• Tidak ada hasil yang ditemukan

(band dependent); 1 day repeat cycle), and IKONOS (4 spectral bands;

4 m spatial resolution; 5 day repeat cycle).

IMAGE PROCESSING

Remote sensing was one of the first disciplines based on analysing and using digital images, and many image-processing techniques were developed under remote sensing projects. Digital images are generated by a combination of source of light and reflection or absorption of energy from that source by the elements of the features being imaged.

The idea to convert energy into image consists of the following three preliminary steps.

• Incoming energy is converted into a voltage by a combination of sensor material, which is responsible for detection of a particular type of energy and input electrical power.

• The output voltage waveform is the response of the sensor(s).

• Digital quantity is obtained from each sensor by digitizing its response.

Pixel, the building block for every digital image, is a tiny small grid structure cell. For example, each of the Landsat TM’s scan lines is approximately 30 m wide. These lines are divided into slices, and so the pixels are roughly 30 m squares. The amount of reflectance for each band generated by each pixel is the genesis for the generation of digital image. Each pixel has a numerical value, called digital number (DN), which records the intensity of the electromagnetic energy measured for the ground resolution cell represented by that pixel (Figure 4.8). Digital numbers range from zero to some higher number on a grey scale.

Digital images are stored in the computer’s memory. They do not have any physical size until they are displayed on a screen or printed

Figure 4.8 Pixel representation of image

on paper. An image refers to a two-dimensional light intensity function f(x, y), where (x, y) denotes spatial coordinates, and the value of f at any point (x, y) is proportional to the brightness or grey levels of the image at that point. A digital image is an image f(x, y) that has been discretized both in spatial coordinates and brightness.

Fundamental Steps in Image Processing

Image processing is a set of techniques that uses computer algorithms to process images as input. The output of processing is an image or set of characteristics of the image. Image processing not only refers to digital image processing but also includes analogue and optical image processing.

• Image acquisition: Data acquisition is the first step of image processing. Usually, sensors are used to acquire data. A monochrome or a colour television camera, common sensors that are generally used for image processing, produces an entire image of the problem domain every 1/30 second. If the output of the camera or any other imaging sensor is not already in digital form, an analogue-to-digital converter digitizes it. Note that the acquisition could be as being given an image that is already in digital form. Generally, the image acquisition stage involves preprocessing, such as scaling.

• Image preprocessing: Image preprocessing techniques are a set of methods that improve the image in ways that increase the chances for success of other processes.

• Image segmentation: Segmentation partitions an input image into its constituent parts or objects.

• Image representation: Representation converts the input data to a form suitable for computer processing.

• Image description: Image description extracts features that result in some quantitative information of interest or features that are basic for differentiating one class of objects from another.

• Image recognition: In image recognition, a label is assigned to an object based on the information provided by its descriptors.

• Image interpretation: Image interpretation assigns meaning to an ensemble of recognized objects.

Thus the fundamental components of an image-processing system include the following subsystems: acquisition, storage, processing, communication, and display.

Histograms and Scatter Plot

A histogram is a statistical technique that represents the distribution of data with the help of graphics. The term was coined by Karl Pearson, and the graph is meant to estimate the probability distribution of continuous values. A histogram can be n-dimensional. The image histogram is a graphical representation of the tonal distribution of the digital image, which plots the number of pixels for each tonal value. The horizontal axis represents tonal variation, while the vertical axis represents the number of pixels. A colour histogram represents colour variations of an image. Thus the histogram of an image with 256 levels of grey will be represented by a graph having 256 values on the x-axis and the number of image pixels on the y-axis (Figure 4.9).

Image editors allow the histogram for an image to be edited. The algorithm incorporated with the editors allows the user to adjust the tonal variation or brightness of the image. Histogram equalization is a technique that maximizes the image contrast by applying a grey- level transformation to try to flatten the resulting histogram. The prime application of histograms in computer vision is thresholding.

Figure 4.9 Image histogram

Source <www.rockware.com/rockworks/revisions/2005_q2.htm>

Thresholding is an image-segmentation technique used to create binary images by identifying object pixels with a pixel value greater than some threshold value. An object pixel is given a value of “1”, while a background pixel is given a value of “0”. Finally, a binary image is created by colouring each pixel white or black, depending on a pixel’s labels. Because the information contained in the graph is a representation of pixel distribution as a function of tonal variation, image histograms can be analysed for peaks and/or valleys, which can then be used to determine a threshold value. This threshold value can then be used for edge detection and image segmentation.

Image Enhancement

Image enhancement is the process of adjusting digital images so that the results are more suitable for display or further analysis. For example, noise can be removed or an image can be brightened, making it easier to identify key features.

Nowadays, digital images have enveloped the complete world. Digital cameras, which are the main source of digital images, are widely available in the market and are inexpensive. Sometimes the image taken from a digital camera is not of good quality and requires some enhancement.

The image-enhancement process consists of a collection of techniques that seek to improve the visual appearance of an image or to convert the image to a form better suited for analysis by a human or machine (Figure 4.10). There exist many techniques that can enhance a digital image without spoiling it.

The enhancement methods can be broadly divided into two categories:

spatial domain methods and frequency domain methods. In spatial domain techniques, one directly deals with image pixels. The pixel values are manipulated to achieve the desired enhancement.

Spatial domain methods: grey-scale manipulation

The simplest spatial domain operations occur when the neighbourhood is simply the pixel itself. The simplest form of operation is when the operator T acts only on a 1 × 1 pixel neighbourhood in the input image,

Figure 4.10 Image enhancement

that is, F(x, y) depends on the value of F only at (x, y). This is a grey-scale transformation or mapping. In this case, T is referred to as a grey-level transformation function or a point-processing operation.

Frequency domain methods

In frequency domain methods, the image is first transferred into a frequency domain. The Fourier transform of the image is computed first.

All the enhancement operations are performed on the Fourier transform of the image. Then the inverse Fourier transform is performed to get the resultant image.

Low-pass Filtering

Low-pass filtering involves the elimination of the high frequency components in the image. An ideal low-pass filter would retain all the low frequency components and eliminate all the high frequency components. However, an ideal filter suffers from two problems—

blurring and ringing.

Contrast Stretching

Contrast stretching (often called normalization) is a simple image- enhancement technique that attempts to improve the contrast in an image by “stretching” the range of the intensity values it contains to span a desired range of values, such as the full range of pixel values that the image type concerned allows. It differs from the more sophisticated histogram equalization in that it can only apply a linear scaling function to the image pixel values. As a result, the “enhancement” is less harsh (most implementations accept a grey-level image as input and produce another grey-level image as output).

This operation is much better for image quality enhancement in comparison to brightness control. If a low-contrast image results due to low light conditions and lack of dynamic range of the camera sensor, contrast-stretching operation results in a good quality image. During the contrast-stretching operation, the dynamic range of the grey values is basically increased.

Image Classification

The classification process categorizes all pixels in a digital image into one of several land cover classes or “themes”. These categorized data may then be used to produce thematic maps of the land cover present in an image. The objective of image classification is to automatically categorize all pixels in an image into land cover classes or information

themes. Informational classes and spectral classes are two major types of classes. Informational classes are based on the interest of users, such as soil type and surface temperature. Spectral classes are groups of pixels that are uniform with respect to brightness values and patterns in their multiple spectral channels. Two main classification methods are supervised classification and unsupervised classification.

Supervised classification

“Supervised” means under the guidance of someone. The classification process for the identification of unknown pixels carried out with the help of a training set of prior identified informational classes is called supervised classification (Figure 4.11). Identification and delineation of training areas are keys to successful implementation of the classification procedure. Supervised classification is driven by data in which the process develops statistical relationships between the input variables and the ground-truth habitats.

The traditional image results from the detection of reflected ambient daylight (visible and near-visible spectra). The spectral response depends on the nature of the objects the light is reflected from. This is used to create a characteristic signature for each habitat type. The training site is similar to a cookie cutter because it cuts through all the spectral image layers and extracts the values for each spectrum. These spectral values

Figure 4.11 Supervised classifcation procedure

are then used to create the habitat signature. The “signature” is in the form of a statistical probability distribution in as many dimensions as there are input images (n dimensions). The probability distribution is calculated using the maximum likelihood estimator. Each habitat will have its own signature, and together they form a signature catalogue.

These signatures are then applied to the whole image (actually, a stack of images, one for each spectrum). The spectral values for each pixel (one value per spectrum) are matched to the signature catalogue, and each pixel is given a probability value of belonging to each habitat category depending on where it lies in the n-dimensional probability distribution.

Usually, the corresponding pixel of the habitat image is assigned to the habitat that has the highest probability. This decision rule is termed a “hard” classifier since it does not take account of uncertainty.

However, the probabilities can be used in other ways to create maps which reflect uncertainty of the classification (for example, through fuzzy classification).

Unsupervised classification

Unsupervised classification is a method that examines a large number of unknown pixels and divides it into a number of classes based on natural groupings present in the image values. The classes that result from unsupervised classification are spectral classes, which are based on natural groupings of the image values. Unsupervised classification is becoming increasingly popular in agencies involved in long-term geographic information system (GIS) database maintenance. The reason is that there are now systems that use clustering procedures that are extremely fast and require little in the nature of operational parameters.

Unsupervised classification yields an output image in which a number of classes are identified and each pixel is assigned to a class.

These classes may or may not correspond well to land cover types of interest, and the user will need to assign meaningful labels to each class. Unsupervised classification often results in too many land cover classes, particularly for heterogeneous land cover types. Classes often need to be combined to create a meaningful map. In other cases, the classification may result in a map that combines multiple land cover classes of interest, and the class must be split into multiple classes in the final map. Unsupervised classification is useful when there is no pre-existing field data or detailed aerial photographs for the image area and the user cannot accurately specify training areas of known cover type. Additionally, this method is often used as an initial step

prior to supervised classification (called hybrid classification). Hybrid classification may be used to determine the spectral class composition of the image before conducting more detailed analysis and to determine how well the intended land cover classes can be defined from the image.