| Previous | Table of Contents | Next |
Image enhancement alters images to make them subjectively more pleasing or improve their utility by making previously invisible (but mathematically present) content in the physical scene visible. Simply put, image enhancement makes images more visually digestible while retaining original image truth. In contrast, image editing and manipulation involves image alterations such as compositing which fundamentally change original image truths.
This chapter presents image enhancement techniques applicable to achromatic images such as gray scale images created by scanning continuous-tone black and white photographs. As described in Section 1, equi-energy white light is achromatic light that contains equal proportions of all light wavelengths. By varying equi-energy white light intensity from black through intervening grays to a maximum value deemed fully white, all present wavelengths remain in the same proportion and vary equally. Since any light source, whether it contains only one light frequency or multiple frequencies, can also preserve light component proportions as the source intensity varies, these techniques also work for images created with light sources other than true white light.
The VIDEO.C sample program shows implementations and provides output illustrating the effect for most of the discussed techniques.
An achromatic image is typically a rectangular integer array whose element values represent sampled and quantized intensity samples called pixels (picture elements). When the quantized sample values are only zero and one (bilevel images), the individual elements are sometimes referred to as pels. Within image files, image data is usually, but not always, stored as complete top-to-bottom image scan lines. Within a scan line, the data is usually stored from left to right. This gives rise to a Cartesian coordinate system with the Y axis inverted where the first (top) scan line is row (scan line) zero and the first positions (column) within all scan lines is column zero. So, the upper left pixel has (row, column) coordinate values (0, 0) instead of the bottom left pixel.
It is convenient to think of achromatic images as three-dimensional objects. The various pixel coordinate positions occupy two dimensions (horizontal row and vertical column positions) while the pixel value found at any (row, column) position denotes a distance into a third dimension. A useful metaphor considers astronauts gazing down from space as the sun sets on the Himalayan Mountains. They sense no vertical surface elevation differences but know white snow marks higher surface elevations while darker areas represent lower elevations—just as white light signifies higher intensities in black and white movies. Analogously, color images can be rendered using three primary colors. Each of the three primary contributions can be regarded as a three-dimensional object that is similarly viewed from above.
As previously discussed, long wavelength waves with high peaks and troughs transmit the most image energy and comprise an image’s foundation. The higher-order bit positions correspond to such waves because they usually tend to change more slowly across an image and have higher peaks and troughs than the low-order bit positions. Lower-order bit positions ride on and refine the higher-order bit positions, adding increasing image accuracy (image detail)—refinements analogous to ocean-wave chop. Literature sometimes refers to these individual bit positions as bit planes.
Viewed differently, instead of cutting an image up into individual pixels —much like slicing a cake—we can consider digital images to be made up of layers of bits—much like the layers of a cake, each layer being a bit plane. An image with 8-bit pixel values (0 to 255 intensity values) has eight bit planes. The bit plane view of an image is mainly used when adding graphics or special effects to an image. However, most continuous tone images are best regarded as pixel arrays rather than bit planes.
The VIDEO.C sample program illustrates the contribution each bit position individually has on a 6-bit (64 intensity level) achromatic image. Beginning with the high-order bit position, the program first visually displays which bits have a zero or one value at each pixel location. One values have a white visual representation and zero values have a dark visual representation. For the high-order bit, this amounts to a threshold process where pixel values greater than 31 receive an “on” value and pixel values less than 32 have an “off” value.
After a brief pause, the program progresses to the next lower bit position and performs the same visual representation (which is now not a digital halftone threshold operation because the bit position is not the high-order bit position). The process continues with the output becoming increasingly unintelligible because successive bit positions constitute refinements to the collective contributions of higher-order bit positions which are not visually provided. Hence, successive visual output increasingly assumes a random numeric-chatter appearance.
After examining all possible bit positions in isolation, VIDEO.C shows the cumulative effects of bit positions when they progressively accumulate, starting with the high-order bit position and progressing to the low-order. This effectively renders the image as a 1-, 2-, 3-, 4-, and 5-bit image. Note the extensive contouring that gradually disappears as successive bit positions refine the image.
| Previous | Table of Contents | Next |