ExpandNet: A Deep Convolutional Neural Network for High Dynamic Range Expansion from Low Dynamic Range Content

Abstract

High dynamic range (HDR) imaging provides the capability of handling real world lighting as opposed to the traditional low dynamic range (LDR) which struggles to accurately represent images with higher dynamic range. However, most imaging content is still available only in LDR. This paper presents a method for generating HDR content from LDR content based on deep Convolutional Neural Networks (CNNs) termed ExpandNet. ExpandNet accepts LDR images as input and generates images with an expanded range in an end‐to‐end fashion. The model attempts to reconstruct missing information that was lost from the original signal due to quantization, clipping, tone mapping or gamma correction. The added information is reconstructed from learned features, as the network is trained in a supervised fashion using a dataset of HDR images. The approach is fully automatic and data driven; it does not require any heuristics or human expertise. ExpandNet uses a multiscale architecture which avoids the use of upsampling layers to improve image quality. The method performs well compared to expansion/inverse tone mapping operators quantitatively on multiple metrics, even for badly exposed inputs.

The Mertens Unrolled Network (MU-Net): A high dynamic range fusion neural network for through the windshield driver recognition

“…Many of them attempt to do this using a single shot 21,22 or by simply having a extreme number of weights. [21][22][23][24] For real-time, robust driver recognition, both of these are unacceptable. In the former case, we expect that a single shot will have a high probability of either misfiring or inaccurately capturing the face, due to the potential for obstructive glare.…”

Deep HDR Video from Sequences with Alternating Exposures

“…Over the past decades, many powerful approaches have been developed to produce still HDR images from sequences with different exposures [DM97, SKY∗12, HGPS13, OLTK15, MLY∗17, KR17, WXTT18], burst images [LYT∗14, HSG∗16], or a single LDR image [EKD∗17, EKM17, MBRHD18]. However, most of these approaches only demonstrate results for generating still HDR images and are not suitable for producing HDR videos [KSB∗13].…”

The high dynamic range imaging pipeline : Tone-mapping, distribution, and single-exposure reconstruction

“…While convincing results can be achieved, the method is limited to textured areas and it requires some manual interaction. More recently, a number of methods employ deep learning strategies for single-exposure HDR image reconstruction [85,151,176,276], including the method of Paper E [83]. The paper is discussed in Chapter 5 and related to the other deep learning reconstruction methods in Section 5.2.…”

“…Marnerides et al [176] proposed to use a multi-level CNN, which processes local and global information in separate branches. These methods are to some extent complementary to Paper E, as they consider the compound problem of transforming an LDR image to HDR.…”