Enhancement of an Image: 6 Techniques

This article throws light upon the six most common techniques used for enhancement of an image. The techniques are: 1. Contrast Enhancement 2. Pseudo-Colour Enhancement—Density Slicing 3. Density Slicing 4. Image Enhancement through Basic Numerical Calculations 5. Edge Enhancement 6. Filtering.

Technique # 1. Contrast Enhancement:

Remote sensing systems record reflected and emitted energy from earth materials. Ideally, one material would reflect a tremendous amount of energy in certain wavelengths, while another material would reflect much less energy in the same wavelengths. This would result in contrast between the two types of materials when recorded by a remote sensing system.

Unfortunately, different materials often reflect similar amounts of radiant flux throughout the visible and near-infrared portion of the electromagnetic spectrum, resulting in a relatively low contrast image. In an image like this, all characteristics of earth (related to the natural and the socio-economic reality and their changes through time etc.), become similar to the environment around them and not discernible.

Technique # 2. Pseudo-Colour Enhancement—Density Slicing:

Pesudo-colour enhancements of an image intend to convert it from grey scale to colour (not necessarily to natural colour or like false infra-red), in order to use all values of the electromagnetic spectrum. The discrimination ability of the human eye is usually limited to 16 different grey levels.

This limitation may be overcome if grey-scale images are converted to pseudo-colour. This conversion allows the utilisation of the entire range of the image brightness values or portions of it in different hue values.

Technique # 3. Density Slicing:

Density Slicing is the technique that assigns different hue values to each grey level, in order to make the discrimination of different homogenous zones easier. Usually, the selection of the successive portions is arbitrary and this causes loss of information.

Studying the histogram and selecting the portion that refers to discernible homogenous zones could prevent this. In this case the final image after density slicing control will be enhanced and much easier to interpret.

Technique # 4. Image Enhancement through Basic Numerical Calculations:

A remotely sensed image of the earth is a 3-D numerical matrix.

i. X, Y axes constitute a Cartesian reference system of spatial features.

ii. Z-axis refers to the pixels’ digital values for the amount of the energy emitted/ reflected from every snapshot of earth surface.

The pixel’s digital value range is determined by the radiometric resolution of the sensor.

0-63 for radiometric resolution of the sensor 6 bits (IRS)

0-127 for radiometric resolution of the sensor 7 bits

0-255 for radiometric resolution of the sensor 8 bits (Landsat TM)

0-511 for radiometric resolution of the sensor 9 bits

0-1023 for radiometric resolution of the sensor 10 bits (NOAA)

0-2047 for radiometric resolution of the sensor 11 bits (Ikonos)

The amount of the radiation emitted/reflected from specific features on the earth surface depends not only on their physical, chemical or biological characteristics, but also on a series of other factors (terrain features, sun ratioing etc.

This could cause difficulties during image interpretation, which may be eliminated by applying basic numerical calculation (subtraction, rationing) on the proper components (bands) of a multi-spectral digital remotely sensed image.

So, we may have image enhancement with the following characteristics:

(a) Addition of Brightness Values:

We can make an enhanced image by addition two or more remotely sensed images of the same date, time, region and dividing the sum with the number of the remotely sensed images. This image is the average of its components and is released from the total noise allocation.

(b) Subtraction of Brightness Values:

By subtracting the appropriate two bands of two multi-spectral satellite images of the same area acquired at the same date in different years or in different seasons of the same year, we can view the annual/seasonal changes in the area, because of the nullification of the same pixel’s digital values.

(c) Ratioing of Remotely Sensed Images:

The image interpretation potential of a multi-spectral digital remotely sensed image is often bounded by the fact that specific features of one homogeneous zone reflect/emit electro-magnetic radiation in different intensity.

So, the features appear on the image in different grey tones. The terrain features cause this— humidity, seasonal change etc. Ratioing the appropriate two components (bands) of the remotely sensed images, we can reduce the impact of the above mentioned factors.

Technique # 5. Edge Enhancement:

Image interpretation of a remotely sensed image may be easier if the edges of the objects/characteristics are enhanced by an edge enhancement operation, so that their shapes and details are enhanced.

Generally, what eyes see as pictorial, edges are simply sharp differences in brightness value between two pixels? The edge enhancement of a remotely sensed image can be performed, either with linear edge enhancement or with non-linear edge enhancement.

Technique # 6. Filtering:

The filtering of a remotely sensed image aims at the enhancement/improvement of the image, either with the elimination or compression of certain spatial frequencies and linear characteristics which abstract to the interpretation of other interesting characteristics (road network etc.) or with the enhancement of spatial frequencies and linear characteristics which concern us the most (boundaries of water resources).

The procedure, which is used to decompose a remotely sensed imagery to its different spatial frequency components, is Fourier analysis. Applying Fourier Transform on the original remotely sensed image is a way to enhance certain interesting spatial frequency components for image interpretation, contrary to others that do not interest us.

The algorithms, used for such as enhancement/improvement, are filters such as:

i. Low pass filters

ii. High pass filters

High pass filter emphasize high frequency spatial characteristics thus enhancing the linear characteristics of a remotely sensed image. Low pass filters, on the other hand, emphasise low frequency spatial characteristics, causing a “smoothing” of the remotely sensed image.

Special Enhancement Transformations:

Some of the special enhancement transformations used for natural resource inventories and monitoring, are related to principles, methods and techniques already mentioned.

So, we can use transformation such as:

i. Principal Component Analysis of a remotely sensed image

ii. Multiple Discriminant Analysis of a remotely sensed image.

Classification:

Remote sensing data are records of reflected and emitted electromagnetic energy presented as picture like images. In order to extract meaningful information out of the data, an interpretation of the image has to be carried out. The image interpretation is made with the help of an electronic computer, where use of mathematical algorithms takes place, provided of course, that the data are in digital form.

The classification methods of the computer assisted approach are:

i. Supervised Classification:

Representative sample sites of known land cover type (from a ground truth survey or from a map e.g. olives, forest, sea), called “training areas”, are selected in order to compile a numerical “interpretation key” that describes the spectral attributes of each feature type of interest.

In other words these pixels from the “training areas” will be used to “train” the computer so that each pixel from the data set will be compared numerically to each category in the interpretation key and labelled with category it “looks most like”.

There are a number of numerical strategies (means, standard deviations etc.) that can be employed to make this comparison between unknown pixels and pixels derived from the “training areas”. The comparison is always made with the use of a specific classification algorithm, like the Gaussion Maximum Likelihood, the Parallelepiped and the minimum distance to means algorithms.

ii. Unsupervised Classification:

In this approach no training samples are used. The pixels from the remote sensing data are classified by the method of cluster analysis, which can identify natural groupings of patterns. The classes that result from unsupervised classification are spectral classes.

Because they are based solely on the natural groupings in the image values, the identity of the spectral classes will not be initially known. The nature of each grouping is determined afterwards by ground truth surveys.

Unsupervised classification is not generally as effective as supervised classification because of the absence of training sets to control the results, especially when classes are only marginally separable. However, supervised classification is still too slow to handle a massive influx of satellite multi-spectral data.

The results of the classification from the computer- assisted approach can be presented as a thematic map, which can be stored in digital form. The computer can also display numerical information on the area of the mapped classes or the frequency of the occurrence of each class and other useful statistical data if required.