To begin with, it is important to understand that image enhancement is applied to facilitate visual interpretation and understanding of images. The enhancement will not change the radiometric values of the objects in the image; it will just allow an observer a better view of these objects. This step, therefore, only serves to help the user define the learning samples and signatures to be used in the classification.
Digital images have the advantage of enabling and easily manipulation of the values recorded for each pixel. Although it is possible to perform radiometric adjustments for the effects of solar illumination, atmospheric conditions and characteristics of the instruments used before distributing the images to users, it may be possible the image is not at its best for visual interpretation. Remote sensing systems, especially those using a spatial platform, must be designed to handle the different energy levels of the targets and their environment, which may be found in normal conditions of use. This significant variation in the spectral response of the different types of targets (eg forest, desert, snow, water, etc.) makes it impossible to apply a general radiometric adjustment likely to optimize the contrast and intensity levels in each one of the different conditions. Therefore, a different adjustment of the tones according to the use and the state of each of the images has to be performed.
Satellite images must, almost always, be improved for different reasons. Firstly, the scene brightness range may not be sufficient to cover the entire range of values supported by the sensor. If it displays no improvement, the scene will seem dark and with poor definition. Secondly, the 16-bit images of modern satellites have to be reduced to fit the 8-bitrange (from 0 to 255) of your computer screen.
You get both results by stretching the image values to use them as different levels of red, green and blue brightness available on your computer screen while remaining in the range supported by your monitor.
In an unprocessed image, the useful information is often included in a restricted set of numerical values among the possible values (256 in the case of 8-bit data). The enhancement of the contrasts is performed by changing the initial values so as to use all the possible values, which increases the contrast between the targets and their environment. To understand how this type of enhancement works, one must, first, understand the concept of an image histogram. A histogram is a graphical representation of the numerical valuesof intensity that make up an image. These values (from 0 to 255 for8-bitdata) appear along the x-axis of the graph. The frequency of occurrence of each of these values is presented along the y-axis.
In the case of a satellite image band, the values will not be in a grey scale but, as in this histogram, in radiance values.
The minimum value present is 0 (NoData) and the radiance values are within a minimum value of 4831and a maximum value of 32767. But the simple observation of the histogram shows that almost all the values are within the 5,500 and 17,000range. The simplest method is a linear enhancement of contrast. In order to apply this method, we must identify the upper and lower intensity limits represented on the histogram (the minimum and maximum values),and using a linear transformation, we stretch these values over all available values.
When using QGis the enhancement functions are found in LayerProperties-> Style->ContrastEnhancement
The following figure illustrates three different ways to stretch the image values of a unique Landsat band of 16bits. Thumbnail images show the results and the histograms show how the stretching has been applied. Input data values (from the file ) appear on the x – axis and the resulting output values (on the screen) are on the y-axis. The blue histograms show the input values distribution and the red lines in diagonal represent the correspondence tables or conversion tables. The stretching is applied by reading each input value (from the X axis), by tracing a vertical line until reaching the conversion table and extending a horizontal line towards the Y axis , where it searches for the corresponding output value , as indicated by the example of the green arrows in (a)
In (a), No enhancements option from the drop-down menu , input pixel values are simply rescheduled for the 16-bit range of 0 to 65535 to 8 bits of the range of 0 to 255. You can see that the histogram does not cover a small part of the brightness range available and the resulting image is very dark.
In (b), Stretch until MinMax option from the drop-down menu ,the input is adjusted so as to extend between the actual input data minimum and maximum values (compare the X-axis labels (a)). Although the histogram covers a larger section of the brightness range than before it, still,does not cover much and the picture is always very dark.
In (c), Stretch and cut until MinMax option from the drop-down menu , the diagonal is adjusted so that its extremities cover 99% of the entrance. The remaining 1% of the data is set to zero for (dark pixels)or 255 (for bright pixels). The blue dotted outline dotted shows the output histogram obtained and you can see that it covers the entire brightness range. The image shows a nice distribution of tones from black to white and is the best value setting of the three available options.
We have not discussed the fourth option, Cut to MinMax because it is seldom used. Display is removed in the same pixels as in the previous options but the retained values are not stretched.
In practice, contrast enhancements are applied to the three color bands of an image , resulting in improved colors.
QGis processing functions are all linear, which is typical for the available improvements in many GIS applications. ArcGIS as well as processing dedicated applications images and photo editing takes in charge various parametric and nonlinear improvements.
To access more powerful processing functions with QGis, it is necessary to install the processing provider Orfeo Toolbox.This library allows specific processing to remote sensing images.
