Premium
Color notation conversion by neural networks
Author(s) -
Tominaga Shoji
Publication year - 1993
Publication title -
color research and application
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.393
H-Index - 62
eISSN - 1520-6378
pISSN - 0361-2317
DOI - 10.1002/col.5080180408
Subject(s) - computer science , artificial neural network , backpropagation , notation , weighting , artificial intelligence , feed forward , nonlinear system , interpolation (computer graphics) , color space , feedforward neural network , pattern recognition (psychology) , algorithm , arithmetic , mathematics , image (mathematics) , medicine , physics , quantum mechanics , control engineering , engineering , radiology
This article describes a new method for color‐notation conversion between the Munsell and CIE color systems by means of neural networks. A multilayer feedforward network is regarded as a nonlinear transformer which is trained to learn a mapping between the two color spaces. We adopt the backpropagation learning rule for the training. The mapping is then realized in a simple network architecture in which nonlinear units are linked in parallel and in layers. The tables of data by Newhall, Nickerson, and Judd are used in computer simulations of training and testing phases. We determine an effective network structure and training strategy based on experiments under different conditions. The conversion accuracy is demonstrated in both directions, Munsell‐to‐CIE and CIE‐to‐Munsell. In the neural network method, it is not necessary to use a special database. The knowledge of the mapping is stored in a small set of weighting parameters in the network. We point out that the proposed method for color‐notation conversion provides several potential advantages over the conventional approach based on interpolation of the data tables.