.After the fusion of high-frequency subbands,the anti-noise performance of the image is enhanced by median filtering.Secondly,in the low-frequency part,superpixels are used to extract salient information and retain the texture and contour of the inner region of the superpixel to ensure the continuity of the texture contour.After the fusion,the low-frequency subband is corrected by gamma curve to improve the image contrast.Finally,the coupling characteristics of SPCNN are utilized to achieve the consistency of the gamut channel region.Experimental results show that the fusion image has obvious contrast enhancement,clear texture and clear boundary.The algorithm has good ability to judge the focusing region,and can avoid the phenomenon of contour discontinuity and artifact.