There are two fundamental ways to obtain digital imagery: Acquire remotely sensed imagery in an analog format (often

n referred to as hard-copy) and then convert it to a digital format through the process of digitization, and Acquire remotely sensed imagery already in a digital format, such as that obtained by the Landsat 7 Enhanced Thematic Mapper Plus (ETM+) sensor system.

Characteristics and statistics of digital remote sensing imagery

Image Digitization: Analog to Digital (A/D) Conversion Using Linear Array Scanners
Image digitization is the process of turning a hard-copy analog imagery into a digital image. (e.g., a desk-top scanner)

A hardcopy image is illuminated by white light. An objective disperses the light reflected from the hardcopy image on to registered red, green, and blue (RGB) linear arrays of detectors. The linear arrays contain thousands of detectors. The analyst sets the output numbers of pixels per inch (dpi) and the scanner samples the linear arrays accordingly. Digitization progresses line-by-line over the hardcopy image. When a color image is scanned, this results in a three-band (RGB) registered datasets.

RGB = True Color. Any addition array that collect spectral information beyond visible light will be able to generate pseudo color image.
Additional array?

Panchromatic

Black & White Infrared

True-color

vs.

Pseudo (false)-color

With all images are digital, knowledge and skills in digital image processing are necessary.

Digital Image With raster data structure, each image is treated as an array of values of the pixels. Image data is organized as rows and columns (or lines and pixels) start from the upper left corner of the image. Each pixel (picture element) is treated as a separate unite.

Statistics of Digital Images: The person responsible for analyzing the digital remote sensor data should first assess its quality and statistical characteristics. This is normally accomplished by: Looking at the frequency of occurrence of individual brightness values in the image displayed in a histogram viewing individual pixel brightness values at specific locations or within a geographic area; Computing univariate descriptive statistics to determine if there are unusual anomalies in the image data; and Computing multivariate statistics to determine the amount of between-band correlation (e.g., to identify redundancy).

Statistics of Digital Images: Looking at the frequency of occurrence of individual brightness values in the image displayed in a histogram Viewing individual pixel brightness values at specific locations or within a geographic area Statistics of Digital Images: Computing univariate descriptive statistics to determine if there are unusual anomalies in the image data; and

Statistics of Digital Images: Computing univariate descriptive statistics to determine if there are unusual anomalies in the image data; and

Statistics of Digital Images: Computing multivariate statistics to determine the amount of between-band correlation (e.g., to identify redundancy).

Forest

Water

Statistics of Digital Images It is useful to calculate fundamental univariate and multivariate statistics of the multispectral remote sensor data. This normally involved computing of maximum and minimum value for each band of imagery, the range, mean, standard deviation, between-band variance-covariance matrix, correlation matrix, and frequencies of brightness values for each band, which are used to produce histograms. Such statistics provide valuable information necessary for processing and analyzing remote sensor data.

A population is an infinite or finite set of elements. A sample is a subset of the elements taken from a population used to make inferences about certain characteristics of the population. (e.g., training signatures) Large samples drawn randomly from natural populations usually produce a symmetrical frequency distribution. Most values are clustered around some central value, and the frequency of occurrence declines away from this central point. A graph of the distribution appears bell shaped is called a normal distribution.

Histogram and Its Significance to Digital Remote Sensing Image Processing Many statistical tests used in the analysis of remotely sensed data assume that the brightness values recorded in a scene are normally distributed. Unfortunately, remotely sensed data may not be normally distributed. In such instance, nonparametric statistical theory may be preferred. The histogram is a useful graphic representation of the information content of a remote sensing image, which provide readers with an appreciation of the quality of the original data, e.g. whether it is low in contrast, high in contrast, or multimodal in nature.

Univariate Descriptive Image Statistics Measures of Central Tendency in Remote Sensor Data Mode: is the value that occurs most frequently in a distribution and is usually the highest point on the curve. Multiple modes are common in image dataset. Median: is the value midway in the frequency distribution. Mean: is the arithmetic average and if defined as the sum of all observations divided by the number of observations.

Measures of Central Tendency in Remote Sensor Data The mean is the arithmetic average and is defined as the sum of all brightness value observations divided by the number of observations. It is the most commonly used measure of central tendency. The mean (k) of a single band of imagery composed of n brightness values (BVik) is computed using the formula:

The sample mean, k, is an unbiased estimate of the population mean. For symmetrical distributions, the sample mean tends to be closer to the population mean than any other unbiased estimate (such as the median or mode).

Sample mean is a poor measure of central tendency when the set of observations is skewed or contains an outlier. 1 1 1 1 1 1 0 0 0 1 1 1 1 1 1 1 1 1 1 1 19 1 1 1 2 2 2 Skewed Mean = 9/9 = 1 Mean = 9/9 = 1 Outlier Mean = 27/9 = 3 (Does not represent the dataset well)

Measures of Dispersion Measures of the dispersion about the mean of a distribution provide valuable information about the image. For example, the range of a band of imagery (rangek) is computed as the difference between the maximum (maxk) and minimum (mink) values:

Unfortunately, when the minimum or maximum values are extreme or unusual observations (i.e., possibly data blunders), the range could be a misleading measure of dispersion. Such extreme values are not uncommon because the remote sensor data are often collected by detector systems with delicate electronics that can experience spikes in voltage and other unfortunate malfunctions. When unusual values are not encountered, the range is a very important statistic often used in image enhancement functions such as minmax contrast stretching.

Standard Deviation (sk): is the positive square root of the variance.

s
k

Standard Deviation (sk): is the positive square root of the variance.

s
19 18 1 1 14 12 13
7.35

var

A small sk. suggests that observations clustered tightly around a central value. A large sk indicates that values are scattered widely about the mean. The sample having the largest variance or standard deviation has the greater spread among the values of the observations.
Mean

var

12

var

( BV
2

ik

i1

( 19 432 8

9)

n 1 (1 9 )

( 12

9)

( 18

9)

(1 9 ) 8

( 13

9)

(1 9 )

( 14

9)

( 12

9)

54

Standard Deviations

var

54

7 . 35

Measures of Dispersion Variance: is the average squared deviation of all possible observations from the sample mean. The variance of a band of imagery, vark, is computed using the equation:
BV11 BV12 BV13 BV21 BV22 BV23 BV31 BV32 BV33

BV11 BV21 BV31

BV12 BV22 BV32

BV13 BV23 BV33

var k
2 ( BV ik k ) i 1 n 1 n

var k

i 1

( BV ik

k )2

1 1 1

1 1 1

1 1

var

(1 1 ) 9 1

i1

n 1

19 1 18

1 1 14

12 13 12

The numerator of the expression is the corrected sum of squares (SS). If the sample mean (mk) were actually the population mean, this would be an accurate measurement of the variance.

var

( BV

i 1

ik

1
k

)2

( 19 9 ) 2 432 8 54

n 1 ( 1 9 ) 2 ( 12 9 ) 2 ( 18 9 ) 2 ( 1 9 ) 2 (13 9 ) 2 ( 1 9 ) 2 ( 14 9 ) 2 ( 12 9 ) 2 8

Measures of Distribution (Histogram) Asymmetry and Peak Sharpness Skewness is a measure of the asymmetry of a histogram and is computed using the formula:
BV11 BV12 BV13 BV21 BV22 BV23 BV31 BV32 BV33

Histogram of A Single Band of Landsat Thematic Mapper Data

A perfectly symmetric histogram has a skewness value of zero.

Max. = 102 Min. = 6 Mean = 27 Median = 25 Mode = 9

Histogram of Thermal Infrared Imagery of a Thermal Plume in the Savannah River

Remote Sensing Multivariate Statistics Remote sensing research is often concerned with the measurement of how much radiant flux is reflected or emitted from an object in more than one band (e.g., in red and near-infrared bands). It is useful to compute multivariate statistical measures such as covariance and correlation among the several bands to determine how the measurements covary. Variancecovariance and correlation matrices are the key for principal components analysis (PCA), feature selection, classification and accuracy assessment.

Max. = 188 Min. = 38 Mean = 73 Median = 68 Mode = 75

Remote Sensing Multivariate Statistics The different remoteremote-sensingsensing-derived spectral measurements for each pixel often change together in some predictable fashion.

Remote Sensing Multivariate Statistics If there is no relationship between the brightness value in one band and that of another for a given pixel, the values are mutually independent; that is, an increase or decrease in one band bands brightness value is not accompanied by a predictable change in another band bands brightness value.

Landsat-7 ETM+ Band 1 (Blue band)

Landsat-7 ETM+ Band 2 (Green band)

Landsat-7 ETM+ Band 3 (Red band)

Landsat-7 ETM+ Band 4 (Near IR band)

Remote Sensing Multivariate Statistics Because spectral measurements of individual pixels may not be independent, some measure of their mutual interaction is needed. This measure, called the covariance, is the joint variation of two variables about their common mean. To calculate covariance, we first compute the corrected sum of products (SP) defined by the equation:

Remote Sensing Multivariate Statistics It is computationally more efficient to use the following formula to arrive at the same result:

This quantity is called the uncorrected sum of products. products.

Remote Sensing Multivariate Statistics Covariance is calculated by dividing SP by (n 1). Therefore, the covariance between brightness values in bands k and l, covkl, is equal to:

Covariance: is the joint variation of two variables about their common mean. SPkl is the corrected Sum of Products between bands k and l. SP kl cov kl n 1
SP kl
19 Band k 18 1

i 1
1 1 14

( BV ik
12

k )( BV il l )
1 Band l 1 1 1 1 1 1 1 1

k = 9

13 12

l = 1

SPkl = (19-9)x(1-1)+(1-9)x(1-1)+(12-9)x(1-1)+(18-9)x(1-1)+(1-9)x(1-1)+(13-9)x(1-1)+ (1-9)x(1-1)+(14-9)x(1-1)+(12-9)x(1-1) = 0 Covkl = 0

Covariance: A more efficient formula is:

SP kl
n

i 1

( BV ik

BV il ) i 1

BV ik n

i 1

BV il

Format of a VarianceVariance-Covariance Matrix

Band 1 (green)
1 1

Band 2 (red) cov1,2

Band 3 (near(nearinfrared) cov1,3 cov2,3

Band 4 (near(nearinfrared) cov1,4 cov2,4 cov3,4

19 Band k Total = 91 1 18

1 1 14

12 13 12 Band l Total = 9

1 1 1

1 1 1

Band 1 Band 2

SS1
cov2,1 cov3,1 cov4,1

SS2
cov3,2 cov4,2

Band 3
SPkl = [(19x1)+(1x1)+(12x1)+(18x1)+(1x1)+(13x1)+(1x1)+(14x1)+(12x1)] [91x9/9] = 0 Covkl = 0

SS3
cov4,3

Band 4

SS4

The sums of products (SP) and sums of squares (SS) can be computed for all possible band combinations.

Correlation between Multiple Bands of Remotely Sensed Data To estimate the degree of interrelation between variables in a manner not influenced by measurement units, the correlation coefficient, r, is commonly used. The correlation between two bands of remotely sensed data, rkl, is the ratio of their covariance (cov (covkl) to the product of their standard deviations (sksl); thus:

Correlation Coefficient: A correlation coefficient of +1 indicates a positive, perfect relationship between the brightness values of the two bands. A correlation coefficient of -1 indicates that the two bands are inversely related. A correlation coefficient of zero suggests that there is no linear relationship between the two bands of data.

r kl

cov kl sk sl