Perceptual Significance Hierarchy: A Co uter Vision Theory for Color Se

A Perceptual Significance Hierarchy (PSH) for line art images is developed which represents the relative perceptual significance of each image component. This is possible through the use of a set of image-features which are used by the human visual system. The PSH and related rho-space computer vrsion algorithms can be used to automate the fake color separation process used by the printing industry. This is accomplished by adding rudimentary visual processing capabilities to a computer graphics system. This paper describes an application of Artificial Intelligence techniques to a pre-press problem in the printing industry, color separation. This application area is interesting as it is one where expert systems techniques are not useful, where rulebased reasoning is inappropriate, and where relational knowledge bases make no sense. Instead, AI techniques based on basic visual perceptual computations and the parallel processing of visual information are required. This can be accomplished through the use of a Perceptual Significance Hierarchy (PSH), as described below. A. The Color Separation Application The printing industry is rapidly becoming an entirely electronic computer-based industry. However not all of the prepress processes have been successfully computerized. Color separation is one such process. Before a colored image can be printed, it must go through the color separation process, which creates three or four separate plates one each for printing cyan, magenta, yellow, and if required, black. There are two types of color separation: process and fake. Process color separation is used for photographic images, and has been successfully automated by using colored filters to optically separate the colors projected from the negative of a color photograph. An example of images printed from process color separation are the color photographs in weekly news magazines. Fake color separation is used for line-art images such as the Sunday comics or commercial art. In this case, the printing company receives black and white art-work, and information about the color for each image region. The task is then to color in, by number, each region in the image. Despite the simple concept, this is a difficult problem. At present there are two techniques for fake color separation: one is completely manual; and the other uses computer graphics techniques. In the manual technique, for each particular desired color in the image, a sheet of transparent are tate is laid on top of the original line drawing art, and blochs of translucent red cellophane tape are laid over each region to be colored the particular color. The red tape is then cut to the exact shape of the areas to be colored with an Xacto knife, and the excess tape peeled away. The acetate sheets are kept aligned with each other and with the original artwork by punching holes in their margins which fit over precisely aligned registration pins. This labor intensive manual color separation is still widely used. In the latter technique the artwork is digitized and displayed on a computer graphics workstation. The user can specify a particular color using a color palette, and then can indicate which region should be filled with that color by pointing to the region with the cursor. A seed-fill algorithm can then be used to fill the desired region with the selected color [l]. After th e user has interactively colored in all of the image, the graphics system can compute the three or four required printing plates. There are several basic problems with the computer graphics based techniques, as illustrated by the simple line drawing in Figure 1: 1) First, extra lines. in a region can cause problems. For example, if the bottom of the container was to be colored blue, then the user would have to separately select and fill each of the Figure 1 Figure 2 I 1 r top level