Hybrid Refractive-Diffractive Lens with Reduced Chromatic and Geometric Aberrations and Learned Image Reconstruction

Abstract

In this paper, we present a hybrid refractive-diffractive lens that, when paired with a deep neural network-based image reconstruction, produces high-quality, real-world images with minimal artifacts, reaching a PSNR of 28 dB on the test set. Our diffractive element compensates for the off-axis aberrations of a single refractive element and has reduced chromatic aberrations across the visible light spectrum. We also describe our training set augmentation and novel quality criteria called “false edge level” (FEL), which validates that the neural network produces visually appealing images without artifacts under a wide range of ISO and exposure settings. Our quality criteria (FEL) enabled us to include real scene images without a corresponding ground truth in the training process.

Achromatic Imaging Systems with Flat Lenses Enabled by Deep Learning

“…Therefore, we have decided to adopt a more challenging evaluation criterion by comparing our results with those achieved by applying deblurring algorithms for chromatic aberrations caused by other lenses with significantly lower chromatic aberrations. In refs , , the U-net architecture was used on a small data set of images captured from a high-resolution monitor using a Hybrid refractive–diffractive lens system. Although the chromatic aberrations in a hybrid lens system are much smaller than in our case, the reported PSNR is 27.71, significantly lower than our result.…”

“…Chromatic aberration caused by single-ball lenses has been corrected to some extent by an image processing algorithm . With the rapid advancement of machine learning, deep-learning models have demonstrated superior performance compared to traditional image restoration methods. − Recent works show a step forward toward facing flat lens challenges by applying algorithms to reduce chromatic aberrations of a hybrid diffractive–refractive imaging system using images that were captured indoors in laboratory conditions. ,, Other works combined computational imaging with advanced meta-optics designs , and were able to reduce chromatic aberrations in the imaging of miniature samples under laboratory conditions. Tseng et al used a deep-learning approach to jointly optimize a metasurface design and its corresponding deconvolution algorithm based on point-spread-function (PSF) estimation and simulated images.…”

Achromatic Single Metalens Imaging via Deep Neural Network

“…Due to significant material dispersion and dispersive responses of metasurfaces, different spectral components passing through metalenses will focus on disparate spatial planes, creating chromatic aberrations and negatively impacting image quality. − The resolution and transmission mismatch between color channels also leads to reduced color contrast and inaccurate color balancing, significantly degrading color-reproduction capabilities and the signal-to-noise ratio (SNR) of full-color imaging. Existing strategies to mitigate chromatic aberration include cascaded multilayer metalenses, − interleaving meta-atoms for different wavelengths, − metalens arrays, dispersion correction phase mask, ,− increased focusing depth, and computational optimization and correction of phase profiles. , But these approaches increase system complexity while sacrificing other performance metrics such as scalable high-yield fabrication, imaging quality at center wavelength, and freedom of material choices. Consequently, a single meta-lens solution capable of color-accurate aberration-free imaging under diverse operating conditions remains elusive.…”

Deep Learning Enables Optofluidic Zoom System with Large Zoom Ratio and High Imaging Resolution

“…In general, the electrowetting liquid lens can be used to correct aberration with its variable optical power, but the limited optical power range affects the ability for correcting aberration. In addition to optimizing the optical structure, some image-processing algorithms were proposed to improve the image quality [ 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 ]. For instance, a real-time digital correction method for transverse chromatic aberration was presented, which can be adapted to multi-resolution foveated laparoscope systems [ 14 ].…”

“…A hybrid refractive-diffractive lens was proposed to be paired with a deep neural-network-based image reconstruction. In addition, it produces high-quality, real-world images with minimal artifacts, reaching a PSNR of 28 dB on the test set [ 19 ]. Moreover, an approach based on a generative adversarial network was proposed to simultaneously denoise and super-resolve OCT images [ 15 ].…”