Accurate Estimation of the Non-Linearity of Input-Output Response for Color Digital Cameras

Many color digital camera systems exhibit a non-linearity between the input intensity and the output response of the color channels. Although the sensor (CCD) material responds to light intensity in a linear way a non-linearity is often added by the camera manufacturer. Recent research has highlighted that in order to be able to estimate this nonlinearity (sometimes termed gamma) it is necessary to know the spectral sensitivities of each of the color channels. The input to the non-linearity for a channel is the raw channel response and the output is the processed channel response and this is inconsistent with the commonly held assumption that the input-output non-linearity and the camera spectral sensitivities curves can be estimated independently. This research employed a computational model of a camera system with known channel spectral sensitivities and nonlinear response. Camera output values were computed for a range of surfaces (defined by spectral reflectance factors) and illumination (defined by spectral power distributions). Different techniques to estimate the non-linearity of the channels were evaluated. These techniques included methods based upon the Luminance and mean reflectance of a set of neutral samples and methods based upon a knowledge (or estimation) of the spectral sensitivities of the channels. The results showed that the difference between the channel-sensitivity-based estimates of the non-linearity and the Luminanceor reflectance-based estimates were small. However, the effect was significant when the linearization was used as part of a device-characterization process. A case study demonstrated that an accurate linearization can reduce characterization errors by approximately 10% compared with traditional techniques.