《Automatic Exposure Correction of Consumer Photographs 》 翻译

Automatic Exposure Correction of Consumer Photographs

Abstract

We study the problem of automatically correcting the exposure of an input image. Generic auto-exposure correction methods usually fail in individual over-/under-exposed regions. Interactive corrections may fix this issue, but adjusting every photograph requires skill and time. This paper will automate the interactive correction technique by estimating the image specific S-shaped nonlinear tone curve that best fits the input image. Our first contribution is a new Zone-based region-level optimal exposure evaluation, which would consider both the visibility of individual regions and relative contrast between regions. Then a detail-preserving S-curve adjustment is applied based on the optimal exposure to obtain the final output. We show that our approach enables better corrections comparing with popular image editing tools and other automatic methods.

本文研究了输入图像的自动曝光校正问题。一般的自动曝光校正方法通常在个别曝光过度/曝光不足的区域失效。交互式矫正可以解决这个问题，但是调整每张照片需要技巧和时间。本文将通过估计最适合输入图像的特定于图像的S形非线性色调曲线来自动化交互校正技术。我们的第一个贡献是一个新的基于区域的区域级最佳曝光评估，它将同时考虑单个区域的可见性和区域之间的相对对比度。然后在最佳曝光的基础上进行细节保持S曲线调整，得到最终输出。结果表明，与常用的图像编辑工具和其他自动方法相比，该方法具有更好的校正效果。

1 Introduction

Exposure is one of the most important factors of determining the quality of a photograph. In over-exposed or under-exposed regions, details are lost, and colors are washed out. Despite that sophisticated metering techniques have been equipped on the cameras, taking well-exposed photos remains a challenge for normal users. There are several reasons: 1) the camera’s metering (e.g., spot, center-weighted, average, or multi-zone metering) is not perfect. If the metering points/areas are not targeting the subject or there are multiple subjects, the metering may fail. Fig. 1(a) is a failure case caused by the backlit; 2) the assumption that the mid-tone of the subject is gray is sometimes invalid due to the complex reflectance of the world (e.g., a snow-white rabbit is often captured as an undesired grayish rabbit without exposure compensation); 3) in-camera post-processing capability is limited, especially for the low-end cameras.

曝光是决定照片质量的最重要因素之一。在过度曝光或曝光不足的区域，细节会丢失，颜色会被冲掉。尽管相机配备了精密的测光技术，但拍摄曝光良好的照片对普通用户来说仍然是一个挑战。有几个原因：1）相机的测光（例如点测光、中心加权测光、平均测光或多区域测光）不完美。如果测光点/区域没有针对被摄体或有多个被摄体，测光可能会失败。图1（a）是由背光引起的故障情况；2） 由于世界的复杂反射，被摄体的中间色调为灰色的假设有时是无效的（例如，雪白兔常常被捕获为不需要曝光补偿的灰色兔子）；3） 相机内后处理能力有限，尤其是低端相机。

To address this issue, some automatic methods like auto-level stretch [1] and histogram equalization [1] have been proposed to correct the exposure. For example, autolevel stretch linearly maps the brightness to the maximum tonal range (e.g., [0, 255]). This method, however, only uses the statistics of the whole image, without considering each image region individually. For the backlit case in Fig. 1, auto-level stretch does not take effect (see Fig. 1 (b)) since the image histogram has reached the maximum tonal range (top-left of Fig. 1(a)). Histogram equalization [1] (and its variations [2]) better distributes the intensity values over the histogram. Unfortunately, it would produce unrealistic effects in photographs (see Fig. 1(c)).

为了解决这个问题，一些自动的方法如自动水平拉伸[1]和直方图均衡化[1]已经被提出来校正曝光。例如，自动水平拉伸将亮度线性映射到最大色调范围（例如，[0，255]）。然而，这种方法只使用了整个图像的统计信息，而没有单独考虑每个图像区域。对于图1中的背光情况，由于图像直方图已经达到最大色调范围（图1（a）的左上角），自动电平拉伸不起作用（参见图1（b））。直方图均衡化[1]（及其变化[2]）更好地将强度值分布在直方图上。不幸的是，它会在照片中产生不切实际的效果（见图1（c））。

If user assistance is allowed, the interactive correction method is more effective. Forinstance, most photo editing software allow the user to manually adjust a non-linear tone curve [3] (e.g., S-curve) to correct the dark/mid-tone/bright regions separately. Fig. 1(d) is the assisted result by expert. But the best shape of the curve varies a lot from image to image. Touching up every single image is impractical for typical consumers.

如果允许用户协助，交互式校正方法更有效。例如，大多数照片编辑软件允许用户手动调整非线性色调曲线[3]（例如，S曲线）以分别校正暗/中色调/亮区域。图1（d）是专家的辅助结果。但曲线的最佳形状因图像而异。对典型的消费者来说，润色每一张图片都是不切实际的。

In this paper, we present an automatic exposure correction method that can estimate the best image specific non-linear tone curve (the S-curve in our case) for a given image. Unlike [4], we need no training data. Note that it is a non-trivial task since the variation of input consumer photographs is so large. The key to the success of an automatic correction is to know what the best exposure should be for every image region.

本文提出了一种自动曝光校正方法，它可以估计给定图像的最佳图像特定非线性色调曲线（本例中的S曲线）。与[4]不同，我们不需要培训数据。注意，这是一个非琐碎的任务，因为输入的消费者照片的变化如此之大。自动校正成功的关键是知道每个图像区域的最佳曝光应该是什么。

To address this fundamental issue, we borrow the concept of “Zone” from the welldeveloped Zone System [5] in photography. The Zone system quantizes the whole exposure range as eleven discrete zones. We formulate the exposure correction as a zone estimation problem - we optimize a global objective function to estimate the desired zone in each image region by simultaneously considering two goals: maximizing the local details in each region, and preserving the relative contrast between regions.

为了解决这个基本问题，我们借用了摄影中发达的区域系统[5]中的“区域”概念。分区系统将整个曝光范围量化为11个离散分区。我们将曝光校正描述为一个区域估计问题，我们同时考虑两个目标：最大化每个区域的局部细节和保持区域之间的相对对比度，优化一个全局目标函数来估计每个图像区域中所需的区域。

After getting the estimated zone of every region, we propose a new non-linear curvebased correction algorithm called detail-preserving S-curve adjustment, to push each region to its desired zone, as much as possible. Compared with generic S-curve adjustment [6][7][8], our detail-preserving S-curve adjustment can maintain local details and avoid halo effects. Fig. 1(e) shows our estimated curve and final corrected result.

在得到每个区域的估计区域后，我们提出了一种新的基于曲线的非线性校正算法，称为保细节S曲线平差，以尽可能地将每个区域推到期望区域。与一般的S曲线平差[6][7][8]相比，我们的细节保持S曲线平差可以保持局部细节，避免光晕效应。图1（e）显示了我们的估计曲线和最终校正结果。

Like most automatic approaches, our approach does not address the user preference issue [9]. The “correct” exposure may be defined as the one that achieves the effect the photographer intended. However, our user studies show that an automatic correction still benefits most typical consumers - especially for their daily photos processing. We also show our new exposure optimization provides significant visual quality improvement over pervious work. Since our correction is simple and robust, it can be chosen as a better alternate in photo editing tools and a built-in camera component.

与大多数自动方法一样，我们的方法没有解决用户偏好问题[9]。“正确”曝光可以定义为达到摄影师预期效果的曝光。然而，我们的用户研究表明，自动更正仍然有利于大多数典型的消费者-尤其是他们的日常照片处理。我们还表明，我们的新的曝光优化提供了显着的视觉质量改善透水工作。由于我们的校正是简单和稳健的，它可以选择作为一个更好的替代在照片编辑工具和内置相机组件。

图1 一张典型的曝光不足的照片。在（a）的左上角，我们显示了输入图像的亮度直方图，该直方图具有最大的色调范围以及阴影和高光中的峰值。

2 Related Work

Automatic exposure control is one of the most essential research issues for camera manufacturers. The majority of developed techniques are hardware-based. Representative work include HP “Adaptive Lighting” technology [10] , Nikon “D-Lighting” technology [11]. These methods compress the luminance range of images by a known tone mapping curve (e.g., Log curve) and further avoid local contrast distortion by “Retinex” processing [12]. Specific hardware has been designed to perform per-pixel exposure control [13] or scene-based (e.g., backlit, frontlit [14] or face [15]) exposure control. Some automatic techniques (e.g. [16]) are proposed to estimate the optimal exposure parameters (shutter speed and aperture) during taking photos.

自动曝光控制是相机制造商最重要的研究课题之一。大多数已开发的技术都是基于硬件的。代表作品包括惠普“自适应照明”技术[10]，尼康“D照明”技术[11]。这些方法通过已知的色调映射曲线（例如，对数曲线）压缩图像的亮度范围，并通过“Retinex”处理进一步避免局部对比度失真[12]。已经设计了特定硬件来执行每像素曝光控制[13]或基于场景（例如，背光、前照[14]或面部[15]）的曝光控制。一些自动技术（如[16]）被提出来估计拍摄过程中的最佳曝光参数（快门速度和光圈）。

There are numerous techniques about software-based exposure adjustment, including most popular global correction (e.g., auto-level stretch, histogram equalization [1]) and local exposure correction [17][18]. However, these methods only use some heuristic histogram analysis to map per-pixel exposure to the desired one, without considering the spatial information of pixels (or regions). An interesting work [19] tries to enhancement image via frequency domain (i.e., block DCT). But some fixed tone curves are used for each image and blocking artifacts occasionally occur in their results.

基于软件的曝光调整有许多技术，包括最流行的全局校正（例如，自动水平拉伸、直方图均衡化[1]）和局部曝光校正[17][18]。然而，这些方法只使用一些启发式直方图分析将每个像素的曝光映射到所需的曝光，而没有考虑像素（或区域）的空间信息。一项有趣的工作[19]试图通过频域（即块DCT）增强图像。但是，对于每幅图像都使用了一些固定色调曲线，结果中偶尔会出现块效应。

Some algorithms [8][20] only consider the exposure of the regions of interest (ROI) and assume it is most important to the whole image correction. Different from ours, they use a known and predefined tone curve but we will estimate the specific curve for every image. Some tone mapping algorithms [21] can also be used to estimate the key of scene and infer a tone curve to map its original exposure to the desired key. However, the key estimation is based on the global histogram analysis and is sometimes inaccurate. Exposure fusion [22] combines well-exposed regions together from an image sequence with bracketed exposures. In contrast, we only use a single image as the input.

一些算法[8][20]只考虑了感兴趣区域（ROI）的曝光，并假设它对整个图像校正最为重要。与我们不同的是，他们使用一个已知的和预定义的色调曲线，但我们将估计每个图像的具体曲线。一些色调映射算法[21]也可用于估计场景的关键点，并推断色调曲线，以将其原始曝光映射到所需的关键点。然而，密钥估计是基于全局直方图分析的，有时是不准确的。曝光融合（Exposure fusion）[22]将图像序列中曝光良好的区域与括号内的曝光结合在一起。相反，我们只使用一个图像作为输入。

Since the exposure correction is kind of subjective, recent methods [23][4][9] enhance the input image using training samples from internet or personalized photos. However, our exposure correction is not relied on the selection of training images and only focuses on the input image itself. Another issue worth mentioning is that our approach does not aim to restore completely saturated pixels like [24].

由于曝光校正是一种主观的，最近的方法[23][4][9]使用来自互联网的训练样本或个性化照片来增强输入图像。然而，我们的曝光校正并不依赖于训练图像的选择，而只关注于输入图像本身。另一个值得一提的问题是，我们的方法并不旨在恢复像[24]这样完全饱和的像素。

3 Automatic Exposure Correction Pipeline

Our exposure correction pipeline is depicted in Fig. 2 and divided to two main steps: exposure evaluation and S-curve adjustment. Both components are performed in the luminance channel. To avoid bias due to different camera metering systems, or user’s manual settings, we would linearly normalize the input tonal range to [0, 1] at first.

我们的曝光校正管道如图2所示，分为两个主要步骤：曝光评估和S曲线调整。这两个组件都在亮度通道中执行。为了避免由于不同的相机测光系统或用户的手动设置而产生偏差，我们首先将输入色调范围线性标准化为[0，1]。

The heart of our system is an optimization-based, region-level exposure evaluation (see Section 4). In the exposure evaluation, we apply a Zone-based exposure analysis to estimate the desired zone (i.e., exposure) for each image region.We first segment the input image into individual regions (i.e., super-pixels). In each region, we measure visible details, region size, and relative contrast between regions. Then we formulate the optimal zone estimation as a global optimization which takes into account all these factors. We also use the high level information (e.g., face) to set the priority of the regions.

我们系统的核心是基于优化的区域级风险评估（见第4节）。在曝光评估中，我们应用基于区域的曝光分析来估计每个图像区域的期望区域（即曝光），我们首先将输入图像分割成单个区域（即超级像素）。在每个区域中，我们测量可见细节、区域大小和区域之间的相对对比度。然后，我们将最优区域估计公式化为一个考虑所有这些因素的全局优化。我们还使用高级信息（例如，人脸）来设置区域的优先级。

After the exposure evaluation, we estimate a best non-linear curve (S-curve) mapping for the entire image to push each region to its optimal zone. We further introduce a detail-preserving S-curve adjustment (see Section 5) instead of na¨ıve S-curve mapping to preserve local details and suppress halo effects in the final result.

在曝光评估之后，我们估计整个图像的最佳非线性曲线（S曲线）映射，以将每个区域推到其最佳区域。我们进一步引入了一个细节保持的S曲线调整（见第5节），而不是简单的S曲线映射，以保持局部细节和抑制最终结果中的光晕效应。

4 Region-level Exposure Evaluation

The aim of our exposure evaluation is to infer the image specific tone curve for the consequent detail-preserving S-curve adjustment. To achieve this goal, we first need to know what is the “best” exposure of each region and how to estimate them all together.

曝光评价的目的是为了推断出图像特定的色调曲线，从而实现后续细节保持S曲线的调整。为了实现这一目标，我们首先需要知道每个区域的“最佳”曝光是什么，以及如何共同估计它们。

4.1 Zone region

To measure the exposure, we borrow the concept of “Zone” from Ansel Adams’ Zone System [5], which is shown in Fig. 3(d). In Zone System, the entire luminance range [0, 1] is equally divided into 11 zones, ranging from O to X denoted by Roman numbers, with O representing black, V middle gray, and X pure white; these values are known as zones. In each zone, the mean intensity value is referred as its corresponding exposure. This concept was also used in recent HDR tone mapping applications [21][25] and realistic image composition [26].

为了测量曝光量，我们借用了Ansel Adams区域系统[5]中的“区域”概念，如图3（d）所示。在区域系统中，整个亮度范围[0，1]被平均划分为11个区域，从O到X用罗马数字表示，O表示黑色，V表示中灰色，X表示纯白色；这些值称为分区。在每个区域中，平均强度值被称为其相应的曝光。这个概念也用于最近的HDR色调映射应用[21][25]和真实图像合成[26]。

We represent the image by a number of zone regions.We first decompose the image into a set of regions by graph-based segmentation [27]. Each region falls into one of the zones. Then, we merge the neighboring regions with the same zone value. To extract high-level information (e.g., face/sky) for high priority of adjustment, we need to detect facial regions [28] and sky regions [29]. All connected regions belonging to face/sky regions are also merged. We call the final merged region as “zone region”. Fig. 3(a-c) shows the procedure of the zone region extraction.

我们用许多区域来表示图像，首先通过基于图的分割将图像分解成一组区域[27]。每个地区分为一个区域。然后，合并具有相同区域值的相邻区域。为了提取高级信息（例如，人脸/天空）以获得高优先级的调整，我们需要检测人脸区域[28]和天空区域[29]。属于面/天空区域的所有连接区域也将合并。我们将最终合并的区域称为“区域”。图3（a-c）示出了区域提取的过程。

4.2 Optimal zones estimation

The optimal zone estimation can be formulated as a global optimization problem by considering two aspects: maximizing the visual details and preserving the original relative contrast between neighboring zone regions.

最优区域估计可以归结为一个全局优化问题，它考虑了两个方面：最大化视觉细节和保持相邻区域之间的原始相对对比度。

Measure of visible details.

The amount of visible details in under-/over-exposed regions can be measured by the difference of the detected edges in these images which are generated by applying different gamma-curves on the input image I (the process is denoted as ). It is based on an observation: in an under-exposed region, we can detect more/less visible edges after the gamma correction when the gamma  is smaller/larger, and the edge number difference between two gamma-corrected images (one with small gamma, the other with large gamma) indicates the amount of recoverable details. The similar process can be applied to the over-exposed region as well.

曝光不足/曝光过度区域中的可见细节量可以通过这些图像中检测到的边缘的差异来测量，这些边缘是通过在输入图像I上应用不同的伽马曲线而生成的（该过程被表示为）。这是基于一个观察：在一个暴露不足的地区， 在gamma校正之后，当gamma 较小/较大，两个gamma校正图像（一个gamma较小，另一个gamma较大）之间的边数差表示可恢复细节的数量。类似的过程也可以应用于过度曝光区域。

 

 

 

 

 

 

 

 

 

 

 

For a color image, we need to compensate the possible reduction of color saturation caused by the luminance adjustment, especially on shadows. To avoid this issue, we transform it to YIQ color space and then scale the corresponding I, Q chroma values by the adjustment of Y luminance values.

对于彩色图像，我们需要补偿亮度调整可能导致的颜色饱和度降低，特别是在阴影上。为了避免这个问题，我们将其转换为YIQ颜色空间，然后通过调整Y亮度值来缩放相应的I，Q色度值。

Efficient implementation. For efficient computation, we enforce two extra constraints to largely reduce our search space of possible zone values: 1) Our adjustment uses the global S-curve which would map the same input pixel values to the same output. Thus we can consider the change of zone should be the same for the regions with the same original zone values; 2) Since our employed S-curve won’t change values across the middle gray (0.5), we can consider that the change of every zone is not allowed across zone V. In addition, our exposure is evaluated on the down-scaled image with their long edge no more than 400 pixels. So our segmentation and face/sky detection can be very efficient. For an 16-megapixel RGB image, the whole evaluation and correction time is 0.3 second on Core2 Duo CPU 3.16GHz with single-thread, no SSE acceleration.

高效实施。为了提高计算效率，我们增加了两个额外的约束，以大大减少可能区域值的搜索空间：1）我们的调整使用全局S曲线，该曲线将相同的输入像素值映射到相同的输出。因此，对于具有相同初始分区值的区域，分区的变化应该是相同的；2） 由于我们采用的S曲线不会改变中间灰色（0.5）的值，因此我们可以认为每个区域的变化不允许穿过V区域。此外，我们的曝光是评估缩小图像与他们的长边不超过400像素。因此，我们的分割和人脸/天空检测可以非常有效。对于1600万像素的RGB图像，在酷睿2双核cpu3.16GHz单线程处理器上，整个评估和校正时间为0.3s，没有SSE加速。

6 Experiments

6.1 Usability Study

Dataset: We perform our evaluation using a database of 4,000 images taken by our friends (including amateur and professional photographers) with direct camera output. These images varies on scenes, locations, lighting conditions and camera models (e.g., DSLR, compact, mobile cameras). We ask five subjects to divide all images into three groups according to different extents of exposure problem. Three groups are “severely badly-exposed, definitely need correction” (Group A), “slightly badly-exposed photos, may require a little correction” (Group B), and “well-exposed, no more correction” (Group C). Finally, we obtain three different datasets respectively: “Group A” (975 images), “Group B” (1,356 images) and “Group C” (1,669 images) according to the majority agreement of five subjects. Fig. 9 (a) shows several examples.

数据集：我们使用4000张由我们的朋友（包括业余和专业摄影师）拍摄的带有直接相机输出的图像数据库进行评估。这些图像因场景、位置、照明条件和相机型号（例如，单反、紧凑型、移动相机）而异。我们要求五名受试者根据不同程度的暴露问题将所有图像分为三组。三组分别为“曝光严重，肯定需要校正”（A组）、“曝光轻微，可能需要稍微校正”（B组）和“曝光良好，不再校正”（C组）。最后，我们根据五位受试者的多数同意，分别获得三个不同的数据集：A组（975张图像）、B组（1356张图像）和C组（1669张图像）。图9（a）示出了几个示例。

可用性研究(a） 三组例子：a组（严重曝光）、B组（轻度曝光）、C组（良好曝光）(b） 在所有的组和三个不同的组中分别对我们的输入、照片库和Picasa进行两两比较。每个颜色条表示受欢迎图像的平均百分比（带有I形标准偏差条）。

Procedure: We will compare with automatic exposure corrections in several popular photo editing tools to manifest our method would become a better candidate. All of results are achieved by default parameters. We invite other 12 volunteers (7 males and 5 females) with balanced expertise in photography and camera use to perform pairwise comparison between our result and one of three other images: 1) input image, 2) result byWindows Live Photo Gallery’s Auto-adjust, exposure only (http://download.live.com/photogallery), 3) result by Google Picasa’s Auto-contrast (http://picasa.google. com/). For each pairwise comparison, the subject has three options: better, or worse, or no preference. Subjects are allowed to view each image pair back and forth for the comparison. To avoid the subjective bias, the group of images, the order of pairs, and the image order within each pair are randomized and unknown to each subject. This usability study is conducted in the same settings (room, light, and monitor).

程序：我们将比较几种流行的照片编辑工具中的自动曝光校正，以表明我们的方法将成为一个更好的候选人。所有结果都是通过默认参数实现的。我们邀请其他12名志愿者（7名男性和5名女性）在摄影和相机使用方面具有平衡的专业知识，将我们的结果与其他三幅图像中的一幅进行两两比较：1）输入图像，2）Windows Live照片库的自动调整结果，仅限曝光(http://download.live.com/photogallery),3）谷歌Picasa的自动对比结果(http://picasa.google. com/）。对于每个两两比较，受试者有三个选择：更好，或更差，或没有偏好。受试者可以前后查看每个图像对进行比较。为了避免主观偏见，图像组，成对的顺序，以及每一对中的图像顺序是随机的，每个受试者都不知道。可用性研究是在相同的环境下进行的（房间、灯光和显示器）。

Usability study results: The main user study results are summarized in Fig. 9 (b). Each color bar is the averaged percentage of the favored image over all 12 subjects (Ishape error bar denotes the standard deviation). From results on “All Groups” (without distinguishing the photos from different groups), we can see that the participants overwhelmingly select our result over the input (70.2% vs. 5.9%), Photo Gallery (60.5% vs. 29.6%), and Picasa (58.3% vs. 12.5%).

可用性研究结果：主要用户研究结果如图9（b）所示。每个颜色条是所有12个受试者喜欢的图像的平均百分比（Ishape误差条表示标准偏差）。从“所有组”（不区分不同组的照片）的结果中，我们可以看到，参与者压倒性地选择我们的结果，而不是输入（70.2%对5.9%）、照片库（60.5%对29.6%）和Picasa（58.3%对12.5%）。

“Group A” results show that our approach works significantly better for severely badly-exposed photos. The participants show a strong bias in preference towards our correction when compared to input images (92.3% vs. 2.7%) and other automatic tools (87% vs. 8.5% against Photo Gallery, 84% vs. 6.4% against Picasa). The results from “Group B” indicate that slightly badly-exposed photos can benefit more from our correction than other methods as well. In “Group C”, our approach also performs very well - for near 92% photos, our method does not make the result worse. It is quite nontrivial and very important for practical use, especially for batch-processing photos.

“A组”的结果表明，我们的方法对严重曝光的照片效果明显更好。与输入图像（92.3%对2.7%）和其他自动工具（87%对8.5%对照片库，84%对6.4%对Picasa）相比，参与者对我们的校正表现出强烈的偏好。“B组”的结果表明，轻微曝光的照片比其他方法更能从我们的校正中获益。在“C组”中，我们的方法也表现得很好——对于接近92%的照片，我们的方法不会使结果变得更糟。对于实际应用，特别是对于批量处理照片，它是非常重要的。

Fig. 9 (b) also graphically show two phenomenons on “Group C” compared with “Group A”: 1) the margin between our result favored and no preference is smaller, and 2) all standard derivations are larger. They both indicate that the exposure correction itself is somewhat subjective especially for “not bad” photos. Subjects show different tastes for good photos correction, which has been discussed in [9][4], but most of these subjects consistently agree with our correction for relatively bad photos.

图9（b）还以图形方式显示了“C组”与“A组”相比的两个现象：1）我们的结果有利与无偏好之间的差距较小，2）所有标准推导都较大。它们都表明曝光校正本身有点主观，特别是对于“不错”的照片。受试者对好照片的纠正表现出不同的兴趣，这在[9][4]中已经讨论过，但是大多数受试者一致同意我们对相对较差照片的纠正。

After the user study, we also ask all participants to articulate the criteria for their feedbacks.We conclude the main criteria: 1) the over-/under-exposed regions of interest should be well corrected; 2) well-exposed regions should not be over-corrected; and 3) the colors in corrected images should look natural. Other feedbacks include “the color of a few individual regions sometimes looks slightly unrealistic”, “in some cases, the corrected results bring in some noise”, and “I want some parameters tuning so that I can control the results.”. Overall, most participants like our correction and want to use it for their daily photos processing.

在用户研究之后，我们还要求所有参与者阐明他们反馈的标准，我们得出的主要标准是：1）过度/不足暴露的兴趣区域应该得到很好的纠正；2） 暴露良好的区域不应过度校正；3）校正后的图像颜色要自然。其他反馈包括“个别区域的颜色有时看起来有点不切实际”，“在某些情况下，校正后的结果会带来一些噪声”，“我需要一些参数调整，以便我可以控制结果。”。总的来说，大多数参与者喜欢我们的更正，并希望将其用于日常照片处理。

Visual quality comparisons: Fig. 10 shows three examples from “Groups A”. These photos show several common badly-exposed scenarios, such as outdoor backlit, dimlight indoor environment, which are very challenging for existing tools. As we can see, their corrections take no effect, but our method brings more visible details into badly-exposed areas while preserving the original appearance in well-exposed areas. Fig. 11(a)(b) show two examples from “Groups B”, whose exposures look somewhat problematic. Our results look much more appealing, especially on important areas, e.g., over-exposed sky (Fig. 11(a)) and under-exposed face (Fig. 11(b)). Fig. 11(c)(d) show two well-exposed examples from “Groups C”. Our corrections seem to be imperceptible because the dark silhouette regions (Fig. 11(c)) have few detectable visible details and the black clothes (Fig. 11(d)) have lower priority than well-exposed faces, which would contribute little to the change of zone in our optimization.

视觉质量比较：图10显示了“A组”的三个例子。这些照片显示了几种常见的严重暴露场景，例如室外背光、室内昏暗环境，这对现有工具来说非常具有挑战性。正如我们所看到的，他们的修正没有效果，但我们的方法带来了更多的可见细节到严重暴露的地区，同时保留原来的外观在良好的暴露地区。图11（a）（b）显示了来自“b组”的两个示例，其暴露看起来有点问题。我们的结果看起来更吸引人，特别是在重要的区域，例如，过度暴露的天空（图11（a））和低于暴露的脸（图11（b））。图11（c）（d）显示了来自“c组”的两个良好暴露的示例。我们的修正似乎是难以察觉的，因为深色轮廓区域（图11（c））几乎没有可检测的可见细节，而黑色衣服（图11（d））的优先级低于曝光良好的人脸，这对我们优化中区域的变化贡献不大。

图10 从A组中随机抽取例子。我们可以注意到更多关于前景脸（a）、前景观众（b）和街道场景（c）的细节。

图11 随机选择B组（a-b）和C组（c-d）各2例。

6.2 Comparisons with other academic methods

In consequent comparisons, our results are generated by the same parameters used in useability study. In Fig. 12(b)(c), we compare with two traditional histogram equalization algorithm [1][2] (by Matlab function histeq, adapthisteq). We can notice local contrast reduction and undesired halo effects in their correction results shown in Fig. 12(b)(c). However, our result shown in Fig. 12(e) looks more natural.We also compare our method with a well-known tone-mapping operator [21] (shown in Fig. 12(d)). Since their automatically estimated scene key is not accurate and tends to be higher than the actual key in this case, their result looks a little over-exposed.

在随后的比较中，我们的结果是由可用性研究中使用的相同参数生成的。在图12（b）（c）中，我们比较了两种传统的直方图均衡化算法[1][2]（通过Matlab函数histeq，adaptehisteq）。我们可以在图12（b）（c）所示的校正结果中注意到局部对比度降低和不期望的光晕效应。然而，我们在图12（e）中所示的结果看起来更自然，我们还将我们的方法与众所周知的色调映射算子[21]（如图12（d）所示）进行了比较。由于它们自动估计的场景关键点不准确，并且在这种情况下往往高于实际关键点，因此它们的结果看起来有点过度曝光。

In Fig. 13, we directly use the image and result from internet-based image restoration [23] for comparison. In this case, we can see our result has more visual details in local under-exposed areas than their provided result. Besides, their approach exaggerates over-exposed sky areas while our method can preserve their original appearance.

在图13中，我们直接使用基于互联网的图像恢复的图像和结果进行比较[23]。在这种情况下，我们可以看到我们的结果在局部曝光不足的区域比他们提供的结果有更多的视觉细节。此外，他们的方法夸大了暴露的天空面积，而我们的方法可以保持其原始外观。

Exposure Fusion [22] is a fairly new concept that fuses all well-exposed regions together from a series of bracketed exposures. The good exposure is measured by some features: contrast, saturation and closeness to middle gray. Fig. 14 shows an example from their paper. We can see our result is visually approaching theirs, but our input is only a single frame from their input sequence. To perceive how well their algorithm works on a single input image, we make a modification of their method for comparison: (1) applying a series of global brightness adjustment (e.g., multiplying luminance with 1/4, 1/2, 1, 2, 4) in Fig. 15(a); (2) applying a set of different gamma curves (e.g., gamma values -3, -1.5, 1, 1.5, 3) in Fig. 15(b). Their results look either less vivid, or have lower global contrast than ours.

曝光融合[22]是一个相当新的概念，它将所有曝光良好的区域从一系列带括号的曝光中融合在一起。良好的曝光度是通过一些特征来衡量的：对比度、饱和度和接近中灰度。图14显示了他们论文中的一个例子。我们可以看到我们的结果在视觉上接近他们，但我们的输入只是他们输入序列中的一帧。为了了解他们的算法在单个输入图像上的工作情况，我们对他们的比较方法进行了修改：（1）在图15（a）中应用一系列全局亮度调整（例如，将亮度乘以1/4、1/2、1、2、4）(2） 在图15（b）中应用一组不同的伽马曲线（例如伽马值-3、-1.5、1、1.5、3）。他们的结果要么不够生动，要么整体对比度比我们的低。

图12 与直方图均衡化[1]、自适应直方图均衡化[2]和色调再现[21]的比较。黄色/红色箭头显示不需要的光晕效果/对比度降低。

图13。与基于互联网的恢复比较[23]。图片（a-c）取自他们的论文。参考结果（b）应用于固定键。黄色/红色箭头显示曝光不足/曝光过度。

图14 与曝光融合[22]在输入图像序列（取自他们的论文）上的比较，我们的算法只使用单帧（b）作为输入。

图15。在单个输入图像上与曝光融合的比较[22]。我们只使用输入图像（用绿色方框表示），而曝光融合则使用与输入图像具有不同曝光的合成图像序列。红色箭头显示不需要的工件。

图16。与基于学习的色调调整的比较[4]。图像（a，b，d）取自[4]。

图17。我们在噪点放大方面的失败案例。

We show the comparison with learning-based adjustment [4] and assisted correction by expert in Fig. 16. As we can see, our result has more luminance details than their result on under-exposed areas and even much closer to the assisted result (from “Retoucher E” mentioned in [4]). Here, please ignore the difference in colors and focus on the luminance modification since the assisted adjustment includes both exposure correction and white balance.Without the need of training images, our approach obtain appealing results as well.

我们在图16中展示了与基于学习的调整[4]和专家辅助纠正的比较。正如我们所看到的，我们的结果比他们在曝光不足区域的结果有更多的亮度细节，甚至更接近于辅助结果（从[4]中提到的“修饰E”）。在这里，请忽略颜色的差异，重点放在亮度调整上，因为辅助调整包括曝光校正和白平衡。在不需要训练图像的情况下，我们的方法也取得了很好的效果。

Fig. 17 (d) shows the limitation of our method. Since the correction does not consider the noise issue in our exposure evaluation, noise would become noticeable after we lighten dark areas. The issue may be addressed by suppressing the excessive noise amplification or applying denoising for these regions as preprocessing. We will further explore this issue in the future work.

图17（d）显示了我们方法的局限性。由于校正没有考虑噪声问题，在我们的曝光评估，噪声将变得明显后，我们照亮黑暗的地区。该问题可以通过抑制过多的噪声放大或在预处理时对这些区域进行去噪来解决。我们将在今后的工作中进一步探讨这个问题。

7 Conclusions

We have presented an automatic method for the exposure correction of consumer photographs. The heart of this method is an optimization-based exposure evaluation and a detail-preserving curve adjustment algorithm. By simultaneously considering visible details in each region and relative contrast between regions, we are able to obtain appropriate exposure at the region level and produce natural-looking results.

我们提出了一种用于消费者照片曝光校正的自动方法。该方法的核心是基于优化的曝光评估和细节保持曲线调整算法。通过同时考虑每个区域的可见细节和区域之间的相对对比度，我们能够在区域级别获得适当的曝光，并产生自然的效果。