Coefficient color constancy
Coefficient color constancy
复制标题
颜色恒常系数
DOI：
--
复制
发表时间：
1995
期刊：
影响因子：
--
通讯作者：
G. Finlayson
中科院分区：
文献类型：
--
作者：
B. Funt;G. Finlayson
颜色恒常性的目标是获取在未知光源下观察的表面的颜色响应（例如相机rgb三元组），并将它们映射到与光源无关的描述符。在现有的理论中，这种映射要么是一个一般的线性33矩阵，要么是一个简单的缩放矩阵的对角矩阵。一般理论的优点是可以准确地折现光源，但缺点是必须恢复9个参数。相反地，虽然古老的理论只有三个未知数，但对角矩阵可能只是部分地忽略了光源。本文的出发点是对实证方法的推广；目标是保留其固有的简单性，同时增加其表现力。在广义coe客户方案下，我提出在应用缩放coe客户之前，视觉系统将响应转换为新的传感器基。我提出了为各种颜色响应的统计模型选择最佳先验基础的方法。这些模型足够丰富，以至于广义先验方法适用于几乎所有可能的传感器集。为了达到颜色的恒定，必须恢复正确的颜色。现有的算法只有在满足强约束的情况下才能做到这一点。例如，通常假设每个场景中都有一个白色反射。在我论文的第二部分，我基于对世界的非常弱（和非常合理）的假设，开发了一种新的高效算法，我称之为透视颜色。我只假设不同反射引起的颜色反应的范围随着光照的变化而变化，并且光照本身只能在一定范围内变化。我在用彩色摄像机拍摄的真实图像上测试了该算法，结果显示，它的稳定性非常好。事实上，这种稳定的程度比理论上可能的最好的程度更有利。本文开发的方法可以应用于各种其他领域，包括色彩图形，色彩再现和色彩外观模型。我首先要感谢我的论文导师Brian V. Funt博士。他一直非常支持我，一开始就指导我阅读相关文献，在每个阶段提供建设性的批评，并全程指导我研究的艺术。在学术方面，我还要感谢很多其他人。我特别感谢马克·德鲁博士，我曾与他合作过许多项目。德鲁博士有着令人羡慕的数学头脑，他耐心地指导我完成了许多数学结果。还要感谢Binay Bhattacharya博士向我介绍了计算几何（我论文的关键组成部分），Michael H. Brill博士与我分享了他在色彩科学方面的百科全书式知识。李泽年和Jacques Vaisey的论文指导。继续去喝茶和喝咖啡：桑吉夫·马哈詹（他开创了这一切）、苏米特·巴瓦、皮纳基·米特拉、希拉·卡彭德尔和苏博·查特吉都是很好的伙伴。在足球场上，我要感谢Pandemonium的队友们；这些人包括雅克·瓦塞、彼得·库邦、让·瓦拉尔迪、沃伦·摩尔斯和埃尔德伊酋长塔马斯。在实验室的帮助和支持：Brigitte Dorner， Kobus Barnard和Janet Dueck是一流的。在众多和各种课外活动的合作伙伴包括T. Pattabhiraman， Glenn Macdonald和Stephan Wehner。感谢塔米·拉伯格请求得到认可。我特别感谢（再次）我来到加拿大后认识的三个好朋友。这些人包括马克•梅佐芬尼、艾伦•班纳特-布朗和吉尔斯•迪翁。最后衷心感谢我远在苏格兰的家人。通常的做法是，一篇论文应该是线性组织的：从介绍到背景，理论结果，实验和结论的顺利进展。这篇论文不是这样组织的。因此，我觉得有必要评论一下我论文的布局。像许多工作一样，我的论文包含的不是一个想法，而是很多想法。虽然这些都是相关的，但它们各自都很有趣和重要。然而，如果我遵循通常的做法，我将在创建线性整体时剥离个人身份。具体来说，每个部分将同时涵盖所有的想法。背景部分将是许多观点的背景材料的结合，理论结果是推导的结合，实验结果是实验的结合。在这篇论文中，我采取了更横向的方法，并单独提出每个想法。具体来说（除了引言和结论），每一章都论述一个主要思想。本文只介绍了与这个想法相关的背景、理论推导和实验。因此，每一个想法都是放在它被调查的背景下，每一章都是完全独立的。此外，由于后面的章节是建立在前面章节的基础上的，所以论文仍然是线性的，尽管在一种不同的（我认为更合适的）意义上。当然，人们可以对这两种组织战略的优点进行争论。例如，在线性模型中，描述是奇异的；在这里，如果它们与一个以上的想法相关，它们必然是重复的。因此，线性命题更加紧凑。然而，我认为在横向论文中少量的重复是值得的，并构成了思想的强化；本质上是让事情变得更清晰。vi此外，手头主题的背景材料总是可以在当地找到；没有必要在前面的章节中进行令人厌烦的搜索。我必须承认横向组织战略不是我自己的主意。相反，直到我阅读了Lucassen的论文[Luc93]，我才完全准备好坚持线性方法的限制。在阅读这篇论文时，我对阐述的清晰印象深刻，我觉得这部分是由于观点的单独表达。七世
The goal of color constancy is to take the color responses (for example camera rgb triplets) of surfaces viewed under an unknown illuminant and map them to illuminant independent descriptors. In existing theories this mapping is either a general linear 3 3 matrix or a simple diagonal matrix of scaling coe cients. The general theories have the advantage that the illuminant can be accurately discounted but have the disadvantage that nine parameters must be recovered. Conversely while the coe cient theories have only three unknowns, a diagonal matrix may only partially discount the illuminant. My staring point in this thesis is to generalize the coe cient approach; the goal is to retain its inherent simplicity while at the same time increasing its expressive power. Under the generalized coe cient scheme, I propose that a visual system transforms responses to a new sensor basis before applying the scaling coe cients. I present methods for choosing the best coe cient basis for a variety of statistical models of color responses. These models are rich enough that the generalized coe cient approach su ces for almost all possible sensor sets. To achieve color constancy the correct coe cients must be recovered. Existing algorithms can do so only when strong constraints are satis ed. For example it is often assumed that there is a white re ectance in every scene. In the second part of my thesis, I develop a new coe cient algorithm, which I call color in perspective, based on very weak (and very reasonable) assumptions about the world. I assume only that the range of color responses induced by di erent re ectances varies with a change in illumination and that illumination itself can vary only within certain bounds. I tested the algorithm on real images taken with a color video camera|extremely good iii constancy is delivered. Indeed the degree of constancy compares favorably with the best which is theoretically possible. The methods developed in this thesis can be applied to a variety of other areas including color graphics, color reproduction and color appearance models. iv Acknowledgements My rst and foremost thanks go to my thesis supervisor Dr. Brian V. Funt. He has been exceptionally supportive, directing me to relevant literature in the beginning, providing constructive criticism at every stage and tutoring me in the art of research throughout. On an academic front I have many other people to thank. I am particularly grateful to Dr. Mark Drew with whom I have collaborated on many projects. Dr. Drew has enviable mathematical acumen and he patiently guided me through many mathematical results. Thanks also go to Dr. Binay Bhattacharya for introducing me to computational geometry (a key component of my thesis), Dr. Michael H. Brill for sharing his encyclopedic knowledge of color science with me and to Drs. Ze-Nian Li and Jacques Vaisey for their thesis supervision. For continued trips for tea and co ee: Sanjeev Mahajan (he started it all o ), Sumeet Bawa, Pinaki Mitra, Sheelagh Carpendale and Subho Chatterjee have provided excellent company. On the football (or soccer) pitch thanks go to my fellow Pandemonium team-mates; these include Jacques Vaisey, Petr Kubon, Jean Varaldi, Warren Moors and Tamas \the chief" Erdelyi. For help and support in the lab: Brigitte Dorner, Kobus Barnard and Janet Dueck have been rst class. Partners on numerous and various extra curricular activities include T. Pattabhiraman, Glenn Macdonald and Stephan Wehner. For asking to be acknowledged thanks go to Tammy Laberge. I am especially grateful (again) to three good friends who I have known since arriving in Canada. These comprise Mark Mezofenyi, Allan Bennet-Brown and Gilles Dionne. A nal and heartfelt thanks go to my family all those miles away in Scotland. v Foreword: Thesis Organization Common practice dictates that a thesis should be organized linearly: a smooth progression from introduction, to background, theoretical results, experiments and conclusion. This thesis is not organized in this manner. As such, I feel compelled to comment on the layout of my thesis. Like many bodies of work my thesis encompasses not a single idea but many. While these are all related they are each individually interesting and important. However if I followed common practice I would strip away individual identity in creating the linear whole. Speci cally each section would cover all ideas at the same time. The background section would be the union of background material for the many ideas, the theoretical results the union of derivations, and the experimental results the union of experiments. In this thesis I take a more lateral approach and present each idea by itself. Specifically (excepting the introduction and conclusion) each chapter tackles a single major idea. Only the background, theoretical derivations and experiments relevant to the idea at hand are presented. As such each idea is placed in the context in which it was investigated and each chapter is completely self contained. Moreover, because later chapters build on earlier ones the thesis is still linear though in a di erent (and I would suggest more appropriate) sense. Of course one can argue the merits for either organizational strategy. For example in the linear model de nitions are presented singularly; here they are necessarily duplicated if they are germane to more than one idea. Thus the linear thesis is more compact. However I would contend that the small degree of repetition in lateral theses is worthwhile and constitutes reinforcement of ideas; in essence making things clearer. vi Moreover background material for the subject at hand can always be found locally; there is no need for a tiresome search through earlier chapters. I have to confess that the lateral oragnizational strategy was not my own idea. Rather I was fully prepared to adhere to the strictures of the linear approach until I read the thesis of Lucassen[Luc93]. In reading that thesis I was impressed at the clarity of exposition and this, I feel, was in part due to the separate presentation of ideas. vii
登录
查看更多内容
DOI：
10.1364/josaa.9.001905
发表时间：
1992-11
期刊：
Journal of the Optical Society of America. A, Optics and image science
影响因子：
--
作者：
D. Marimont;B. Wandell
通讯作者：
D. Marimont;B. Wandell
DOI：
10.1364/josaa.3.000029
发表时间：
1986-01-01
影响因子：
1.9
作者：
MALONEY, LT;WANDELL, BA
通讯作者：
WANDELL, BA
DOI：
10.1364/josaa.3.001651
发表时间：
1986-10-01
影响因子：
1.9
作者：
BRAINARD, DH;WANDELL, BA
通讯作者：
WANDELL, BA
DOI：
10.1364/josaa.3.001662
发表时间：
1986-10-01
影响因子：
1.9
作者：
DZMURA, M;LENNIE, P
通讯作者：
LENNIE, P
任务相关的颜色辨别。
DOI：
10.1364/josaa.7.000776
发表时间：
1990
期刊：
Journal of the Optical Society of America. A, Optics and image science
影响因子：
--
作者：
Poirson,AB;Wandell,BA
通讯作者：
Wandell,BA