Today, we received confirmation our overview-paper on Color Constancy is accepted for publication in IEEE Transactions on Image Processing. The manuscript (written by myself, Theo Gevers and Joost van de Weijer) is titled “Computational Color Constancy: Survey and Experiments”. It conveys a survey of many recent developments and state-of-the-art methods. Several criteria are proposed that are used to assess the approaches. A taxonomy of existing algorithms is proposed and methods are separated in three groups: static methods, gamut-based methods and learning-based methods. The experimental setup is discussed including an overview of publicly available data sets. Finally, various freely available methods are evaluated on two data sets. These results will be made available on www.colorconstancy.com in due time.

Thanks to Michael Metternich for taking the pictures.

On behalf of the Vision and Color division of the OSA, David Brainard and Alex Wade organize the 2011 Spectrum Recovery Competition. The challenge is to recover the relative illuminant spectra from a set of 10 rendered visual scenes. The winners will be awarded a cash prize of $1000, and will be invited to present their method and solution at the 2011 OSA Vision Meeting in Seattle, Washington. The deadline for final submission is July 14th, 2011, and details of the competition can be found here.

In order to accumulate the knowledge on color constancy we developed over the past years, we created a website that summarizes as much as possible on this topic. This website presents a concise overview of the problem and some of its solutions. Further, a list of publicly available data sets and source-code is provided. In the future, we will add results of color constancy methods applied to some of the publicly available data sets. Researchers interested in the topic can download these results, e.g. for comparison to their own method or to develop new methods. Anyone interesting in contributing to this site is kindly requested to contact the contact person of the website. The website can be found here.

Last Friday (September 24th), I successfully defended my thesis against a committee consisting of Arnold Smeulders (as my promotor), Theo Gevers (as my co-promotor), Graham Finlayson, Jan Koenderink, Lucas van Vliet and Jan-Mark Geusebroek. Those who are interested in my work can drop me a line to request a hard copy of the thesis (for free). Digital copies can be found here. A trimmed version of the slides that I used for the “laymen-talk” prior to the opposition can be found here. Please note that the talk was in Dutch, so the slides are in dutch too (although the slides only contain a minimal amount of text).

I just received the printed version of my thesis. The title of my thesis is “Edge-driven Color Constancy”. In this thesis, created under supervision of Theo Gevers, methods and techniques are proposed to improve the accuracy of color constancy algorithms. The first objective of this thesis is to improve existing algorithms by using spatial relations between pixels rather than pixel values alone. The second objective of this thesis is to analyze existing color constancy methods, with the intention to combine the algorithms dynamically, based on image content. Finally, the third objective is to evaluate color constancy performance using a performance measure that correlates with human vision as much as possible. More information can be found in my thesis; if you would like to receive a (free) hard-copy, just drop me a line and I will make sure you receive a copy as soon as possible. The defense will be held on September 24 at 14.00 in “De Agnietenkapel“, and the official announcement can be found here. Of course, everyone is cordially invited to attend the defense and the reception afterward!

Joint work with Marcel Lucassen and Theo Gevers on texture and color emotions is accepted for publication in Color Research and Application. Part of this work is currently presented at IS&T’s European Conference on Color in Graphics, Imaging and Vision (CGIV 2010), and investigates whether texture patterns affect the (human) perception of color emotions. Results of psychophysical experiments are reported, suggesting a strong influence. These results can be of importance during the development of applications working with color emotions formulae, like affective image retrieval.

Edit: A video of the presentation at CGIV on this topic by Marcel Lucassen can be found here!

Recently, our paper on Color constancy using natural image statistics and scene semantics was accepted for publication in Transactions on Pattern Analysis and Machine Intelligence. As promised, the source-code is made available too, click here to download. Note that, for proper functions, the following external Matlab-packages are required:

• Edge-based Color Constancy, which can be found here (thanks to Joost van de Weijer).
• Weibull Fitting to Natural Image Statistics, which can be found here (thanks to Jan-Mark Geusebroek and Victoria Yanulevskaya).
• LIBSVM.

Provided all packages are located in your Matlab-path, the source-code should work properly. Any questions and suggestions can be placed in the comments-section below.

Our manuscript on Color Constancy using Natural Image Statistics and Scene Semantics is accepted for publication in IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI)! After a thorough review process, we finally received confirmation from the associate editor: “I am pleased to recommend that this manuscript is published without the need of further review.” In this paper, an extension of our CVPR-2007 paper, we analyzed existing color constancy methods and discovered that the Weibull parameterization (grain size and contrast) is related to the number of edges, amount of texture and signal-to-noise ratio of images. Furthermore, these aspects are found to have an influence on the performance of different color constancy algorithms, so based on the statistics of natural images we can select the most appropriate color constancy algorithm for every image dynamically. One of the conclusions that is drawn is that methods using 2nd-order statistics perform best, but put high demands on the information content of an image (e.g. high signal-to-noise ratio is required). The performance of pixel-based methods is in principle not as high, but especially images with significant fewer edges are better solved with pixel-based methods than higher-order statistics-based methods. A trade-off is obtained for methods using 1st-order statistics. The full paper can be found here, source code of the proposed method will be made available later.

This month, our long expected paper on Generalized Gamut Mapping appeared in IJCV. Now, the source-code is made available too. With this code, it should be possible to reproduce the results that are reported in the paper. The code is written using Matlab, and requires the Convex Programming Toolbox CVX. Note that a different implementation can be found here that includes the possibility to preprocess the image with segmentation-algorithms, but it does not include the derivative-based variants that are described in our recent IJCV-paper. Our matlab-code provides two demo-scripts: one script facilitates the learning phase of the canonical gamut, the second script attempts to estimate the color of the light source for some demo-images using a simple canonical gamut. Note that to actually reproduce the results, the canonical gamuts should be learned as specified in the paper. Furthermore, for usage in real-world situations, careful thought must be put into selecting appropriate images for the canonical gamut. The source-code can be downloaded here. Any questions and suggestions can be placed in the comments-section below.