Print

发布时间: 2019-08-16
摘要点击次数:
全文下载次数:
DOI: 10.11834/jig.180608
2019 | Volume 24 | Number 8




    图像处理和编码    




  <<上一篇 




  下一篇>> 





局部自适应的灰度图像彩色化
expand article info 曹丽琴1, 商永星1, 刘婷婷2, 李治江1, 马爱龙3
1. 武汉大学印刷与包装系, 武汉 430079;
2. 武汉大学中国南极测绘研究中心, 武汉 430079;
3. 武汉大学测绘遥感信息工程国家重点实验室, 武汉 430079

摘要

目的 现有的灰度图像彩色化方法为了保证彩色化结果在颜色空间上的一致性,往往采用全局优化的算法,使得图像边界区域易产生过渡平滑现象。为此提出一种局部自适应的灰度图像彩色化方法,在迁移过程中考虑局部邻域像素信息,同时自动调节邻域像素权重,在颜色正确迁移的同时保证清晰的边界信息。方法 首先结合SVM(support vector machine)和ISLIC(improved simple linear iterative clustering)算法获取彩色图像和灰度图像分类结果图;然后在分类基础上,确定灰度图像高置信度像素点,并根据图像纹理特征,在彩色图像中寻找灰度图像的像素匹配点;最后利用自适应权重均值滤波实现高置信度匹配像素点的颜色迁移,并利用迁移结果对低置信度像素点进行颜色扩散,以完成灰度图像彩色化。结果 实验结果显示,本文方法获得的彩色化迁移结果评分均高于3.5分,特别是局部放大区域评价结果均接近或高于4.0分,高于其他现有彩色化方法评价分数。表明本文方法不仅能够保证颜色迁移的准确性和颜色空间的一致性,同时也能获取颜色区分度高的边界细节信息。与现有的典型灰度图像彩色化方法相比,彩色化结果图在颜色迁移的正确性和抑制边界区域颜色的过渡平滑上都有更优的表现。结论 本文算法为灰度图像彩色化过程中抑制颜色越界问题提供了新的指导方法,能有效地应用于遥感、黑白图像/视频处理、医学图像着色等领域。

关键词

彩色化; 颜色迁移; 局部自适应; 一致性; 平滑

Novelimage colorization of a local adaptive weighted average filter
expand article info Cao Liqin1, Shang Yongxing1, Liu Tingting2, Li Zhijiang1, Ma Ailong3
1. School of Printing and Packaging, Wuhan University, Wuhan 430079, China;
2. Chinese Antarctic Center of Surveying and Mapping, Wuhan University, Wuhan 430079, China;
3. State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China
Supported by: National Key Research and Development Program of China (2017YFB0504202);Fundamental Research Funds for the Central Universities (2042018kf0229);National Natural Science Foundation of China (41671441, 41771385)

Abstract

Objective Image colorization is the process of assigning color information to grayscale images and retains grayscale image texture information. The aim of colorization is to increase the visual appeal of an image. This technology is widely used in many areas, such as medical image illustrations, remote sensing images, and old black-and-white photos. Colorization methods have to main categories, namely, user-assisted and automatic colorization methods. User-assisted colorization methods require users to manually define a layer mask or mark color scribbles on a grayscale image. This method is time-consuming and cannot provide sufficient and desirable color scribbles. Automatic colorization methods can reduce user effort and transfer color from a sample color image. The color image is called the reference/source image, and the grayscale image to be colorized is called the target image. The primary difficulty of these automatic colorization methods is to accurately transfer colors and satisfy spatial consistency. Most of these approaches achieve spatial coherency by using weighted filter or global optimization algorithms during colorizing. However, these methods may result in oversmoothed colorization or blur color in edge regions. Method We use a grayscale image colorization approach based on a local adaptive weighted average filter. This proposed method considers local neighborhood pixel information and automatically adjusts the domain pixel weights to ensure correct color migration and clear boundary results. A reference image with similar contents as the target image is provided to achieve color transfer. The method includes the following steps:First, the class probability distribution and classification are obtained. Support vector machine (SVM) is adopted to calculate class probability based on feature descriptors, mean luminance, entropy, variance, and local binary pattern (LBP) and Gabor features. The probability results and classification are post-processed to enhance the spatial coherency combined with superpixels that are extracted based on improved simple linear iterative clustering (ISLIC). Second, the color candidate in the reference image is determined based on matching low-level features in the corresponding class. Thereafter, each pixel with high-confidence class probability is assigned a color from the candidate pixel by using an adaptive weight filter. The adaptive weight, which is defined by the class probability of small neighborhood pixels around the corresponding pixel, can improve local spatial consistency and avoid confusion colorization in the boundary region. Finally, the optimization-based colorization algorithm is used on the remaining unassigned pixels with low-confidence class probability. Result This paper analyzes single pixel-based, weighted average, and adaptive weighted average methods. Results demonstrate that the adaptive weighted average method is better than the other strategies. The colorized images illustrate that our method takes advantage of the other strategies and that it not only has high spatial consistence but also ensures the boundary detail information with high color discrimination. Compared with previous colorization methods, our method works well in colorization. Colorized images achieved by Gupta's method and Irony's method have obvious erroneous colors due to inaccurate matching or corresponding pixels. Charpiat's method produce oversmoothed color on the boundary regions. Images colorized by using Zhang's method are extracted by training more than a million color images and diversities of colors based on CNN. However, some unreliable and undistinguished colors appear on the boundary contours. The colorization results obtained by using a local adaptive weighted average filter ensures the correctness of color transfer and spatial consistence and avoids oversmoothing simultaneously in the edge areas. Thus, our proposed method performs better than the existing methods. The evaluation scores for experimental images using our method are higher than 3.5, especially in local areas, and the evaluation results are close to or higher than 4.0, which is greater than the results of those using the existing colorization method. Conclusion A new colorization approach using color reference images is presented in this paper. The proposed method combines SVM and ISLIC to determine the class probabilities and classifications with high spatial coherence for images. The corresponding pixels are matched based on the space features according to the same class label between the reference and the target images. A local adaptive weighted average filter is defined to transfer the chrominance from the source image to the grayscale image with high-confidence pixels to facilitate spatially coherent colorization and avoid oversmoothing. The colorized pixels are considered automatic scribbles to be spread across all pixels by using global optimization to obtain the final colorization result. Experimental results demonstrate that our proposed method can achieve satisfactory results and is competitive with the existing methods. However, several limitations are observed. First, this method is not fully automated, requiring some human intervention to provide class samples during the process of calculating class probability. Second, the selected space features are not optimal for all images, especially for images with complex textures and rich colors. We will focus on fully automatic operation and general features in our future work.

Key words

colorization; color transfer; local spatial coherency; consistency; smooth

0 引言

灰度图像彩色化方法是利用计算机给灰度图像加上颜色信息的处理过程,目的是增强图像的视觉感知能力[1]。该技术广泛应用于老旧黑白照片加工、经典电影颜色再现制作和医学图像着色等领域。

现有的灰度图像彩色化方法有两大类:用户辅助着色法和自动着色法。

用户辅助着色方法需要用户手动标记颜色[2-6]或标记颜色迁移区域[1, 7-9]。如Levin等人[2]利用手工着色的方法将彩色样本标记至灰度图像,然后根据着色像素与临近像素的相关关系,通过最小二乘法进行优化,实现全局图像的彩色化。手工提供颜色样本虽然可以按照需求提供用户满意的颜色信息,但大量的样本标记耗时耗力。基于此,学者们提出利用参考图像进行颜色迁移,即将用户提供的与灰度图像场景相近或相似的彩色图像(也称为源图像)的颜色信息迁移至目标灰度图像。如Gupta等人[1]采用分割方法,结合人工标记区域,寻找各分割区域中心像素的匹配点,并进行着色,利用Levin等人[2]的优化方法将中心像素的颜色进行扩散,完成整幅图像的颜色迁移,并通过扩散保证局部空间颜色的一致性。Irony等人[7]采用监督分类的策略,结合图像纹理特征寻找匹配点,利用均值滤波方法将源图像中匹配像素点的彩色信息迁移至目标图像。这些方法均需要先在源图像中找到与目标图像匹配的像素点或所属颜色类别的概率分布,然后基于匹配结果实现颜色的迁移。

自动着色方法指利用机器学习或统计学习的方法实现自动颜色迁移。典型的方法有Welsh算法[10]、直方图统计法[11]、基于多模型预测方法[12]、深度学习方法[13-14]等。Welsh算法[10]由于未考虑像素空间结构信息,仅利用亮度信息进行像素匹配,迁移结果容易缺乏空间一致性。直方图统计法[11]和基于多模型预测方法[12]考虑局部特征信息,能准确地传输颜色并满足空间一致性,但这些方法在着色过程中通过加权滤波实现空间的一致性,容易导致边缘区域的着色出现越界或颜色模糊。基于深度学习的灰度图像彩色化是一种新的颜色迁移方法,优势是能通过大规模的源图像样本获取足够多的颜色信息,但大量的样本对某些特定的对象可能造成颜色迁移偏差,如衣服、头发颜色等[14]

为了解决颜色迁移过程中出现的边界区域颜色过渡平滑的问题,本文提出一种局部自适应的灰度图像彩色化方法。该方法在图像分类结果的基础上,结合纹理特征寻找匹配像素点,以保证迁移结果在颜色空间上的局部一致性。在颜色迁移过程中,利用局部自适应权重均值滤波,对置信度高的匹配像素点进行颜色迁移,避免边界区域的着色出现过界,同时通过高置信度参数确定匹配像素点的准确性。

1 着色方法描述

图 1为本文灰度图像着色流程图,具体为:

图 1 灰度图像着色流程图
Fig. 1 Flowchart of grayscale image colorization

1) 类别概率计算及分类结果获取。利用支持向量机(SVM)计算初始类别概率并进行初始分类。为了提高分类空间的一致性,结合图像超像素,对初始类别概率及分类结果进行重新计算。图像超像素由改进的SLIC(simple linear iterative clustering)方法生成。

2) 实现目标图像至源图像的像素点匹配。通过亮度和纹理特征,结合分类类别信息,在源图像中寻找匹配像素点,同时通过类别概率信息,确定置信度参数值,对置信度高的像素进行匹配。

3) 对灰度图像中高置信度像素点进行着色。着色的方法考虑局部邻域像素信息,采用本文提出的自适应权重均值滤波,将源图像彩色信息迁移至灰度图像。

4) 利用Levin等人[2]的优化算法对目标图像中的低置信度像素进行着色。

1.1 图像分类及类别概率计算

1) 特征提取。本文融合图像的多特征信息进行图像分类及像素点匹配。特征包括:均值亮度、图像熵、方差、局部二值模式(LBP)和Gabor特征。均值亮度能较好地反映图像的全局性,可以有效地用于像素点的匹配及颜色的扩散[2, 10]。图像熵、方差和局部二值模式(LBP)是描述图像局部纹理特征的重要参数,Gabor特征用以表达图像的方向性和空间频率,描述图像的纹理特性。

2) 超像素提取。图像超像素是指具有相似纹理、颜色、亮度等特征的相邻像素构成的有一定视觉意义的不规则像素块。本文利用改进的简单线性迭代聚类(ISLIC)方法进行超像素的提取。简单线性迭代聚类(SLIC)方法的思想是通过线性迭代聚类实现图像分割生成超像素,该方法运算速度快,且能够生成紧凑、大小形状近似均匀的超像素。为了保证在实现过程中超像素边界能很好地贴合物体边缘,本文采用Kim等人[15]提出的基于Sigma滤波的ISLIC方法。

3) 分类及概率计算。在已提取的图像特征的基础上,利用SVM进行分类分析,同时计算各像素类别概率分布值。在分类过程中,训练样本通过人为选择获得。由于基于像素的分类容易引起单像素类别的错分或误分,导致分类空间不连续,因此结合超像素对分类结果进行后处理。因为超像素内的所有像素均具有相似的颜色和亮度,类别信息相同或相近,因此在任一超像素中,对不在超像素边界区域内的像素,若满足以下任一假设条件,则定义为存在误分类的像素:a)像素的类标签与邻域像素的类标签均不同;b)具有相同类标签的像素所占比例小于15%。

对于误分类的像素调整为与邻域像素同类别,并通过权重平均方法调整其类别概率。对存在误分类的像素$x_i$,其属于第$k$类别的概率为$p_k(x_i)$,调整后的类别概率为

$ {p_k}\left( {{x_i}} \right) = \sum\limits_{{x_j} \in {\mathit{\boldsymbol{ \boldsymbol{\varPsi} }}_n}} {{w_{ij}}} {p_k}\left( {{x_j}} \right) $ (1)

式中,$x_j$为像素$x_i$所属超像素$\mathit{\pmb{Ψ}}_{\rm{n}}$内的其他邻域像素,$p_k(x_j)$为属于类别$k$的概率。$w_{ij}$为权重

$ {w_{ij}} = \frac{{\exp \left( { - \left| {{L_{{x_i}}} - {L_{{x_j}}}} \right|} \right)}}{{\sum\limits_{{x_m} \in {\mathit{\boldsymbol{ \boldsymbol{\varPsi} }}_n}} {\exp } \left( { - \left| {{L_{{x_i}}} - {L_{{x_m}}}} \right|} \right)}} $

式中,$L_{x_{i}}$和$L_{x_j}$分别为像素$x_i$和$x_j$的亮度值。参见图 1中的概率类别分布图,其为3个类别概率的假彩色合成图。

1.2 像素点匹配

对目标图像和源图像的像素进行匹配,即为目标图像中的每一个像素在源图像中寻找最相似的像素。对于目标图像中的任意像素$x_i$,在源图像中寻找与其具有相同类别的像素作为匹配像素集$\bf{R}$,并在匹配像素集中寻找与像素$x_i$特征最为相近的像素$y_j$作为匹配像素点,$y_j∈\bf{R}$,即

$ {y_j} = \mathop {\arg \;\mathop {\min }\limits_{{y_j} \in {\bf{R}}} }\limits_{} F\left( {{x_i},{y_j}} \right) $ (2)

$ F\left( {{x_i},{y_j}} \right) = \sum\limits_{\eta = 1}^N {{\varphi _\eta }} {D_\eta }\left( {{x_i},{y_j}} \right) $ (3)

式中,$D_η(x_i, y_j)$表示像素$x_i$和像素$y_j$的特征$η$的相似度,指归一化后的欧氏距离。$N$表示选取的特征类别数,本文选择平均亮度、熵、方差、LBP和Gabor共5个特征。图 2为选择单一特征进行匹配得到的彩色化结果,通过分析发现,单一的Gabor特征和平均亮度特征相较于其他单一特征更易获得更好的彩色化效果。基于这一结果,利用试错法,通过不断试验和消除误差,将获得较好彩色化结果的特征权重赋值较大,其他特征权重赋值较小,通过人工干预,多次尝试,确定平均亮度、熵、方差、LBP和Gabor的特征权重$φ_η$分别为0.2、0.1、0.1、0.1、0.5。

图 2 基于单一特征的彩色化结果
Fig. 2 Colorization using a single feature ((a) mean luminance; (b) entropy; (c) variance; (d) LBP; (e) Gabor)

在获得匹配像素对后,将源图像中匹配像素$y_j$及其邻近像素颜色信息迁移至目标图像像素$x_i$,对目标图像进行着色。

1.3 目标图像初始着色

为了保证迁移颜色的准确性和空间的一致性,采用置信度参数确定初始着色像素。置信度值低的像素,较难通过匹配像素点进行颜色迁移,对其采用Levin的优化算法进行扩散。置信度定义为

$ con f\left( {{p_k}\left( {{x_i}} \right)} \right) = \mathop {\max }\limits_k {p_k}\left( {{x_i}} \right) $ (4)

式中,$p_k(x_i)$为目标图像像素$x_i$属于第$k$类别的概率,$k=1, 2, …, K$,$K$为类别总数。

本文设定置信度阈值为0.85,即对于置信度超过0.85的像素点进行颜色迁移。图 1中的置信度分布图中,白色区域为置信度值超过阈值的像素。迁移颜色空间为$CIELab$,其中亮度空间保持不变,计算待匹配点的$a, b$值。在颜色迁移中,为了增强颜色空间的一致性,部分学者考虑邻近像素点通过均值滤波的方法实现[7, 10],但这类方法在边界区域容易出现过平滑的现象。本文采用自适应权重均值滤波,对具有同类别$k$的匹配像素点对$(x_i, y_j)$,考虑源图像$y_j$的$n×n$邻域$\mathit{\pmb{Υ}}(y_j)$,目标中待着色点的颜色值$C(x_i)$为

$ C\left( {{x_i}} \right) = \left\{ {\begin{array}{*{20}{l}} {\sum\limits_{{y_m} \in \mathit{\boldsymbol{Y}}\left( {{y_j}} \right)} {{\varphi _m}} C\left( {{y_m}} \right)}& {F\left( {{x_i},{y_j}} \right)\ge\bar F\left( {{x_i},{y_j}} \right)}\\ {C\left( {{y_j}} \right)}&{F\left( {{x_i},{y_j}} \right) < \bar F\left( {{x_i},{y_j}} \right)} \end{array}} \right. $ (5)

式中,

$ \bar F\left( {{x_i},{y_j}} \right) = \sum\limits_{{y_m} \in \mathit{\boldsymbol{Y}}\left( {{y_j}} \right)} {{\varphi _m}} F\left( {{x_i},{y_m}} \right) $ (6)

$ {\varphi _m} = \frac{{\exp \left( { - \left| {{p_k}\left( {{x_i}} \right) - {p_k}\left( {{y_m}} \right)} \right|} \right)}}{{\sum\limits_{{y_m} \in \mathit{\boldsymbol{Y}}\left( {{y_j}} \right)} {\exp } \left( { - \left| {{p_k}\left( {{x_i}} \right) - {p_k}\left( {{y_m}} \right)} \right|} \right)}} $ (7)

式中,$C(y_i)$为源图像像素点$y_j$的颜色信息,$F(x_i, y_j)$为匹配像素对的特征差值,$\overline{F}\left(x_{i}, y_{j}\right)$为像素点$x_i$与匹配像素点的邻域平均差值,权重$φ_m$由类别概率值确定。当匹配像素对特征差值小于待匹配点与邻域均值特征差时,邻域像素权重值均为0,将颜色进行直接迁移;当匹配像素点对特征差值大于平均特征差值时,将均值颜色迁移至待匹配点。颜色迁移结果见图 1中初始着色图部分。

1.4 着色像素点扩散

本文将已着色的点作为标记的着色样本(micro scribbles),采用Levin的优化方法将颜色自适应地扩散至整幅图像。Levin优化的基本思想是:具有相同亮度的邻域像素应该具有相同的颜色信息。通过这一思想,构造一个优化函数,将像素与其邻域像素颜色的加权平均值的差最小化[2]。扩散后的结果见图 1中的彩色化结果图部分。

2 实验结果及分析

2.1 实验结果

图 3为灰度图像彩色化结果图。为了验证算法的有效性,将自适应方法与未考虑邻域像素迁移方法和考虑邻域像素的均值滤波进行比较,结果见图 3(a)—(c),其源图像见图 1

图 3 灰度图像彩色化结果图
Fig. 3 Results of grayscale image colorization
((a) pixel-based method without neighboring information; (b) weighted average method; (c) ours)

图 3可以看出,未考虑邻域像素的信息的迁移结果(图 3(a))中出现了明显错误迁移的区域,如树林中间出现了明显的蓝色区域,房屋顶部和墙面也有不连续的蓝色像素点出现,图像颜色空间连续性不足。对均值滤波方法得到的图像(图 3(b)),图像颜色有很好的空间一致性,但在部分区域出现了过平滑现象,如房屋右边界与天空相连接的地方、天空中云的颜色也呈现与房屋颜色相近的黄色。本文提出的方法综合了以上两种方法的优点,迁移结果不仅具有较好的空间一致性,颜色空间过渡平滑,同时图像中边界区域能保持较好的颜色细节,得到清晰自然的彩色化图像(图 3中红圈区域)。

2.2 对比分析

将灰度图像彩色化结果图与文献[1, 7, 12, 14]的结果进行比较,文献[1, 7]与本文方法均需要少量人工干预,文献[12, 14]为全自动化彩色化方法。图 4为彩色化结果对比图,为了更清楚地对比分析各彩色化方法,将部分区域放大显示,见图 4中红色矩形框。

图 4 彩色化结果图对比
Fig. 4 Colorization results comparison ((a) source images; (b) target images; (c) ours; (d) reference [1]; (e) reference [7]; (f) reference [12]; (g) reference [14])

从彩色化结果图可知,本文得到的结果在颜色迁移正确性上表现较好,且能有效避免区域空间颜色越界。从彩色化结果的正确性来看,文献[7, 14]均存在大量的错误迁移颜色(图 4(e)中的风景图和图 4(g)中的蒙娜丽莎图),原因在于文献[7]采用单像素匹配过程中出现较多的错误匹配点对,而文献[14]通过深度学习的方法,虽然可以对大量的样本图像进行学习,但同时也迁移了较多与实际情况不符的颜色。在颜色边界越界上,文献[1, 12]均有不同程度的越界区域(图 4(d)斑马头部边界区域存在一定的颜色模糊)。在文献[12]中,迁移过程中所有像素均考虑邻域信息,并利用图割的方法增强全局颜色的一致性,因此在边界区域容易产生模糊,如蒙娜丽莎头部边沿、山顶边界等(图 4(f)中红色矩形区域)。

为了量化结果优劣,参考文献[16]的评价方法,邀请了89位年龄在12~51岁的观察者(56名男性和33名女性)对彩色化结果进行评价。图 4中不同方法的实验结果将随机展示,对颜色的自然度、纹理及与参考图像颜色的相似性综合考虑,给出评价分数,分值范围为1~5,其中1分为最差结果,5分为最优结果。表 1为各实验结果的平均分值。

表 1 图 4中不同方法彩色化结果评价
Table 1 Average scores of evaluation for images in Fig. 4

下载CSV
/分
方法 蒙娜丽莎图 蒙娜丽莎局部图 风景图 风景局部图 斑马图 斑马局部图
文献[1] 4.32 4.1 4.34 3.9 3.76 3.52
文献[7] 2.84 3.25 2.84 2.83 4.11 4.01
文献[12] 3.64 3.37 4.23 4.05 4.13 3.95
文献[14] 3.52 3.77 3.84 3.52 3.76 3.43
本文 4.23 4.23 3.8 3.96 4.27 4.04
注:加粗字体为最优结果。

表 1可以看出,本文和文献[1]方法在整体和局部颜色迁移中的分值均高于3.5分,表明这两种方法均能取得较好的彩色化结果。尤其是本文方法在局部颜色迁移中的分值均接近或高于4.0,明显优于现有的其他方法,说明本文方法在局部及边界颜色迁移上相较其他方法存在较大优势。为了进一步比较本文和文献[1]方法在局部及边界上颜色迁移的表现,选择另外3组图像进行实验分析,结果见图 5

图 5 本文方法与文献[1]方法结果对比图
Fig. 5 Comparison of results between our method and reference [1] ((a) origin images; (b) target images; (c) ours; (d) reference [1])

文献[1]的颜色迁移思想是在目标灰度图像和源彩色图像中匹配超像素对,并将源图像超像素平均颜色信息迁移至灰度图像超像素的中心像素点,选择这些着色的中心像素点作为微涂鸦样本(micro-scribbles),通过Levin扩散方法[2]完成着色。其中超像素匹配确保迁移颜色空间的局部一致性,其匹配的准确性决定颜色迁移结果,直接影响局部区域着色的正确性。如图 5(d)第2行胳膊周围的头发颜色、第3行桌子的颜色均出现了明显的错误着色区域。此外在利用全局优化的方法实现颜色扩散过程中也出现了颜色模糊化,如图 5(d)第1行云彩的颜色、第3行狗的腿及腹部轮廓边界均出现了颜色过界。而本文方法先匹配颜色类别,在同类别中寻找待匹配像素,结合置信度值,确保了局部区域内颜色空间的一致性和匹配的正确性,在迁移过程中,通过自适应权重均值滤波避免出现颜色的模糊和过界,从图 5(c)看,在局部和边界区域其彩色化结果均优于文献[1]结果。

3 结语

本文提出一种局部自适应的灰度图像彩色化方法。该方法在匹配的颜色类别基础上,利用纹理特征确定源图像和目标图像的像素匹配点对,并通过可变参数的均值滤波实现匹配像素点的局部自适应的颜色迁移。实验结果表明,本文方法可以获取颜色空间一致性较高且整体平滑的结果,与现有的典型灰度图像彩色化方法相比,能有效抑制边界区域颜色的过渡平滑和模糊,得到更自然的彩色化结果图。然而与近年来深度学习等全自动化方法相比,本文方法中部分参数仍需要人工干预,因此,对于参数的选择和自动化设定是下一步关注和研究的内容。

参考文献

  • [1] Gupta R K, Chia A Y S, Rajan D, et al. Image colorization using similar images[C]//Proceedings of the 20th ACM International Conference on Multimedia. Nara, Japan: ACM, 2012: 369-378.[DOI: 10.1145/2393347.2393402]
  • [2] Levin A, Lischinski D, Weiss Y. Colorization using optimization[J]. ACM Transactions on Graphics, 2004, 23(3): 689–698. [DOI:10.1145/1015706.1015780]
  • [3] Chen Y, Wang S Z. Image colorization without edge crossover of colors based on anisotropic diffusion[J]. Journal of Electronics & Information Technology, 2014, 36(2): 271–276. [陈颖, 王朔中. 避免颜色越界的各向异性扩散图像彩色化[J]. 电子与信息学报, 2014, 36(2): 271–276. ] [DOI:10.3724/SP.J.1146.2013.00558]
  • [4] Huang Y C, Tung Y S, Chen J C, et al. An adaptive edge detection based colorization algorithm and its applications[C]//Proceedings of the 13th Annual ACM International Conference on Multimedia. Hilton, Singapore: ACM, 2005: 351-354.[DOI: 10.1145/1101149.1101223]
  • [5] Horiuchi T, Kotera H. Colorization algorithm for monochrome video by sowing color seeds[J]. Journal of Imaging Science and Technology, 2006, 50(3): 243–250. [DOI:10.2352/J.ImagingSci.Technol.(2006)50:3(243)]
  • [6] Li X J, Zhao H L, Huang H. Local linear model optimization based grayscale image colorization[J]. Journal of Image and Graphics, 2013, 18(4): 460–466. [厉旭杰, 赵汉理, 黄辉. 局部线性模型优化的灰度图像彩色化[J]. 中国图象图形学报, 2013, 18(4): 460–466. ] [DOI:10.11834/jig.20130414]
  • [7] Irony R, Cohen-Or D, Lischinski D. Colorization by example[C]//Proceedings of the 16th Eurographics Conference on Rendering Techniques. Konstanz, Germany: ACM, 2005: 201-210.[DOI: 10.2312/EGWR/EGSR05/201-210]
  • [8] Kuzovkin D, Chamaret C, Pouli T. Descriptor-based image colorization and regularization[C]//Proceedings of the 5th International Workshop on Computational Color Imaging. Saint Etienne, France: Springer, 2015: 59-68.[DOI: 10.1007/978-3-319-15979-9_6]
  • [9] Zhang X, Zhao R Z, Cen Y G, et al. Color reconstruction algorithm for grayscale images based on dictionary learning and sparse representation[J]. Journal of Computer-Aided Design & Computer Graphics, 2014, 26(7): 1092–1098, 1108. [张勋, 赵瑞珍, 岑翼刚, 等. 基于字典学习与稀疏表示的灰度图像颜色重建算法[J]. 计算机辅助设计与图形学学报, 2014, 26(7): 1092–1098, 1108. ]
  • [10] Welsh T, Ashikhmin M, Mueller K. Transferring color to greyscale images[J]. ACM Transactions on Graphics (TOG), 2002, 21(3): 277–280. [DOI:10.1145/566654.566576]
  • [11] Liu S G, Zhang X. Automatic grayscale image colorization using histogram regression[J]. Pattern Recognition Letters, 2012, 33(13): 1673–1678. [DOI:10.1016/j.patrec.2012.06.001]
  • [12] Charpiat G, Hofmann M, Schölkopf B. Automatic image colorization via multimodal predictions[C]//Proceedings of the 10th European Conference on Computer Vision: Part Ⅲ. Marseille, France: ACM, 2008: 126-139.[DOI: 10.1007/978-3-540-88690-7_10]
  • [13] Larsson G, Maire M, Shakhnarovich G. Learning representations for automatic colorization[C]//Proceedings of the 14th European Conference on Computer Vision. Amsterdam, The Netherlands: Springer, 2016: 577-593.[DOI: 10.1007/978-3-319-46493-0_35]
  • [14] Zhang R C, Isola P, Efros A A. Colorful image colorization[C]//Proceedings of the 14th European Conference on Computer Vision. Amsterdam, The Netherlands: Springer, 2016: 649-666.[DOI: 10.1007/978-3-319-46487-9_40]
  • [15] Kim K S, Zhang D N, Kang M C, et al. Improved simple linear iterative clustering superpixels[C]//Proceedings of 2013 IEEE International Symposium on Consumer Electronics. Hsinchu, Taiwan, China: IEEE, 2013: 259-260.[DOI: 10.1109/ISCE.2013.6570216]
  • [16] Henz B, Limberger F A, Oliveira M M. Independent color-channel adjustment for seamless cloning based on Laplacian-membrane modulation[J]. Computers & Graphics, 2016, 57: 46–54. [DOI:10.1016/j.cag.2016.03.004]