IMAGE ENHANCEMENT

Dr. ABHISHEK RAWAT

• The primary goal of image enhancement is to improve the visual
interpretability of an image by increasing the apparent distinction
between the features in the scene.
• The range of possible image enhancement and display options
available to the image analyst is virtually limitless.
• Most enhancement techniques may be categorized as either point or
neighborhood operations.
• Point operations modify the brightness value of each pixel in an
image data set independently.
• Neighborhood operations modify the value of each pixel based on
neighboring brightness values.
• Either form of enhancement can be performed on single-band
(monochrome) images or on the individual components of multi
image composites.
• Choosing the appropriate enhancement(s) for any particular
application is an art and often a matter of personal preference.
• Below, we discuss the most commonly applied digital enhancement
techniques. These techniques can be categorized as contrast
manipulation, spatial feature manipulation, or multi-image
manipulation.
Image Processing
• Contrast Manipulation
Grey Level Thresholding, Level Slicing, and Contrast Stretching
• Spatial Feature Manipulation
• Spatial filtering, edge enhancement, and Fourier analysis
• Multi Image Manipulation
• Multispectral band rationing and differencing,
• vegetation and other indices, principal components, canonical components,
• vegetation components, intensity–hue–saturation (IHS) and other color
space transformations, and decorrelation stretching.
• Gray-level Thresholding
Gray-level thresholding is used to segment an input image into two classes—
one for those pixels having values below an analyst-defined gray level and
one for those above this value.

• Level Slicing
• Level slicing is an enhancement technique whereby the DNs distributed
along the x axis of an image histogram are divided into a series of analyst-
specified intervals or “slices.” All of the DNs falling within a given interval in
the input image are then displayed at a single DN in the output image.
• Contrast Stretching
Image display and recording devices often operate over a range of 256
brightness levels (the maximum number represented in 8-bit computer
encoding). Sensor data in a single image rarely cover this exact range—
they may utilize only a small part of this 8-bit range.
Hence, the intent of contrast stretching is to alter the range of
brightness values present in an input image so as to optimally utilize
the full 8-bit range of display values.
DN’ = ((DN – MIN)/(MAX-MIN))*255
• SPATIAL FEATURE MANIPULATION
• Spatial Filtering
• spatial filters emphasize or deemphasize image data of various spatial
frequencies.
• Spatial frequency refers to the “roughness” of the tonal variations
occurring in an image.
• Image areas of high spatial frequency are tonally “rough.”
• the gray levels in these areas change abruptly over a relatively small
number of pixels.
• “Smooth” image areas are those of low spatial frequency, where gray
levels vary only gradually over a relatively large number of pixels.
• Spatial filtering is a “neighborhood” operation in that pixel values in
an original image are modified on the basis of the gray levels of
neighboring pixels.
• A simple low-pass filter may be implemented by passing a moving
window throughout an original image and creating a second image
whose DN at each pixel corresponds to the neighborhood average
within the moving window at each of its positions in the original
image.
 Convolution
• Spatial filtering is one special application of the generic image processing
operation called convolution. Convolving an image involves the following
procedures:
• A moving window is established that contains an array of coefficients or
weighting factors. Such arrays are referred to as masks, operators, or kernels,
and they are normally an odd number of pixels in size (e.g., 3.3, 5.5, 7.7).
• The kernel is moved throughout the original image, and the DN at the center of
the kernel in a second (convoluted) output image is obtained by multiplying
each coefficient in the kernel by the corresponding DN in the original image and
adding all the resulting products.
• A simple low-pass filter may be implemented by passing a moving window
throughout an original image and creating a second image whose DN at
each pixel corresponds to the neighborhood average within the moving
window at each of its positions in the original image.
• A simple high-pass filter may be implemented by subtracting a low-pass
filtered image (pixel by pixel) from the original, unprocessed image.
 Edge Enhancement
• Edge enhancement is a digital image processing technique used to
emphasize the edges and boundaries within an image, making them more
prominent and easier to perceive.
• This enhancement is achieved by emphasizing the high-frequency
components (the rapid changes in intensity or color) that typically
correspond to edges and fine details in an image.
• Edge enhancement is commonly used in various fields, including image
processing, computer vision, and photography.
MULTI-IMAGE MANIPULATION
Spectral Ratioing
• Ratio images are enhancements resulting from the division of DN values in
one spectral band by the corresponding values in another band.
• A major advantage of ratio images is that they convey the spectral or color
characteristics of image features, regardless of variations in scene
illumination conditions.
• Ratioed images are often useful for discriminating subtle spectral variations
in a scene that are masked by the brightness variations in images from
individual spectral bands or in standard color composites.
Simple Ratio (SR)
• Many indices make use of the inverse relationship between red and near-
infrared reflectance associated with healthy green vegetation. Cohen
(1991) suggests that the first true vegetation index was the Simple Ratio
(SR), which is the ratio of red reflected radiant flux ( red) to near-infrared
radiant flux ( nir)
SR = ρRed / ρNIR
• Normalized Difference Ratios and Other Indices
• One of the most widely used normalized difference indices is the
Normalized Difference Vegetation Index (NDVI).
• This index is based on the difference of reflectance in the near-infrared and
red bands.
NDVI = ρNIR - ρRed / ρNIR + ρRed

• High NDVI values will result from the combination of a high reflectance in
the near infrared and lower reflectance in the red band.
• This combination is typical of the spectral “signature” of vegetation.
• Non-vegetated areas, including bare soil, open water, snow / ice, and most
construction materials, will have much lower NDVI values.
• Seasonal and inter-annual changes in vegetation growth and activity can be
monitored.
• The ratioing reduces many forms of multiplicative noise (Sun illumination
differences, cloud shadows, some atmospheric attenuation, some
topographic variations) present in multiple bands of multiple date imagery.
• It has been shown to be well correlated not only with crop biomass
accumulation, but also with leaf chlorophyll levels, leaf area index values,
and the photosynthetically active radiation absorbed by a crop canopy.
• Soil Adjusted Vegetation Index (SAVI)

SAVI = (ρNIR – ρRed) /( ρNIR + ρRed + L) * (1+L)

• where L is a correction factor between 0 and 1, with a default value of

0.5 (as L approaches 0, SAVI becomes close to NDVI).
• Normalized Difference Moisture or Water Index (NDMI or NDWI)

NDMI= ρNIR – ρSWIR / ρNIR + ρSWIR

Landsat TM near- and middle-infrared bands, was highly correlated with

canopy water content and more closely tracked changes in plant biomass
and water stress than did the NDVI
• Normalized Difference Built-Up Index NDBI
• Many professionals working on urban/suburban problems are interested in
monitoring the spatial distribution and growth of urban built-up areas.
• These data can be used for watershed runoff prediction and other planning
applications.
NDBI = Bu – NDVI
Bu = NIR TM4 – Mid IR TM5 / NIR TM4 + Mid IRTM5
• This resulted in an output image that contained only built-up and barren
pixels having positive values while all other land cover had a value of 0 or
±254. The technique was reported to be 92% accurate.
• Normalized Burn Ratio (NBR)
• One of the most widely used spectral indices for mapping burn
severity is the Normalized B urn Ratio (NBR) (Brewer et al., 2005).
• The NBR combines the reflectances in the near-infrared ( nir) and the
shortwavelength infrared bands ( swir)
NBR = ρNIR - ρSWIR / ρNIR + ρSWIR

• The NBR is sometimes used to map fire-affected areas using a single

post-fire image, but is more commonly applied to quantify burn
severity as the difference (𝛥) between pre- and post-fire values
observed in registered multiple-date images
𝛥 NBR = NBRprefire – NBRpostfire
Principal Components Analysis
(The ability to reduce the dimensionality.)
• Images generated by digital data from various wavelength bands often
appear similar and convey essentially the same information.
• Principal components analysis is a technique that transforms the original
remotely sensed dataset into a substantially smaller and easier to
interpret set of uncorrelated variables that represents most of the
information present in the original dataset.
• The transformations generally increase the computational efficiency of the
classification process because both principal and canonical component
analyses may result in a reduction in the dimensionality of the original
data set.
• The purpose of these procedures is to compress all of the information
contained in an original n-band data set into fewer than n “new bands”.
• Sometimes, variables are highly correlated in such a way that it would be
duplicate information found in another variable. The principal component
analysis identifies duplicate data over several datasets. Then, PCA
aggregates only essential information into groups called “principal
components“.
• The power of PCA is that it creates a new dataset with only the essential
information.
• The bottom line is that you reduce redundancy when using PCA.
• To perform principal components analysis we apply a transformation to a
correlated set of multispectral data.
• The application of the transformation to the correlated remote sensor data
will result in an uncorrelated multispectral dataset that has certain ordered
variance properties.
• This transformation is conceptualized by considering the two-dimensional
distribution of pixel values obtained in two TM bands, which we will label
X1 and X2.
• Vegetation Component
• In addition to the vegetation indices described previously, various
other forms of linear data transformations have been developed for
vegetation monitoring, with differing sensors and vegetation
conditions dictating different transformations.
• The “tasseled cap” transformation produces a set of vegetation
components useful for agricultural crop monitoring.
• The majority of information is contained in two or three components
that are directly related to physical scene characteristics.
• Brightness, the first component, is a weighted sum of all bands and is
defined in the direction of the principal variation in soil reflectance.
• The second component, greenness, is approximately orthogonal to
brightness and is a contrast between the near-infrared and visible bands.
• Greenness is strongly related to the amount of green vegetation present in
the scene.
• A third component, called wetness, relates to canopy and soil moisture.
• Urbanized areas are particularly evident in the brightness image. The greater
the biomass, the brighter the pixel value in the greenness image. The
wetness image provides subtle information concerning the moisture status
of the wetland environment.
• Intensity–Hue–Saturation Color Space Transformation
• Digital images are typically displayed as additive color composites
using the three primary colors: red, green, and blue (RGB).
• For a display with 8-bit-perpixel data encoding, the range of possible
DNs for each color component is 0 to 255.
• Hence, there are 2563 (or 16,777,216) possible combinations of red,
green, and blue DNs that can be displayed by such a device.
• Every pixel in a composited display may be represented by a three-
dimensional coordinate position somewhere within the color cube.
• The line from the origin of the cube to the opposite corner is known
as the gray line because DNs that lie on this line have equal
components of red, green, and blue.
• The RGB displays are used extensively in digital processing to display
normal color, false color infrared, and arbitrary color composites.
• For example, a normal color composite may be displayed by assigning
TM or ETM+ bands 1, 2, and 3 to the blue, green, and re components,
respectively.
• An alternative to describing colors by their RGB components is the
use of the intensity–hue–saturation (IHS) system.
• Intensity relates to the total brightness of a color.
• Hue refers to the dominant or average wavelength of ligh contributing
to a color.
• Saturation specifies the purity of color relative to gray.
• Transforming RGB components into IHS components before
processing may provide more control over color enhancements.
• Transforming RGB components into IHS components. This particular
approach is called the hexcone model, and it involves the projection
of the RGB color cube onto a plane that is perpendicular to the gray
line and tangent to the cube at the corner farthest from the origin.
The resulting projection is a hexagon.
• If the plane of projection is moved from white to black along the gray
line, successively smaller color subcubes are projected and a series of
hexagons of decreasing size result. The hexagon at white is the largest
and the hexagon at black degenerates to a point.
• The series of hexagons developed in this manner define a solid called
the hexcone.
• In the hexcone model intensity is defined by the distance along the
gray line from black to any given hexagonal projection.
• Hue and saturation are defined at a given intensity, within the
appropriate hexagon.
• Hue is expressed by the angle around the hexagon, and saturation is
defined by the distance from the gray point at the center of the
hexagon.
• The farther a point lies away from the gray point, the more saturated
the color.
• Decorrelation Stretching
• Decorrelation stretching is a form of multi-image manipulation that is
particularly useful when displaying multispectral data that are highly
correlated.
• Traditional contrast stretching of highly correlated data as R, G, and B
displays normally only expands the range of intensities; it does little to
expand the range of colors displayed, and the stretched image still
contains only pastel hues.
• Decorrelation stretching involves exaggeration of the least correlated
information in an image primarily in terms of saturation, with minimal
change in image intensity and hue.
• decorrelated information
• exaggerated primarily in terms of saturation.