Recovering spectral data from digital prints with an RGB camera using multi-exposure method

The spectral data of a surface can be reconstructed from RGB camera response when reflectance spectrum of the surface and radiance spectrum of the light source are smooth enough. One factor that lowers the reconstruction performance of consumer level cameras is limited dynamic range. Details or intensity levels in low and high luminance regions can not be reliably detected from a single image. With multiple exposures it is possible to capture high dynamic range images. The aim of this study was to apply a multi-exposure method for spectral data reconstruction of prints and find how accuracy it is. Our intended application area is printed image digitization for image quality calculations. We measured the intensity levels and recovered spectral data of printed samples by taking two images using different exposure times. Based on the results multi-exposure method improves the accuracy of spectral data reconstruction of print samples compared to traditional methods and is the preferred choice for printed image digitization process. Introduction Spectral data provide the most useful information for color reproduction measurements. The spectral data of a sample can be measured by spectroradiometer. Spectroradiometer measurements are precise but time-consuming and point-wise. Point-wise measurements are not applicable for some applications. For example spatial variations of spectral data in a natural scene cannot be measured using a point-wise device. Respectively spectral information of a printed photograph is difficult to measure using a point-wise device. Our intended application area is printed image digitization. This digitization process has been developed for our printed image quality calculation system [1]. The idea in our printed image quality calculation system is to digitize the printed image and calculate the quality of printed image by comparing the features of original digital image and printed images. When printed image quality is calculated the accuracy and spatial sampling frequency of colour measurement should be high enough. Earlier we have used (RGB to XYZ) characterised digital camera for spatial sampling of printed images [1]. In some other studies [2],[3] reflective scanners have been used for spatial sampling. There are plenty of earlier studies where a RGB camera (spatially sampling device) is used as a spectral measurement device. This is possible because the spectral data of natural scenes, objects and illumination are smooth. It is a known fact that the dynamic range of consumer digital cameras is the limiting factor of their imaging performance. Often consumer digital cameras are not capable of detecting, in a single image, the entire intensity range hitting the sensor. Reliable detection is possible only if more than one exposure is used. In this paper we describe our experiments in which we investigated the performance of a multi-exposure method for spectral data reconstruction of printed samples. In previous studies the reconstruction accuracy has been improved by increasing the number of response channels using different filter combinations. Valero et. al. [4] measured that two or three filters improved significantly the reconstruction performance. In some studies the number of channels was increased by producing images under spectrally different light sources [5]. The principal component analysis (PCA) is the traditional linear method for reconstruction of the spectral data from a camera response [5] or to analyse the dimensionality of a sample spectra set [6]. Other base function methods have also been studied. Li and Berns [7] studied the performance of ICA analysis compared to the PCA. Mansouri et. al. [8] studied the performance of Fourier and Wavelet bases. The second linear method to estimate the spectral data from camera response is based on the camera’s filter functions. Camera is a linear system which can be described by the Equation (1):