Remote Sensing

Practical 3: Image Enhancement

Part 1: Contrast and Spatial Enhancement
Aims and Objectives
This practical is the first of two which aim to introduce you to a number of common image
enhancement techniques.

Core Tasks for this session

1. Undertake contrast enhancement on an image using linear stretches and image histogram
equalization
2. Spatially enhance an image using convolution filtering
At the end of the following 2 practical sessions you should have some understanding of the
processes and issues associated with image enhancement and be able to carry out a number of
these techniques using ENVI.

Introduction
“The term enhancement is used to mean the alteration of the appearance of an image in such a way
that the information contained in that image is more readily interpreted.” (Mather, 1999)
One type of image enhancement with which you are already familiar from the previous lab is image
reduction and magnification, this is a simple example but makes the point of how you can gather
different information from the same image by simply zooming in and out of an image. In this lab,
we will introduce more advanced image enhancements. The basic idea it to use either global
statistics (those from the entire image) or neighborhood statistics (these from a smaller number of
pixels) to modify or adjust each pixel’s value. Not that after image enhancements are performed,
you no longer have the same digital numbers for the bands.
 INFORMATION BOX
There are 3 main types of image enhancement:
1. Contrast Enhancement
Contrast enhancement is only intended to improve the visual quality of a displayed
image. It is a process generally required for all images because satellite-based sensors
with global coverage need to image a wide range of scenes. From very low radiance
(oceans, low solar elevation angles, high latitudes etc) to very high radiance (snow,
sand, high solar elevation angles, high altitudes). Any one scene will generally have a
radiance range much less than the full range. When such a scene is imaged and
converted to digital numbers (DNs), it uses less than the full quantization range
(usually 0-255). When the image is displayed it will have low contrast, appearing quite
dull, making visual interpretation difficult. Contrast stretching involves increasing the
range (spreading or stretching) of data values to occupy the available image display
range (usually 0-255). The result is displayed image with a greater level of contrast.

2. Spatial Enhancement
Spatial enhancement techniques modify the value of a given pixel based on the values
of surrounding pixels.

3. Spectral Enhancement
Spectral enhancement techniques are mainly transforms that are used to aid in
interpreting image data

In this practical we will specifically concentrate on contrast and spatial enhancement techniques.
We will not cover all the available enhancement methods, instead we will be concentrate on the
following:
1. Contrast Enhancement
• Simple linear stretch
• Linear stretch with saturation
• Histogram equalization
2. Spatial Enhancement
• High pass convolution filter
• Low pass convolution filter
Data for this Practical

Task 1: Contrast Enhancement- Contrast Stretches

A. Simple linear stretch
This is the simplest type of stretch, it uses the minimum and maximum data values as stretch
endpoints. Basically, the endpoints of the data distribution are pulled to the endpoint of the palette
and all the values in between are re-scaled accordingly.

Before we can apply a contrast stretch to our image, we must first view the image histogram.
1. Load the Murrells inlet image into a false colour display as RGB = 6,4,2.

2. Click on custom stretch tool to appear custom stretch dialog box.

 (a) (b)

The default dialog box is as figure (a) which is the look-up table (LUT) that govern visual display
and you can see the histogram of each data values by clicking the button encircled by black ellipse
in figure (a).
The text box immediately below the histogram plot of each band represents the minimum and
maximum values which participates in the stretch i.e. in this case 3 and 160. These values are
represented on the plots by two vertical lines, which we will call range bar (maximum and
minimum range bar).
The button encircled by black ellipse allows you to toggle between the bands that you have
assigned to R G B in your image i.e. bands 6,4,2.
This is the display contrast set by default by computer. At the present display, you will notice that
the contrast in the image is poor, especially for wetlands.

Question 1:
Explain the shape of the Red and NIR histograms based on your knowledge of the
electromagnetic spectrum.
B. Applying a Linear Stretch
Now, you can experiment using interactive stretching.
1. Set the method to Linear and linear (percent value) to 1.
2. To apply linear stretch, move the maximum and minimum range bars to the left and right to find
a level which best enhances the details of the wetlands within Murrell’s inlet colour. Remember
that you can alters the stretches in each of the RGB bands.

Questions 2:
What were the stretch parameters (numerical values below the histograms) that you choose
to enhance detail within the wetlands of Murrell’s Inlet and why did your chosen parameters
improve the contrast in this region?

C. Apply Histogram Equalization

Histogram equalization is another form of contrast stretching.
1. Remove all the stretches by reloading the image into a new display (RGB = 6,4,2)
2. Choose custom stretch and change the method to the equalization.
3. Change the minimum and maximum range bar as per requirements.
3. See the results in image.

Question 3:
What happens to the images when using histogram equalization? When would histogram
equalization not be appropriate to use?
Other technique:
There are a range of other contrast enhancement techniques and you can try them in your own if
you wish.

Task 2: Spatial Enhancement- Convolution Filtering

A convolution filter is the process of averaging small sets of pixels across and image. A
convolution kernel is a matrix of numbers that is used to average the value of each pixel with the
values surrounding pixels in a particular way. The numbers in the kernel act to weigh this average
toward particular pixels.

A. High pass convolution filter

1. Load the Murrell Inlet image and select a single band to display as a grey scale image 9at this
point it does not matter which band you choose).
2. From the toolbox, choose Convolution and Morphology. The Convolution and morphology
tool window appear.

3. Click on Convolution > High Pass.

4. You can change the size of the kernel. For now, keep it as 3*3.
5. Click on apply to file to save the results to a new file. The convolution input file window
appears.
6. Select the input file as Murrells inlet image. Rather than applying the filter to all bands (the
default), to save time select the spectral subset button and then in the spectral subset window,
select the Red band (band 3). Select Ok and then Ok. You should then see the following window.
7. Enter a name for you output file – i.e. something that tells you what the input is, the filter type
is and the size. Make sure that you save it in your home file store directory.
8. From the available bands list, load up the resulting image in a new display to explore the results,
link the displays.

Question 4:
Explain how the high pass convolution kernel works and what it does to the image?

B. Low pass convolution filter

9. Repeat the above procedure, but this time select convolutions > low pass from the convolutions
window toolbar.
Question 5:
Explain how the low pass convolution kernel works and what does it do to the image?

10. Now repeat both processes but change the kernel size to 7*7 for each.

Question 6:
What differences do increasing the kernel size from 3*3 to 7*7 have on the images?

11. Try other type of the filtering such as directional, Laplacian, median, sobel, Robert filter etc if
you wish.