Psy 221 Project Report

Introduction


Digital cameras use an array of Charge Coupled Devices (CCDs) to collect information about an image scene. A CCD is essentially a silicon integrated circuit of the MOS type, comprising of an oxide (SiO2) covered silicon substrate upon which is formed an array of closely spaced electrodes. Signal information in each CCD is carried in the form of electric charges, usually electrons, stored in a "potential well." Hence, inherently each CCD in the digital camera can store only one-dimensional information, i.e., the number of photoelectrons on the CCD created from an image falling on the CCD array. To obtain full-color images, we need, at each pixel (the location of each CCD), more than the charge information alone. In fact, because of the trichromatic nature of the human visual system, we need at least (and only) three pieces of data, from which we can deduce the intensity of three independent colors (e.g., red, green, and blue).

Color encoding in digital cameras relies on the use of color filters, which are placed somewhere between the incoming light and the CCD arrays. So, each CCD now records the local intensity information of a particular color (that of the filter) present in the incoming light. The trichromatic information at each pixel can be acquired in several ways.

First, we can obtain full-color information with three exposures, using a different kind of color filter for each one. This approach produced very high qualify images, but is slower and cannot be used to acquire fast-changing images.

Second, we can optically split the exposing light into the three primary colors, and then use three sets of CCD arrays to acquire the intensity information for each color. This approach requires only one exposure, but has some drawbacks. It "wastes" many incoming photons (absorbed by the color filters) and hence creates problems when the scene is not well illuminated or illuminating. More importantly, this design incurs very high production cost because of the extra optics and CCD arrays required.

Third, we can use a color filter array (CFA), so that each CCD (or pixel) is covered by only one kind of filter. Certainly we have only information of one color at each pixel. But we can make the CFA in such a way that the missing intensity information at each pixel can be inferred (to a reasonable degree of accuracy) from the intensity information around that pixel. Such inference is of course not perfect, but either is our visual system*s ability to see small discrepancies. So with some care, we may still be able to produce high-quality (to our eyes) full-color images.

This third approach is the most common in digital camera design, and the most commonly used CFA is known as the Bayer pattern, part of which is as shown here. Notice that there are twice as many green filters as red or blue ones. The reason for such a design is that the human visual system is more sensitive to green patterns than blue or red ones.

To display the acquired image in full color, we need to up-sample the image, a process that is often referred to as demosaicking or color interpolation. Two commonly used color interpolation algorithms for Bayer pattern CFAs are the median method and the bilinear methods. These two are very easy to implement in digital camera design, and are very computationally efficient. In a recent paper, Acharya and Tsai (1999) proposed a new block matching algorithm. The goal of this paper is to examine this new algorithm and to compare its performance with the commonly used algorithms.