GamutNet: Restoring Wide-Gamut Colors for Camera-Captured Images

Abstract

Most cameras still encode images in the small-gamut sRGB color space. The reliance on sRGB is disappointing as modern display hardware and image-editing software are capable of using wider-gamut color spaces. Converting a small-gamut image to a wider-gamut is a challenging problem. Many devices and software use colorimetric strategies that map colors from the small gamut to their equivalent colors in the wider gamut. This colorimetric approach avoids visual changes in the image but leaves much of the target wide-gamut space unused. Noncolorimetric approaches stretch or expand the small-gamut colors to enhance image colors while risking color distortions. We take a unique approach to gamut expansion by treating it as a restoration problem. A key insight used in our approach is that cameras internally encode images in a wide-gamut color space (i.e., ProPhoto) before compressing and clipping the colors to sRGB's smaller gamut. Based on this insight, we use a softwarebased camera ISP to generate a dataset of 5,000 image pairs of images encoded in both sRGB and ProPhoto. This dataset enables us to train a neural network to perform wide-gamut color restoration. Our deep-learning strategy achieves significant improvements over existing solutions and produces color-rich images with few to no visual artifacts.

A novel gamut expansion method based on combined global–local mapping for sRGB‐to‐ProPhoto conversion

“…Some DL-based GE methods have been proposed. Hoang et al [11] proposed a deep neural network (DNN) for GE, in which they first processed the input sRGB image in a clipped-ProPhoto version and the input clipped-ProPhoto images are predicted by the DNN. Takeuchi et al [20] proposed GE methods considering colour information restoration, in which they proposed two methods to realize the expansion of BT.709 gamut [21] to BT.2020 gamut [22].…”

“…After the linear mapping, we concatenate the out‐of‐gamut mask to $${I_{clipped}}$$. Based on [11]’s definition, in the input $${I_{sRGB}}$$, the pixels with one or more possibly saturated values in their three RGB channels as potentially out‐of‐gamut, and the other pixels including pure white and pure black colours are preserved during the colour space conversion. So, the out‐of‐gamut mask m can be calculated as follows: $$$$\begin{equation} m\ =\left\{ \def\eqcellsep{&}\begin{array}{c}0,\ \textit{where}\ {I}_{\textit{sRGB}}\ \textit{is}\ \textit{black},\ \textit{white},\ \textit{or}\ \textit{in}\ \textit{sRGB}\\ 1,\ \textit{potentially}\ \textit{out-of-gamut}\end{array} \right.\ \end{equation}$$$$We concatenate the out‐of‐gamut mask m with the $${I_{clipped}}$$ as network input $${I_{input}}$$.…”

“…After the linear mapping, we concatenate the out-of-gamut mask to I clipped . Based on [11]'s definition, in the input I sRGB , the pixels with one or more possibly saturated values in their three RGB channels as potentially out-of-gamut, and the other pixels including pure white and pure black colours are preserved during the colour space conversion. So, the out-of-gamut mask We concatenate the out-of-gamut mask m with the I clipped as network input I input .…”

“…The purpose of GE is similar to image restoration. According to the literature [1,11], GE is to restore the colour values of the input small gamut to a wider gamut space. Because many limited gamut images are converted by the original wider gamut images.…”

Learning a Practical SDR-to-HDRTV Up-conversion using New Dataset and Degradation Models

“…out-of-gamut (OOG) pixels will fall outside valid range after CST to a narrower volume. Dealing OOG is the key of gamut mapping, instead of softmapping [63,75] (used by [76]'s DM) preserving as much WCG information to narrow gamut, we use hard-clipping which clips all OOG pixels to BT.709 boundary (Fig. 5):…”