Here you can find the algorithms (Matlab files [M-Files] and the references [Ref] to the original papers) tested in the work of B. Ortiz-Jaramillo, A. Kumcu and W. Philips, "Evaluating color difference measures in images", International Conference on Quality of Multimedia Experience (QoMEX) 2016. Please reference this work if you use any of the code available in this page. To download the files, please navigate to the links in the Table with the state-of-the-art CD measures. For the login information (user name and password) please contact Benhur Ortiz Jaramillo Asli.Kumcu@UGent.be indicating the purpose of using this code.
The most well known and widely used method for comparing two homogeneous color samples is the CIEDE2000 color difference formula because of its strong agreement with human perception. However, the formula is unreliable when applied over images and its spatial extensions have shown little improvement compared with the original formula. Hence, researchers have proposed many methods intending to measure color differences (CDs) in natural scene color images. However, these existing methods have not yet been rigorously compared.
Nowadays, the color-related aspect of image difference assessment has become an active area in the research of color science and imaging technology due to its wide range of applications such as color correction, color quantization, color image similarity and retrieval, image segmentation, gamut mapping, among others. For instance, in multiview imaging, color correction is used to eliminate color inconsistencies between views. Then, the assessment of color corrected images can be used to find the color correction algorithm that produces the smallest difference in terms of color. Color image similarity and retrieval is a process where all images with similar color composition to a query image are retrieved from a database. Thus, the assessment of CDs between images is very important to obtain those images with the minimum perceived CD with respect to the query image. Gamut mapping and color quantization algorithms replace pixel colors following certain criteria to ensure a good correspondence in terms of color between an original image and its reproduction. That is, CD assessment can be used to find the quantization step size and/or the range of displayable colors to obtain the reproduction with the minimum perceived CD. Color image segmentation divides images into regions displaying homogeneous colors. Hence, a CD measure can be used to find the regions with minimum perceived CD between pixels within the same region.
Traditionally, computing CDs in images has been accomplished by using a CD formula on a pixel-by-pixel basis and then examining statistics such as mean, median or maximum. For instance, the CIEDE2000 formula can be used for computing CD in natural scene color images. However, it is well know that the use of this procedure produces big estimation errors because the CIEDE2000 formula was specifically designed for homogeneous color samples. Additionally, there is not standard procedure for computing CDs in images. In search for an adequate solution of this problem, the study of CD measures in natural scene color images is an active area because its wide range of applications such as color correction, color quantization, color image similarity and retrieval, image segmentation, gamut mapping, among others. The most well-known and widely used CD measures for natural scene color images are listed in the following Table.
CD measure |
Symbol |
Appearance model |
Type |
Spatial processing |
---|---|---|---|---|
Δ E00 |
Full reference |
No |
||
Δ ES |
CIELAB |
Full reference |
Yes |
|
Δ EH |
Reduced reference |
No |
||
cssim |
Reduced reference |
Yes |
||
Ch |
Reduced reference |
Yes |
||
Image CD measure based on image appearance models [Ref] [M-File] |
Δ EI |
Full reference |
Yes |
|
Δ EJ |
CIELAB |
Full reference |
Yes |
|
Δ EA |
Full reference |
Yes |
the observer makes the color sensation from a number of pixels and not a single pixel color.