Image Enhancement

Image enhancement is applied to improve the overall visual quality of an image—to smooth or sharpen image features such as edges for better visual interpretation or better subsequent digital image analysis. It is applied to many types of images, such as medical images, digital photography, and remotely sensed imagery.

In geography, Earth's surface information can be obtained by the interpretation and understanding of images acquired from sensor devices on board satellites and airplanes. Recently, many new remote sensors have been developed that can detect electromagnetic signals in different wavelength ranges with various spatial and spectral resolutions. However, the original images delivered to general users may not display all the surface objects clearly. Human analysts may not be able to distinguish slight brightness differences for different objects on the original images. The objective of image enhancement is to optimize the image appearance and to aid human analysts with particular image interpretation tasks. The evaluation of the quality of the enhancement is usually subjective; it depends on the human eye and on the particular application. Various image enhancement techniques have been developed for remotely sensed imagery. The commonly used methods include contrast enhancement, spatial filtering, and Fourier transform.

In original images, the brightness values rarely cover the entire range of the gray levels (commonly 8 bits or 256 levels). Therefore, the image contrast may be low, or the image appears dark. Contrast enhancement increases the contrast between image objects and between objects and their surroundings by modifying the brightness values of the image pixels. The brightness value frequency distributions or histograms will be changed. The result is an output image with better image contrast for optimal visualization and interpretation. There are several contrast enhancement methods. For example, the linear contrast stretch methods uniformly expand the range of the brightness values to fill the entire range of the gray levels. The histogram equalization method is a nonlinear contrast stretch method, which considers the frequency distribution or histogram of the image brightness values and assigns pixel values based on [Page 1533]equalizing the histogram. As a result, the pixel values are redistributed according to the histogram. Therefore, the contrast is stretched for pixels with brightness values that appear more frequently and reduced for pixels with infrequent brightness values, such as very dark or very bright pixels.

The contrast enhancement assigns a new value to a pixel based on the brightness value of this single pixel. Spatial filtering assigns a new value to a pixel based on the values of a local neighborhood, or a window, of this pixel. Spatial filters include low-pass filters, high-pass filters, and directional filters. Low-pass filters emphasize low-frequency features and have a smoothing effect on the image. High-pass filters emphasize high-frequency features and may be used for edge enhancement and detection. Directional filters can be used to detect edges in specific orientations.

Fourier transform is a mathematical operation to transform the original image from the spatial domain—the (x, y) coordinate space of the image—to the frequency domain. In the transformed Fourier spectrum, the lower frequencies in the original image are plotted at the center of the spectrum, and the high frequencies are plotted away from the center. Various filters can be designed and applied to the Fourier spectrum. For example, a circular low-frequency blocking filter blocks the center of the Fourier spectrum and can be used as a high-pass filter. After filtering in the frequency domain, an inverse Fourier transform is performed to transform the filtered version back to the spatial domain. With the high-pass filter, the result is an edge-enhanced image. With the low-pass filter, the result is a smoothed image.

...