Research On Image Fusion And Image Translation Algorithms Under Different Lighting Environments

Illumination always plays a vital role in the image acquisition process.In order to research the impact of illumination on image information,this thesis uses image fusion and image translation algorithms to learn the impact of illumination from intensity and direction.In the field of image fusion,multi-exposure image fusion is selected as the entry point,and in the image translation algorithm based on generative adversarial networks,two experiments are designed for the sequential verification of the conversion of multiple light source images to simple light source images.The main research work of the thesis includes the following two aspects:1.In order to research the influence of light intensity on image information,a multi-exposure image fusion algorithm based on independent component analysis is proposed.Aiming at the problem that some detail information cannot be effectively retained and color distortion in multi-exposure image fusion,a multi-exposure image fusion method combining signal decomposition is proposed.The key to multi-exposure images is to fuse the luminance channels,so different fusion methods are used for the luminance and chrominance channels.In the luminance channel fusion,as much as possible details are extracted by adding independent component analysis to decompose the signal,and then combined with the Hybrid HDR method to keep it in the resulting image.Experimental results show that the proposed method can improve the overall quality of the final fusion result,and in some scenes,it has more outstanding detail retention ability than other existing methods,while it can still maintain the color of the original exposure image.information.2.In order to research the influence of light direction on image information,two different image translation algorithms and data sets were used for experiments.In the experiment of multi-light source image conversion to single light source,the image translation algorithm based on the detection and repair of the deformation area was first applied to the analysis of the impact on lighting.Through the continuous refinement of the local block with the worst image quality using the generation network,the overall quality of the image is improved.The experimental results show that the overall characteristics of the single light source image converted from the multi-light source image are basically consistent with the real image.In the experiment of conversion ofcomplex natural light images to simple light sources,the improved Sim GAN algorithm is used to reduce the effect of complex natural light in the line-of-sight estimation data set.In order to ensure the integrity of key information such as line of sight and improve the quality of the generated image,an encoder-decoder network is used to replace the original generator,and a relative discriminator network is used to replace the original discriminator.The experimental results further reduce the shadow area and noise interference in the original human eye image,remove part of the interference information for subsequent line-of-sight estimation,and verify that the image conversion algorithm can effectively reduce the impact of irregular lighting.