High dynamic range and super-resolution from raw image bursts

Abstract

Photographs captured by smartphones and mid-range cameras have limited spatial resolution and dynamic range, with noisy response in underexposed regions and color artefacts in saturated areas. This paper introduces the first approach (to the best of our knowledge) to the reconstruction of highresolution, high-dynamic range color images from raw photographic bursts captured by a handheld camera with exposure bracketing. This method uses a physically-accurate model of image formation to combine an iterative optimization algorithm for solving the corresponding inverse problem with a learned image representation for robust alignment and a learned natural image prior. The proposed algorithm is fast, with low memory requirements compared to state-of-the-art learning-based approaches to image restoration, and features that are learned end to end from synthetic yet realistic data. Extensive experiments demonstrate its excellent performance with super-resolution factors of up to ×4 on real photographs taken in the wild with hand-held cameras, and high robustness to low-light conditions, noise, camera shake, and moderate object motion.

Exploiting Light Polarization for Deep HDR Imaging from a Single Exposure

“…After flow estimation, the FlexHDR model addresses uncertainties caused by exposure and alignment via an attention network, and the final HDR is generated by a multi-stage fusion-based merging network. Other learning-based methods can be found at [ 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 ].…”

Breaking down Polyblur: Fast Blind Correction of Small Anisotropic Blurs

“…It may be used to enhance the images obtained with a super-resolution (SR) algorithm. We improve the contrast of two examples obtained with the technique of Lecouat et al [11], achieving better results than a typical sharpening filter designed to boost the mid-frequencies with a Gaussian mask (we used the implementation of skimage.filters.unsharp mask). We show these images in Figure 11 for a SR result provided by the authors of [11], prior to any sharpening.…”

“…We improve the contrast of two examples obtained with the technique of Lecouat et al [11], achieving better results than a typical sharpening filter designed to boost the mid-frequencies with a Gaussian mask (we used the implementation of skimage.filters.unsharp mask). We show these images in Figure 11 for a SR result provided by the authors of [11], prior to any sharpening. The version predicted by Polyblur is much sharper than the one obtained with the typical sharpening technique.…”

“…The blurry images in Figures 9 and 10 come from the supplemental materials of Schuler et al [14,15]. The SR image in Figure 11 is courtesy of Lecouat et al [11].…”

“…The functions GaussianBlurEstimation and Deconvolution are detailed in Section 3. Section 4 presents qualitative results for defocus and lens blur correction, and for sharpening of images predicted with a super-resolution (SR) algorithm, for instance in [11].…”