Overview of Image Processing
79
2.7
Image Change Detection
The objects on the earth surface change with time due to the economic, social
and environmental pressures. These changes need to be recorded periodically for planning, management and monitoring programs. Some programs
require estimation of changes over long time intervals whereas others need the
recording of short-term changes. The change in time is typically referred to as
temporal resolution.
To perform a systematic change detection study, two maps prepared at
different times are required. Aerial photographs and remote sensing images
that provide synoptic view of the terrain are regarded as attractive sources for
studying the changes on the earth's surface. This is due the availability of these
data at many different temporal resolutions. While the temporal resolution
of data from aircrafts may be altered as per the needs of the project, satellite
based data products have fixed temporal resolution. Once the data at two
different times are available, these can be analyzed to detect the changes using
three approaches - manual compilation, photographic compilation and digital
processing. Manual and photographic compilations are laborious and time
consuming. Therefore, digital methods of change detection are gaining a lot of
importance.
Before detecting changes using digital images, these need to be accurately
registered with each other (see Sect. 2.4). For example, registration accuracy
ofless than one-fifth of a pixel is required to achieve a change detection error
of less than 10% (Dai and Khorram 1999). Either feature based or intensity
based registration may be performed.
The fundamental assumption in applying any digital change detection algorithm is that there exists a difference in spectral response of a pixel (i. e.
intensity value) on images of two dates if there is a change in two objects from
one date to the other. There are many digital change detection algorithms that
can be used to derive change maps; three widely used ones are,
1. Image differencing
2. Image ratioing
3. Post classification comparisons
In image differencing, one digital image is simply subtracted from another
digital image of the same area acquired at different times. The difference of
intensity values is stored as a third image in which features with no change will
have near zero values. The third image is thresholded in such a way that pixels
showing change have a value one and the pixels with no change have a value
zero thereby creating a binary change image.
The image ratioing approach involves the division of intensity values of
pixels in one image to the other image. The ratios are used to generate a third
image. In this image, if the intensity values are close to 1, there is no change in
objects. This method has an important advantage in that by taking ratios the
variations in the illumination can be minimized.
79
2.7
Image Change Detection
The objects on the earth surface change with time due to the economic, social
and environmental pressures. These changes need to be recorded periodically for planning, management and monitoring programs. Some programs
require estimation of changes over long time intervals whereas others need the
recording of short-term changes. The change in time is typically referred to as
temporal resolution.
To perform a systematic change detection study, two maps prepared at
different times are required. Aerial photographs and remote sensing images
that provide synoptic view of the terrain are regarded as attractive sources for
studying the changes on the earth's surface. This is due the availability of these
data at many different temporal resolutions. While the temporal resolution
of data from aircrafts may be altered as per the needs of the project, satellite
based data products have fixed temporal resolution. Once the data at two
different times are available, these can be analyzed to detect the changes using
three approaches - manual compilation, photographic compilation and digital
processing. Manual and photographic compilations are laborious and time
consuming. Therefore, digital methods of change detection are gaining a lot of
importance.
Before detecting changes using digital images, these need to be accurately
registered with each other (see Sect. 2.4). For example, registration accuracy
ofless than one-fifth of a pixel is required to achieve a change detection error
of less than 10% (Dai and Khorram 1999). Either feature based or intensity
based registration may be performed.
The fundamental assumption in applying any digital change detection algorithm is that there exists a difference in spectral response of a pixel (i. e.
intensity value) on images of two dates if there is a change in two objects from
one date to the other. There are many digital change detection algorithms that
can be used to derive change maps; three widely used ones are,
1. Image differencing
2. Image ratioing
3. Post classification comparisons
In image differencing, one digital image is simply subtracted from another
digital image of the same area acquired at different times. The difference of
intensity values is stored as a third image in which features with no change will
have near zero values. The third image is thresholded in such a way that pixels
showing change have a value one and the pixels with no change have a value
zero thereby creating a binary change image.
The image ratioing approach involves the division of intensity values of
pixels in one image to the other image. The ratios are used to generate a third
image. In this image, if the intensity values are close to 1, there is no change in
objects. This method has an important advantage in that by taking ratios the
variations in the illumination can be minimized.