Evaluation of preferred lightness rescaling methods for colour reproduction

摘要

In cross-media colour reproduction, it is common goal achieving media-relative reproduction. From the ICC specification, this often accomplished by linearly scaling XYZ data so that the media white of the source data matches that of the destination data. However, in this approach the media black points are not explicitly aligned. To compensate this problem, it is common to apply a black point compensation (BPC) procedure to improve the mapping of the black points. First, three lightness rescaling methods were chosen: linear, sigmoidal and spline. CIECAM02 was also implemented in an approach of a lightness rescaling method; simply, lightness values from results produced by CIECAM02 handle as if reproduced lightness values of an output image. With a chosen image set, above five different methods were implemented. A paired-comparison psychophysical experiment was performed to evaluate performances of the lightness rescaling methods. In most images, the Adobe's BPC, linear and Spline lightness rescaling methods are preferred over the CIECAM02 and sigmoidal lightness rescaling methods. The confidence interval for the single image set is ±0.36. With this confidence interval, it is difficult to conclude the Adobe BPC method works better, but not significantly so. However, for the overall results, as every single observation is independent to each other, the result was presented with the confidence interval of ±0.0763. Based on the overall result, the Adobe's BPC method performs best.