Accurate facial skin colour representation is highly required for an increasing number of applications, such as the solution of cosmetic products, the diagnosis of cutaneous diseases, and the manufacture of soft tissue prostheses. This study presents a novel camera colour characterisation model with higher predictive accuracy for the image-based colour measurement of human skin. More specifically, a digital imaging system was developed to collect the facial images of sixty human subjects from four ethnic groups. The newly collected human facial skin colour data and a conventional colour chart were selected as the training dataset, respectively, and three general techniques (linear transformation, polynomial regression, and root-polynomial regression) were utilised to derive the characterisation model by mapping camera digital signals to CIE XYZ tristimulus values. The predictive accuracy of each model was then verified using the mean CIELAB colour difference between actual skin colour measurements and the corresponding predictions from colour images. Results showed that the best model performance was achieved when the human skin colours of real subjects were used as the training samples and first order polynomial regression was used as the mapping algorithm.