Sample Essay Technology is Rapidly Advancing over the past years various applications and discoveries have been obtained and studied

Utilisation of Image Texture in Multi-spectral Analysis of satellite images for classifying land cover

            Technology is rapidly advancing over the past years, various applications and discoveries have been obtained and studied. Multi-spectral analysis of satellite images is one of the new discoveries in this century and it stimulates new applications. Multi-spectral analysis poses a huge potential in the field of forestry, geology, and even agriculture. According to Natural Resources Canada (2007) multi-spectral analysis is the study of factual information in various spectral bands. Wenger (1984) discussed that in multispectral analysis, every Land sat scenes of 85x85 mi is separated into thousands of pixels which is around 1 acre in size wherein each of the pixel has an equivalent location within the scene. In addition, Wenger stated that multispectral analysis cannot be anticipated to stand alone and must be utilised with other pieces of information sources like the topographic maps and aerial photographs, moreover multispectral analysis gives reliable basic information for areas that are large in just a short span of time.

            The objectives of this paper are the following:

• The main objective of this paper is to examine image texture in multi-spectral analysis of the satellite images of land cover.
• This paper also gives discussion on texture analysis and the various measure used in the image texture.
• And finally, a conclusion will be made.

 

Image Texture

            Tuckerman and Jain (1998) defined image texture as a function of the spatial differentiation in the intensities of the pixels; it is also very helpful in numerous applications and has been a subject of intense study and scrutiny by a number of researchers. CIVIL (1999) discussed that texture models differentiate local spatial information in an image, moreover texture can be a useful tool in identifying things particularly in natural scenes such as the land cover, and however the texture of an image relies on factors such as scene geometry and illumination conditions.   

            Multi-spectral analysis uses various types of multi-spectral scanners in determining the vegetation mapping, fire mapping, and monitoring soil moisture. Wenger enumerated and discussed the instruments and these are:

Multispectral Scanners- it is an optical-mechanical electronic device that observes the scene under an aircraft or satellite platform in a number of discrete bands of the ultraviolet, visible, and reflected, near and middle infrared portions of the electromagnetic spectrum. Wegner discussed that a normal multispectral scanner is composed of a rotating mirror and a telescope in order to emphasis radiation reflected from a small part of the surface of the earth on an array of detectors that are sensitive to energy. Every detector in the system observes the same element of resolution of the scene below however in different bands of wavelength.

 

 

Types of Multispectral Scanners

            Airborne Multispectral Scanners- the commercial airborne multispectral scanners gather electromagnetic energy in 5-12 bands. The five most appropriately chosen bands are the most efficient wherein the two are the most visible, one is near the infrared, one middle infrared, and one thermal infrared. The airborne multispectral scanners could deliver spatial and spectral resolutions necessary to precisely map wild land resources; on the other hand this type of multispectral scanner is not economical.

Land sat Multispectral Scanner- the launching of the Land sat-I in the year 1972 and Landsat-2 and 3 in the year 1975 and 1978 simultaneously, have given the resource managers a comparatively inexpensive multispectral scanner (MSS) data on a customary basis which is every 9 to 18 days. The data from the Land sat are duplicated in either on the digital form or the photographic form on magnetic tapes that are compatible in computers. The data from the Land sat Multispectral scanners are specified in five distinct bands of the electromagnetic spectrum. Wenger stated that the Land sat satellite multispectral data is helpful in the following areas:

• Vegetation mapping in wider classes.
• Stratification for sampling in broad renewable natural resource inventories.
• Mapping defoliation.
• Mapping and estimating clear-cut areas.
• Mapping the surface water.

 

Aside from the Land sat Multispectral Scanner, another Land sat instrument is

being utilised by geographers, researchers, etc. and this is the Land sat Thematic mapped (TM) which is a multispectral satellite that measures the electromagnetic energy in seven spectral bands that is composed of seven spectral bands that extends from visible to thermal infrared and each pixel indicates an area of 30 m by 30 m in six out of the seven bands and the pixels in the thermal band indicates an area of 120 m by 120 m (Romero, 2001).

Figure 1: Land sat Thematic mapped image of Jackson Purchase area of western Kentucky (Source: Romero, 2001).

Texture Analysis

            Texture analysis is the segmentation or classification of textural features with consideration to the shape of the small element, density and direction of prevalence (Hogg, nod.).

            Chen, Pau, and Wang (1998) discussed that the model based texture analysis techniques’ foundations are on the development of an image model that could be utilised to describe and synthesise the texture.

            There are three types of measures utilised by researchers, geographers, agriculturists, etc. in the image texture and these are the statistical measures, spectral measures, and structural measures.

Statistical Measures- according to Hogg in the statistical methods, the grey level-histogram, statistics based in grey-level occurrence matrix are calculated in order to make a clear distinction on the different textures of the image. In the Statistical Approach have four types according to Hogg and these are:

First Order Histogram- According to Hogg the first order histogram describes the regularity of presence of each grey level in a local area. The most widely used statistical measures are the average, variance, entropy, and coefficient of variation. 

Average:                                                                   Variance:

                             

Entropy:                                                                   Co-efficient of variation:

               

fi =  frequency of grey level i that occurs in a pixel window

quantk= quantization level of the image

W= total number of pixels in a window.

            Second-order grey level-co-occurrence matrix- the second-order set of texture measures which is based on the brightness value spatial-dependency grey-level co-occurrence matrices (Heraldic, 1973). The six most widely used measures are the angular second moment, the contrast, the correlation, the entropy, the inverse difference moment, and the covariance.

Angular second moment:                                                Contrast:

  

Correlation:                                                                         Entropy:

     

Inverse difference moment:                                            Covariance:

        

quantk= quantization level of the image

hc(i)= the (i,j)th entry in one of the grey level co-occurrence matrix

            Third order grey level co-occurrence matrix- it is a higher-order of texture measures that was based on third-order histograms, the measures used for this technique are the angular second moment contrast, correlation, entropy, the inverse and the covariance difference moment.

Angular second moment:                       

 

Contrast:

Correlation:

Entropy:

Inverse:

Covariance:

Phi, j, k) = the (i, j, kith entry of the third order grey level co-occurrence matrix

quantk=quantization level of the image

            Grey level difference vector- it is the sum of the diagonals of the grey-level co-occurrence matrices. The measures used for this technique are angular second moment, contrast, correlation, entropy, inverse difference moment, and variance.

Angular Second Moment:                                    Contrast:     

                                         

Correlation:                                                             Entropy:

                                             

Inverse difference moment:                                Variance:

                                              

chi=(m) = the probability that a pair of brightness values having an absolute difference of m occurs at separation c.

            In the study of Alaska, (1989) in classifying the land cover using the statistical measure there were three difference statistics and these are the inters ample spacing distance, Band, and Features. Alaska et al stated that the statistical measures are capable for identifying the information regarding the texture of the land cover; on the other hand the simple introduction of the information on texture by statistical measures in the spectral information could not make the classification accurate.        

Spectral Measures

            Aside from the statistical measures, researchers also utilize the spectral approaches in analysis of the texture. In this approach, Hogg discussed that the textured image is changed into frequency domain, afterwards the extraction of texture features could be done through analysis of the power spectrum.  The spectral measures have three classifications and these are the Fourier Transform, Gabor filter bank, and wavelet decomposition.

Fourier Transform- Morita (1995) it is a calculation needed to see a wave not only in the time domain but also in the frequency domain. Darken (1985) stated that the Fourier transform is a generalization of the complicated Fourier series in the limit as . The common Fourier transform pairs according to Darken are the following:

Function
Fourier transform--1	1
Fourier transform--Cosine
Fourier transform--Delta function
Fourier transform--Exponential function
Fourier transform--Gaussian
Fourier transform--Heaviside step function
Fourier transform--Inverse function
Fourier transform--Lorentz Ian function
Fourier transform--Ramp function
Fourier transform--Sine

 

Gabor Filter Bank- Prasad and Dome (2005) discussed that the Gabor filter is acquired through modulation of a sinusoid with a Gaussian.

Wavelet decomposition-