| Compared with the limited dynamic range of a camera or display,the dynamic range of light intensity in a natural scene can easily span several orders of magnitude,so a high dynamic range image is needed to record this dynamic range.Usually,high dynamic range images are obtained by fusing multiple exposure images,and the fused high dynamic range images consistently exceed the display’s dynamic range.Therefore,a tone mapping algorithm is proposed to compress the intensity distribution of high dynamic range images.Generally,compression dynamic range means that the low-frequency component of an image is compressed while the high-frequency component is preserved.To effectively compress the dynamic range,this paper mainly does the following research work:(1)A multi-scale cross-decomposition tone mapping algorithm is proposed based on the analysis of image edge conditions.First,the chromaticity and brightness information are separated through the chromaticity brightness color space.Secondly,a hybrid crossdecomposition method based on Gaussian and Bilateral filters is proposed to decompose the brightness information into the base,detail,and edge layers.Then,the reconstruction coefficients of each scale layer are discussed in detail through an experimental case,and the selection rules of coefficients for tone mapping are formulated.Finally,the compressed and reconstructed brightness and chromaticity images are fused to complete the mapping process.A large number of subjective and objective experiments show that the algorithm can effectively compress the image’s dynamic range,effectively preserve the color information,and achieve the effect of enhancing details and highlighting borders.(2)To overcome the shortage of edge preservation ability and severe halo artifacts in the current hierarchical tone mapping algorithm,a total bending filter for solving gradients with a polygonal window is presented.A hierarchical tone mapping is proposed algorithm is constructed on this basis.The filter is constructed using the second fundamental type in differential geometry,combines the first derivative with the second derivative information using a multiplicative strategy to characterize piecewise linear images,and designs a multiwindow technique to solve gradient dispersion.In the design of the algorithm,a chromaticity brightness color space is used to effectively decompose the input high dynamic range image to maximize the color integrity.Through many subjective and objective experiments and data analysis,the advantages of the total bending tone mapping algorithm compared with the current advanced tone mapping algorithm are proved.In addition,a large number of qualitative experiments further prove the advancement of the total bending filter in image smoothing,image denoising,and image enhancement applications,highlighting the ability of the total bending filter to maintain edges.(3)Due to the limited dynamic range of digital imaging equipment,the images taken under low-light conditions often have problems such as insufficient brightness and severe noise.A channel splitting attention network for low-light image enhancement is proposed,which requires supervised dataset learning.The network first divides shallow features into two branches,residual branches and dense branches,to promote the transmission of different feature information.Residual branching facilitates the reuse of old features,while dense branching facilitates the exploration and mining of new features.In addition,the network uses merge-run mappings to facilitate information integration between different branches,and an attention block is designed to differentiate information characteristics between different branches.Many qualitative and quantitative experiments have proved that the proposed network is superior to the most advanced methods in all indicators.It can suppress color difference and noise while enhancing the low-light image.This paper has 33 figures,13 tables,and 86 references. |
