As computer processing power soars and the cost of storage drops, Grayscale Image Processing will become even more important. Grayscale Image Processing presents new challenges to software developers and opportunities for new applications for imaging.
Most image processing is performed on 1-bit black and white scans. 1-bit scans are compact. They can be sent across networks quickly and processed quickly. But the quality of a 1-bit scan depends on how well the scanner is set up and the quality of the original image. Black and white scans can be hard to look at on image viewers.
Grayscale images provide a number of advantages over 1-bit black and white scans. The first and most obvious is that there is much more information in an 8-bit scan. The extra information can be used to improve optical character recognition and handwriting recognition. Grayscale images are easier to read. It is true for both humans and computers.
Grayscale images make non-text images look better. Photos and artwork are much better looking in grayscale. This is important in applications that need to store photographic data. Engineering and scientific applications require grayscale scans. Some of these applications go beyond the need for basic 8-bit scans. For these applications 12-bit grayscale images, more than 4,000 shades of gray, are not uncommon.
Image processing takes on a new dimension of complexity with grayscale imaging. How should noise be cleaned up from a grayscale image? The methods used to make 1-bit scans are inadequate. The usual practice is to establish a simple threshold. If a pixel is darker than the threshold the the 1-bit processor sets the pixel to black. If it is lighter than the threshold, the pixel is set to white. Traditional image processing tries to recognize patterns in the black and white scan.
Grayscale Image Processing tries to recognize patterns. One way is to examine the pixels surrounding the pixel being processed. This surrounding environment "weighs" the image processing engine's decision towards darkening or lightening the pixel being processed. It is more than a simple contrast adjustment. This approach requires a lot of computer power and memory. At the bottom of it all, the user still has to determine a threshold -- how many pixels may be examined at one time.
Another approach to Grayscale Image Processing is to have the computer look at patterns similar to how recognition engines work. Handwriting recognition engines build a database on how handwritten letters look. They then try to match scanned images to this database. Most handwriting recognition systems try to make the handwriting conform to patterns the recognition engine expects to see.
A popular way of pursuing pattern recognition in grayscale images is through the use of neural networks. Neural networks allow computers to recognize patterns the same way humans do. With grayscale imaging, pattern recognition can be put to spectacular applications. Miros (Wellesley, MA 617-235-0330) uses neural networks for face recognition. Face recognition is used in security applications. It can be used to restrict access to buildings or to computers. One application of Miros' Trueface recognition engine is for automated check cashing. A customer submits a check to an automated teller machine. The machine snaps a picture of the customer and compares it to a photo kept on file. If the images match, the customer gets the money. Truface comes as an API ($2,000). The company has shrink wrapped applications for building (Gatewatch $3,500) and computer (Cyberwatch $60 per seat) security.
Some industry reps say grayscale imaging wastes too much memory. On the other hand, it provides more information. As the speed of scanning equipment increases and hardware compression improves, expect to see more grayscale scanning and image processing.
