Retaining Color Fidelity in Photo CD Image Migration

The Photo CD system, originally aimed at consumers, has also been used by cultural institutions to store digital images from scanned photographic film, prints, and documents. With improvements in technology and anticipated adoption of new digital image standards, e.g., JPEG 2000, retention of image fidelity during migration of digital collections is often a concern. In a previous report (Archiving Conf. 2005), we described the format and colorimetric definition of images stored on Photo CD disks. Here we address several steps that can be taken to ensure the retention of the original subject color information, and the straightforward display and exchange of derivative images. The rationale and practical considerations for using embedded ICC color profiles will also be discussed. Introduction The Photo CD system was developed to support the capture and display of images from a wide variety of imaging media and devices. This was accomplished using a unique color-encoding and image file format. Originally aimed at consumers and professional photographers, the system was also used by cultural institutions to store images from scanned material. With the development of, and migration to, new digital imaging standards, retention of both spatial and color image fidelity is of concern to many institutions. In this paper we address several steps that can be taken for successful reformatting. Here our focus is on the retention of color fidelity during image migration. The Photo CD color-encoding specification, which is consistent with color standards for imaging devices, was designed to provide high quality images on color displays without requiring special adjustments or extensive image processing. The Photo CD Image Pac file format involves storing the digital image at several sampling resolutions. This facilitates image viewing and printing using a wide range of devices. The Image Pac elements for about one hundred digital images are stored on a single Photo CD compact disk. PhotoYCC Color-Encoding Specification The PhotoYCC color-encoding specification is used to express the original-subject colorimetry in terms of RGB exposure factor values. These values represent those that would have been produced by a specified reference image-capture device, had it captured the same subject, illuminated by CIE Standard Illuminant D65. The defined red, green and blue spectral responsivities of the reference image-capture device were chosen to be equivalent to the color-matching functions corresponding to the reference primaries specified for ITU Recommendation ITU-R BT.709. Signal values from practical video cameras that conform to this standard produce all-positive signals. However, since the reference capture device is defined by its theoretical characteristics, it can form negative signal values. The PhotoYCC specification accommodates this situation, which corresponds to the capture of original-subject color information outside the color gamut of a video display. In addition to this extended color-gamut, the PhotoYCC color-encoding specification deals with the extensive range of luminance information encountered in much original-subject matter and captured by input photographic media. This includes specular hightlights and image areas more highly illuminated than the main subject. The color encoding accommodates luminancefactor values up to two times that produced by a perfect white reflector in the principal subject area. In summary, PhotoYCC-encoded image data represent original-subject color image information in a way that accommodates a wide range of scene exposures and allows rapid display of high-quality photographic images. A way to preserve this input-referred color information, extended dynamic range and color gamut is to migrate to the RIMM color space. Before discussing recommendations for such a migration, we review the color transformation equations that define the relationship between stored PhotoYCC image data values and the RIMM R,G,B (tristimulus) values. Objectives for Conversion A color encoding for conversion of Photo CD image files amenable to objectives for archiving and preservation should ideally retain the salient original-subject nature and extensive color gamut and luminance dynamic range characteristics of the extant image files. Additionally, such a conversion should allow for straightforward production of appropriately rendered images in a variety of modalities without imposing any loss of color information to the converted image data. Moreover, it would be convenient if such a conversion could be accomplished using an existing color-management infrastructure. The Eastman Kodak Company developed the Reference Input Medium Metric RGB (RIMM RGB) color encoding to be used with digital images expressed in an unrendered, or scene-referred, image state. As with PhotoYCC, the RIMM encoding provides a color gamut encompassing virtually all real-world surface colors encountered in typical imaging applications and supports a luminance range sufficient to include critical specular-highlight information. RIMM is also well suited as an input space to the ICC profile connection space (PCS), both of which have identically defined adaptive white point chromaticities. RIMM RGB is applicable for numerous imaging functions including the archiving and interchanging of unrendered-state images. The 8-bit version of this color encoding is particularly well suited for conversions of legacy PhotoYCC images with minimal loss. Definition of RIMM RGB Color Encoding The reference viewing conditions associated with RIMM RGB are consistent with those of typical original subject matter viewing environments, Proc. Archiving 2006 Conf., IS&T, 2006 • Luminance level >1,600 cd/m. • Average image surround (i.e. area surrounding the scene is similar in luminance and chromaticity to that of the overall scene). • No viewing flare. • Aadaptive white-point chromaticities corresponding to those of CIE Standard Illuminant D50 (x = 0.3457, y = 0.3585). RIMM RGB color values are expressed in terms of flareless colorimetric measurements referenced to a set of RGB primaries capable of expressing virtually all real-world surface colors using all-positive exposure-factor values. The reference primaries and the white produced by their additive combination are defined by their CIE x, y chromaticity values given in Table 1. Table 1: RIMM RGB reference primaries and white point chromaticities Color x y Red 0.7347 0.2653 Green 0.1596 0.8404 Blue 0.0366 0.0001 White 0.3457 0.3585 General Architecture for Conversion The architecture shown in Fig. 1 accomplishes these objectives within an ICC-based color-management framework. The required color-signal processing is embodied in two distinct types of profiles. The first step (unpack) involves the assembly of the multi-resolution pyramid elements of each digital image into a single image array. This would usually result in a (2048 lines × 3072 pixels x 3 colors) array for 16Base Photo CD image. Photo CD file PhotoYCC array YCC to RIMM unpack RIMM to PCS Apply ICC Device-Link profile Assign ICC profile RIMM8 with Embedded profile Figure 1: Transformation from Photo CD image file to RIMM A device-link profile, which defines the relationship between the PhotoYCC and RIMM color encoding specifications, executes the desired data metric conversion, retaining the salient PhotoYCC original-subject nature of the image data. The use of a device-link profile is necessary to retain PhotoYCC’s extended luminance dynamic range and color gamut in the RIMM RGB encoding. This information would be irretrievably lost if the conversion were embodied in a composite profile produced by concatenation of an input PhotoYCC-to-PCS profile and an output PCS-to-RIMM profile. The problem results when the image data passes through the International Color Consortium (ICC) Profile Connection Space (PCS). The PCS comprises a color encoding specification of colors associated with rendered images to be viewed in a particular reproduced-image viewing environment. While the PCS was intended to be a color space where input and output device profiles could be joined to form composite input-to-output color transforms, it does not deal well with the extended nature of the encodings of interest in this paper. Use of a device-link profile avoids passing through the PCS, yet can be applied in an ICCbased color-management environment. In our case, the PCS is recommended in the second profile of the conversion process, due to the different purpose of that profile. An assigned input profile, embedded in the image file itself, accomplishes the rendering of RIMM-encoded original-subject color values to ICC PCS values appropriate for display on imaging devices and media viewed in average-surround conditions. This rendering process entails necessary increases in the luminance contrast and chroma of the original subject colors, as well as imposing limits on the luminance dynamic range and color gamut dictated by the selected output medium or device capabilities. By relegating this lossy process to an assigned profile, the integrity of the stored image data is preserved and the rendering is executed on demand. The color-signal processing embodied in the device-link profile describing the relationship between PhotoYCC and RIMM is defined by the following transformation sequence. Transformation of PhotoYCC values to originalsubject RIMM tristimulus values In the first step of the transformation, luma and chroma values are computed from PhotoYCC Y, C1, and C2 digital color values. For 24-bit (8-bits per color channel) encoding, luma and chroma values are computed according to the following equations, Y Luma 255 402 . 1 = ( ) 40 . 114 156 1 1 − = C Chroma (1)     Z ( ) 64 . 135 137 2 2 − = C Chroma . The resulting Luma, Chroma1, and Chroma2 values are converted to nonlinear values, , using the following matrix transformation, 709 ' ' ' B G R