Overview of Image Processing
57
An advantage of working with control points and windows centered on them
lies not only in the reduced computation but also in enabling registration of
relevant portions of the images. This is important when the images are not
of the same size or the two images have only a partial overlap in the scenes
captured. The correlation approach is obviously most applicable when one
image or a portion of it can be regarded as a shifted version of another image or a portion thereof. However, there are several situations where this is
not true. For example, even if a geographical area is captured with the same
imaging system with perfect camera registration but at widely separated times,
say in summer and winter, the intensity values will vary substantially due to
seasonal variations. Trying to register images through the maximization of
correlation would then be meaningless. An alternative approach in such cases
is shape matching which correlates features such as coastlines that do not show
seasonal variations.
A more general model is to account for differences in images that arise
not only due to translation but also due to scaling, rotation or shear. Such
effects complicate the registration process substantially. Feature based techniques that apply polynomial coordinate transformation identical to that described in the context of rectifying geometrical distortion, may be useful here.
Multiresolution techniques based on the wavelet transform have also been
developed to perform feature-based registration of images. For a review on
registration methods the reader is referred to (Brown 1992; Zitova and Flusser
2003).
2.S
Image Enhancement
The aim of image enhancement is to improve the quality of the image so
that it is more easily interpreted (e. g. certain objects become more distinct).
The enhancement may, however, occur at the expense of the quality of other
objects, which may become subdued in the enhancement operation. Image
enhancement can be performed using point operations and geometric operations, which are described next. It may, however, be mentioned that before applying these operations, the images are assumed to have been corrected for radiometric distortions. Otherwise, these distortions will also be
enhanced.
2.S.1
Point Operations
We begin by focusing on scalar image oriented processing, in particular point
operations where modification of a pixel value is on the basis of its current
value independent of other pixel values. A simple example is thresholding
where pixels of a specified value or greater are forced to one gray level and
the rest to another as shown in Fig. 2.2. Point operations are typically used to
enhance an image or to isolate pixel values on the basis of the range of values
57
An advantage of working with control points and windows centered on them
lies not only in the reduced computation but also in enabling registration of
relevant portions of the images. This is important when the images are not
of the same size or the two images have only a partial overlap in the scenes
captured. The correlation approach is obviously most applicable when one
image or a portion of it can be regarded as a shifted version of another image or a portion thereof. However, there are several situations where this is
not true. For example, even if a geographical area is captured with the same
imaging system with perfect camera registration but at widely separated times,
say in summer and winter, the intensity values will vary substantially due to
seasonal variations. Trying to register images through the maximization of
correlation would then be meaningless. An alternative approach in such cases
is shape matching which correlates features such as coastlines that do not show
seasonal variations.
A more general model is to account for differences in images that arise
not only due to translation but also due to scaling, rotation or shear. Such
effects complicate the registration process substantially. Feature based techniques that apply polynomial coordinate transformation identical to that described in the context of rectifying geometrical distortion, may be useful here.
Multiresolution techniques based on the wavelet transform have also been
developed to perform feature-based registration of images. For a review on
registration methods the reader is referred to (Brown 1992; Zitova and Flusser
2003).
2.S
Image Enhancement
The aim of image enhancement is to improve the quality of the image so
that it is more easily interpreted (e. g. certain objects become more distinct).
The enhancement may, however, occur at the expense of the quality of other
objects, which may become subdued in the enhancement operation. Image
enhancement can be performed using point operations and geometric operations, which are described next. It may, however, be mentioned that before applying these operations, the images are assumed to have been corrected for radiometric distortions. Otherwise, these distortions will also be
enhanced.
2.S.1
Point Operations
We begin by focusing on scalar image oriented processing, in particular point
operations where modification of a pixel value is on the basis of its current
value independent of other pixel values. A simple example is thresholding
where pixels of a specified value or greater are forced to one gray level and
the rest to another as shown in Fig. 2.2. Point operations are typically used to
enhance an image or to isolate pixel values on the basis of the range of values