Colour in Context
Research group
Computer Vision Center


TSE Colour-Naming Procedure

Robert Benavente and Maria Vanrell



The TSE colour-naming model

A full parametric model has been defined on the CIE Lab space. On this model, each one of the 11 basic colour categories (i.e. White, Black, Red, Green, Yellow, Blue, Brown, Purple, Pink, Orange and Grey) is modelled with a fuzzy set characterized by a combination of sigmoids as membership function. Hence, for any given image pixel, the memberships to the 11 basic categories can be obtained and used by a decision function to assign a linguistic laber to the pixel.

The model has several applications such as segmentation, automatic image description, retrieval, tracking and human-machine interaction.

Plot of the membership values of the TSE model in the 6 Lightness planes  
Example of color naming on real images - Original Image  
Example of color naming on real images - Annotated Image

BLUE
GREEN
PINK
RED

 

Example of color naming on real images - Original image Example of color naming on real images - Green Example of color naming on real images - Pink Example of color naming on real images - White
Original image
Green
Pink
White
Example of color naming on real images - Original image Example of color naming on real images - Red Example of color naming on real images - Orange Example of color naming on real images - Brown
Labelled image
Red
Orange
Brown

 

In order to refine the proposed model, we have done psychophysical experiments to accurately define the inter-colour regions of the colour-naming space. These experiments are done in our Psychophysics Lab in fully-controlled conditions. The goal is to precisely define the boundaries between basic colours for a mean English speaker.

Results of a psychophysical experiment on color naming

 


Matlab code for colour naming

An version of the color naming software for matlab can be found here. The code consist out of two functions:

'SampleColorNaming' applies the TSE model to a single sRGB value. It returns the membership values to the 11 basic color categories and the color term corresponding to the highest membership value.

'ImColorNaming' applies the TSE model to a sRGB image. It returns a matrix with the membership values to the 11 basic color categories for every pixel in the image, and an image with each pixel labelled with the representative color corresponding to the category with the highest membership value at the pixel.


References

Robert Benavente, Maria Vanrell, and Ramon Baldrich. Parametric fuzzy sets for automatic color naming. Journal of the Optical Society of America A, 31(1): 48-56, 2008.

C. Alejandro Párraga, Robert Benavente, Maria Vanrell, and Ramon Baldrich. Psychophysical measurements to model inter-colour regions of colour-naming space. Journal of Imaging Science and Technology, 53(3):031106-(8), 2009.

 

  2008 Colour in context Group | Computer Vision Center. All rights reserved | Contact webmaster |  Last updated: Monday 11 May 2009     eXTReMe Tracker