Spectral Based Color Image Editing (SBCIE)

Spectral based color image editing systems include methods for generating spectral representations of surfaces and illuminants from input signals of image capture devices (such as digital cameras and scanners), editing spectral representations of surfaces and illuminants in a scene, and transforming spectral representations of surfaces and illuminants to output signals of image devices (such as CRT and LCD displays and color printers). Introduction Computer-assisted color image editing systems include methods for image capture, manipulation and reproduction. Until recently, color image editing systems were commercially available only as part of large graphics systems. Today, such systems are widely available in desktop publishing software or embedded in drivers for digital cameras, scanners and printers, providing users with the means to 1) generate digital representations of color images, 2) edit digital representations to make the image look visually appealing, and 3) reproduce digital representations on different media. Whether they are part of an elaborate publishing system or a software driver for imaging peripherals, all color image editing systems are dependent on and limited by devices that either capture or generate color images. Most image capture devices have only three color sensors and therefore compress a multidimensional spectral signal into the output of three color channels -usually, R, G and B. Most color image editing systems convert the device RGB signals into a human-based representation, such as CIE XYZ, and operate on this representation. For example, Neugebauer [1] and Schreiber [2] described color image editing systems based on converting devicedependent image representations, such as scanner or display RGB values, into human-based tristimulus signals, such as XYZ. These color image editing systems provide a means of changing and displaying the human-based tristimulus signals to enable interactive editorial color corrections. The objective of these systems is to visualize or simulate enhancements to the printed output before the actual image is printed. This paper presents a different approach to color image editing. Rather than convert device-dependent signals or vectors into human-based vectors based on the CIE XYZ color matching functions, as is traditionally performed in today’s color image editing systems, it is possible to use device-dependent sensor responses to estimate deviceindependent spectral reflectances of surfaces and illuminants in the captured and/or rendered scene. To illustrate this approach, I describe several methods for generating spectral representations of surfaces and illuminants in a scene. I introduce the concept of spectral-based color image editing (SBCIE) systems that embody methods to generate, edit and reproduce spectral representations of surfaces and illuminants based on the analysis of digital images generated by scanners, digital cameras and displays. Background In this section, I introduce the linear algebraic notation for spectral representations of illuminants and surfaces used throughout this paper. Then I describe the more practical low-dimensional linear spectral representations that can be used to approximate spectral representations. These low-dimensional linear models are necessary because most often we do not have complete spectral information about the illuminants and surfaces in a scene. I describe how the accuracy of spectral representations are limited by the dimensionality of our image capture data and by our