Image acquisition, representation and filtering.

We first list the different types of images that we may have to process. In order to be processed by a computing system, an image must be represented by a discrete ensemble of values, the pixels. Each pixel is associated with one or several values depending on the chosen representation.

The most elementary representation corresponds to the binary image, for which each pixel can take only one value among two. For monochrome images, each pixel can take one value among N. N is generally a power of 2, thus facilitating the image representation in the computer. For example, for a gray-level image, each pixel can take one value among 256: this value is then encoded by a data byte. This representation is frequently used and finds a justification involving the human visual system and the physical characteristics of the image support. This point is detailed in the Sampling and quantization section.

A tri-chrome image (or color image) is a superposition of three gray-level images corresponding to three basic colors. For images that will be displayed on a computer screen, the RGB (red, green, blue) representation is used. Each pixel of a tri-chrome image is thus associated with a triplet of values corresponding to the luminance of the basic colors.

This representation is not the only one allowing color images to be processed by a computing system. An alternative consists in using indexed color images, associating the matrix of pixels with a color table (colormap). More precisely, in this case, each pixel value is an index pointing to the color table. The color table is composed of three columns corresponding to the three basic colors. The number of rows in the table is equal to the total number of colors used for the representation. The indexed color representation is more cost-effective in terms of memory occupancy than the RGB representation since the number of colors is voluntarily limited. Therefore, it is also less precise in terms of image definition.

Images obtained through snapshots in the visible range are not the only ones to find applications, and multi-spectral images are a generalization of the previous case. They are represented by n tables of numbers.

Let us end this section by mentioning some quantitative information regarding image representation in the computer. An image is composed of a pixel matrix of N rows by M columns, where each pixel value is encoded in x bits.

For example, a tri-chrome image 512x512 for which each color is encoded in 8 bits requires 786 Ko of data for storage. Such an image is represented on figure 2. On figure 3, we represented the same image after applying the JPEG algorithm. This later image is stored in a memory space of only 15 Ko.

This example illustrates the significant improvements that have been made in the area of digital image representation within the past fifteen years. These improvements have been applied to the video domain and have lead to new encoding norms, such as mpeg2 and mpeg4 for the most famous ones, which are widely used for digital television broadcasting.