Spektrale Charakterisierung von Farbbildsensoren zur anwendungsspezifischen Kameraparametrierung (English shop)

Preview

Table of Contents, Datei (33 KB)
Extract, Datei (180 KB)

Determining the spectral sensitivity of a colour image acquisition system, for example a digital camera, is an important prerequisite for the model-based adaptation of colour correction to the illumination conditions in specific applications, or for generating device profiles for a colour management system.

In this work, a method for the indirect determination of spectral sensitivity was developed, implemented in practice and investigated for the first time, based on a small set of colour samples selected by means of optimization methods. Compared with a direct method, the indirect method has the advantages that the broadband colour samples used result in a comparatively higher radiant power at the image sensor, that these samples correspond more closely to the conditions encountered in practical use, and that the method is more cost-effective to implement. An essential contribution to the improvement of the indirect measurement method lies in the new method for the automatic selection of the colour samples best suited to the procedure. This selection approach, termed the Lagrange multiplier method, is based on an analytical examination of the quadratic optimization problem and provides an optimal sample set for the chosen measurement task. The results of the indirect method were compared with those of a direct method. The two results exhibit a very high degree of similarity and can be regarded as identical within the limits of measurement uncertainty. Further confirmation of the quality of the results was provided by a comparison based on a test colour sample set. The results presented qualify the indirect measurement method developed for use in everyday practice. In this context, the additional advantage of a low calibration effort could be demonstrated.

As a further key aspect, new methods for the model-based optimization of colour correction transformations were developed and compared with previous approaches from the literature. Using the indirectly determined sensitivities, the spectral model applied permits a high degree of flexibility in taking into account real light sources, defined standard illuminants, test charts and measured real as well as generated virtual object spectra, in order to optimally adapt colour reproduction to particular application conditions. In order to optimize not only colour fidelity but also the effect of a transformation on the signal-to-noise ratio in the image data, a new approach for integrating a perceptual evaluation of noise into the quadratic optimization was presented. Furthermore, a new method for optimal colour correction for multiple and varying acquisition illuminations, termed the triple matrix, was introduced. For the dynamic adaptation of colour correction, i.e. an automatic white balance, the triple matrix can be used in direct combination with various published colour constancy algorithms. The practical applicability of the methods developed was demonstrated, for example, using a dermatoscopy system.