Noise Reduction Method for Spectral Holographic Reconstructed Images Based on Noise Discreteness
Objective As high-resolution image sensors and computer technology develop,significant applications for holography have emerged in the fields of three-dimensional imaging and display,optical information processing,and intelligent optical computing.However,numerous challenges remain unresolved in both digital holography and computational holography.During the recording process of digital holography,as a kind of multiplicative noise,speckle noise becomes a prominent problem and its removal is more challenging than that of additive noise,drastically compromising the quality of the reconstructed image.Consequently,noise reduction in holograms and reconstructed images is of particular urgency.Current noise reduction methods are primarily categorized into two categories:optical-based methods and digital-based methods.In the optical-based methods,one limitation involves the cumbersome recording process of multiple holograms with speckle diversity by multiple mechanical motions,which could lower the system stability.In the digital-based methods,complex algorithm commonly reaches better noise reduction effects.However,the cost of increased processing time may impede the real-time capability of the system.Therefore,an integrated approach combining optical and digital methods can maximize noise reduction while maintaining a focus on speed.Therefore,we combine Fourier transform spectroscopy technology and digital holography technology to gain speckle diversity holograms.Then,utilizing the noise difference analysis of decrypted images at several wavelengths,we propose a weighted summation average(WSA)noise reduction method,in combination with the block matching 3D(BM3D)algorithm.As a result,the optimized effect of noise reduction can be achieved.Methods First,we calculate the normalized monochromatic peak signal-to-noise ratio(M-PSNR)between the reconstructed images and take the original images as a representation of the noise intensity level of the reconstructed images,which are used as an initial weighting factor.Subsequently,within a given spectral range,the wavelength centers of the three RGB components are selected respectively according to the CIE international standard.A uniform interval radius is selected for the three RGB components,and a binary weighting factor is applied to weight the selected wavelength intervals,achieving the waveband optimization.Finally,the BM3D algorithm is combined with the WSA algorithm to further reduce the noise,with the sequence of utilization also being analyzed to achieve the optimum denoising effect.Results and Discussions To verify the feasibility of the algorithm,we take the decryption process of the proposed optical cryptosystem as the testbed(Fig.1).Under specific conditions,89 single-wavelength reconstructed images,spanning a wavelength range of 449-801 nm with a 4 nm wavelength interval are processed to analyze the noise reduction effect.First,the normalized M-PSNR between the reconstructed images and the original images is calculated to be used as the weighting factor(Fig.4).The suboptimal denoising effect of this direct method is analyzed by examining the deviation of the average intensity ratio of the three RGB components from the ground truth.Hence,it is imperative to select the waveband that is closer to the true average intensity value.Second,according to the CIE international standard,initial wavelength centers of three RGB components(633 nm,553 nm,and 453 nm)are selected.When the interval radius of 25 is chosen,the selected intervals for RGB components approximately encompass the entire waveband(Fig.7),and optimization is performed to identify the optimal wavelength center of the three RGB components(621 nm,549 nm,and 449 nm).Third,we perform a comparative analysis considering the symmetry of the intervals,the uniformity of the selected interval radius for the three RGB components,and the adopted weighting method,aiming to maximize the color-PSNR(C-PSNR)value between the noise reduction result and the original color image(Fig.9).As a result,the C-PSNR reaches to 78.59 dB when the abovementioned three parameters are determined in which a symmetric interval radius of 26 for the three RGB components and a binary weight factor for weighting is utilized.Fourth,the noise reduction result is compared with that achieved by the classical color BM3D algorithm which reaches 79.15 dB.In comparison,the WSA algorithm demonstrates a faster noise reduction speed(0.75 s vs 4.14 s),while the C-PSNR value obtained by the CBM3D algorithm is comparatively larger(79.15 dB vs 78.59 dB).Considering both the noise reduction effect and processing time,we combine the two algorithms to analyze the order in which the two algorithms are used,and choose the method with the best denoising effect.Specifically,the images denoised by the WSA algorithm should be further denoised by the CBM3D algorithm to obtain the final color-denoised image.Following this sequence of the two algorithms,the C-PSNR between the final denoised image and the original color image reaches 91.11 dB.Conclusions Based on the combination of Fourier transform spectroscopy technology and digital holography technology,we propose a method for noise reduction that makes full use of the noise diversity cross all reconstructed images at varying wavelengths with hyperspectral resolution.Our WSA algorithm analyzes the difference in the noise intensity level of several wavelengths to determine the center and interval radius of the three RGB components to optimize the waveband and reduce noise.The BM3D algorithm is further applied to reduce the noise.Numerical simulation and experimental results indicate that a maximum C-PSNR value of 91.11 dB is attainable by reasonably employing the WSA and BM3D algorithms.Our composite algorithm can effectively reduce speckle noise based on the optimal selection of optical waveband and weighted factors.This method provides new insights for the noise reduction of color digital holography.
NETL
NSTL
万方数据
