3
Image Filtering,
Enhancement,
and Restoration
3.1 INTRODUCTION AND OVERVIEW
The equipments with which we capture images are often influenced by measurement
noise; therefore, the resulting images may not provide the quality needed for desired
analysis. In addition, even for images with acceptable quality, it is often the case that
certain regions, features, or components of the image need to be emphasized and
highlighted. As an example, when processing medical images, it is often the case that
highlighting certain parts of the image such as tumor-like regions would help physicians make a better diagnosis. In this chapter, the main computational techniques commonly used in image processing to restore and enhance images are discussed.
Image restoration and enhancement techniques can be regarded as computational
algorithms that receive an image as their input and generate an enhanced or restored
version of this image as their output. This is shown in Figure 3.1.
Image restoration techniques are often described either in space domain or in frequency (Fourier) domain. While almost any space-domain technique has an equivalent
frequency-domain method and vice versa, some techniques are easier to understand
and/or implement in one domain than another. Some frequency-domain filtering techniques were described in the previous chapter, and in this chapter the focus is given to
space-domain techniques. It is not surprising to see that some of the space-domain techniques described in this chapter carry the same name as the ones described in the previous
chapter (such as low-pass filtering and high-pass filtering). As we will discuss later, these
filters are the space-domain equivalents of the filters described in the previous chapter.
Space-domain image enhancement techniques can be further classified into two
general categories: point processing and mask processing techniques. The point processing techniques are the ones in which each pixel of the original (input) image at
coordinates (x, y) is processed to create the corresponding pixel at coordinates (x, y)
in the enhanced image. This means that the only pixel in the original image that has
a role in determining the value of the corresponding pixel in the enhanced image is
the pixel located at the exact same coordinate in the original image. In contrast, in
mark processing techniques, not only the pixel at (x, y) coordinates of the original
image but also some neighboring pixels of this point are involved in generating the
pixel at (x, y) coordinates in the enhanced image. The schematic diagrams of the
point processing and mask processing techniques are compared with each other in
Figure 3.2. These two categories of space-domain image enhancement techniques
are further described in this chapter.
39