Image Enhancements Indices and Transformations Remote Sensing Process

Image Enhancements, Indices and Transformations

Remote Sensing Process (A) Energy Source or Illumination Recording of Energy by the Sensor (D) Transmission, Reception, and Processing (E) Interpretation and Analysis (F) Radiation and the Atmosphere (B) Application (G) Interaction with the Target (C) Reference: CCRS/CCT

Remote Sensing Process (A) Energy Source or Illumination Recording of Energy by the Sensor (D) Transmission, Reception, and Processing (E) Interpretation and Analysis (F) (B) Radiation and the Atmosphere Application (G) Interaction with the Target (C) Reference: CCRS/CCT

Remote Sensing Process (A) Energy Source or Illumination Recording of Energy by the Sensor (D) Interpretation and Analysis (F) (B) Radiation and the Atmosphere (C) Interaction with the Target Transmission, Reception, and Processing (E) Application (G) Reference: CCRS/CCT

Remote Sensing Process (A) Energy Source or Illumination (D) Recording of Energy by the Sensor Interpretation and Analysis (F) (B) Radiation and the Atmosphere (C) Interaction with the Target Transmission, Reception, and Processing (E) Application (G) Reference: CCRS/CCT

Remote Sensing Process (A) Energy Source or Illumination (D) Recording of Energy by the Sensor Interpretation and Analysis (F) (B) Radiation and the Atmosphere (C) Interaction with the Target (E) Transmission, Reception, and Processing Application (G) Reference: CCRS/CCT

Remote Sensing Process (A) Energy Source or Illumination (B) Radiation and the Atmosphere (D) Recording of Energy by the Sensor (E) Transmission, Reception, and Processing Interpretation and Analysis (F) Interpretation and Analysis (C) Interaction with the Target Reference: CCRS/CCT

Remote Sensing Process Energy Source or Illumination (A) Radiation and the Atmosphere (B) Interaction with the Target (C) Recording of Energy by the Sensor (D) Transmission, Reception, and Processing (E) Interpretation and Analysis (F) (G) Application Reference: CCRS/CCT

Applications Carbon Management Public Health Energy Management Aviation Water Management Homeland Security Coastal Management Disaster Management Agricultural Efficiency Invasive Species Ecological Forecasting Air Quality 9

Image enhancement Alteration of the image in such a way that the information contained in the image is easier to visually interpret or systematically analyze 10

Types of image enhancement § Radiometric enhancement § Spatial enhancement § Spectral enhancement 11

Types of image enhancement § Radiometric enhancement § Spatial enhancement § Spectral enhancement 12

Radiometric enhancement • Compensates for inadequacies in the image contrast (too dark, too bright, too little difference between the brightness of features in the image) • Attempts to optimize the distribution of pixel values over the radiometric range of the image 13

Radiometric enhancement Often increases contrast for some image pixels while decreasing it for others. 14

Types of radiometric enhancement 1. Linear stretch 2. Piecewise linear stretch 3. Histogram equalization (non-linear stretch) 15

Linear stretch • Simple method that expands the range of original image pixel values to the full radiometric range of the image; • Best applied to images where pixel values are normally distributed 16

Minimum/maximum linear stretch 17

Minimum/maximum linear stretch no stretch linear stretch 18

Original Minimummaximum +1 standard deviation 19 Contrast Stretching of Predawn Thermal Infrared Data of the Savannah River

Piecewise linear stretch Allows for enhancement of a specific range of pixel values 20

Piecewise linear stretch • Slope of the linear contrast enhancement changes • Piecewise contrast stretching (sometimes referred to as using breakpoints) 21

Piecewise Linear Contrast Stretching

Histogram equalization (non-linear stretch) • Redistributes pixel values so that there are roughly the same number of pixels with each value within a range • Applies greatest contrast enhancement at the peaks of the histogram 23

Histogram equalization Dark Most populated 24 Light

Histogram matching Convert the histogram of one image to match the histogram of another 25

Histogram matching rules § General shape of histograms should be similar § Relative dark/light features should be the same § Spatial resolution should be the same § Same relative distribution of land cover 26

Histogram matching rules • Histogram matching is useful for matching data of the same or adjacent scenes that were scanned on separate days, or are slightly different because of sun angle or atmospheric effects • Especially useful for mosaicing or change detection 27

Histogram matching + input image = match image LUT 28 output image

Types of image enhancement § Radiometric enhancement § Spatial enhancement § Spectral enhancement 29

Spatial enhancement • Modifies pixel values based on the values of surrounding pixels • Changes the “spatial frequency” of an image 30

Spatial frequency • The number of changes in pixel value per unit distance for any particular part of an image • Few changes – low frequency area • Dramatic changes – high frequency area 31

Spatial frequency Neighboring pixel brightness values rather than an independent pixel value 32

Types of spatial enhancement 1. Convolution filtering 2. Resolution merge 33

Convolution filtering • Process of assigning a new value for an image pixel based on a weighted average of surrounding pixels • Can be used to visually enhance an image OR to prepare an image for classification 34

Kernel • A matrix of coefficients used to average the value of each image pixel with the neighborhood of pixels surrounding it • Kernel is systematically moved across the image and a new value is calculated for each input image pixel (at the center of the kernel) 35

Kernel 36

Convolution Formula the kernel coefficient at column i, row j the pixel value at column i, row j the dimension of the kernel (i. e. , 3 X 3) the sum of the kernel coefficients (if 0, then 1) the output pixel value 37

High-frequency (high-pass) kernel • Increase spatial frequency • used to enhance “edges” between nonhomogeneous groups of image pixels • Not often used prior to classification 39

High-frequency (high-pass) kernel before filtering after filtering 40

Zero-sum kernel • Sum of all kernel coefficients is zero • Output pixel values are zero where equal • Low values become much lower, high values become much higher • Used as an edge detector • Can be biased to detect edges in a certain direction • Kernel above is biased towards the south • Stream delineation, fault mapping 41

Zero-sum kernel before filtering after filtering 42

Low-frequency (low-pass) kernel • Kernel coefficients are usually equal • Simply averages pixel values • Results in increased pixel homogeneity and a “smoother” image • Most widely-used filtering mechanism • Smooth terrain; reduce noise; generalize land cover (post-classification) • Kernel: 3 X 3 or 5 X 5 43

Low-frequency (low-pass) kernel before filtering after filtering 44

Resolution merge Using an image with high spatial resolution to increase the spatial resolution of a lower spatial resolution image of the same area (a. k. a. , “pan sharpening”) 45

Resolution merge + original MS (30 m) = panchromatic (15 m) output image (15 m) § note that this changes the input image pixel values 46

Types of image enhancement § Radiometric enhancement § Spatial enhancement § Spectral enhancement 47

Spectral enhancement • Create, expand, transform, analyze or compress multiple bands of image data • Can be used to both visually enhance data and prepare it for image classification 48

Types of spectral enhancement 1. Principal component analysis 2. Tasseled cap 3. Indices 49

Principal Components Analysis (PCA) • Transforms a multi-band image into a series of uncorrelated images (“components”) that represent most of the information present in the original dataset • Can be more useful for analysis than the original source data 50

Principal Components Analysis (PCA) • The first one or two components represent most of the information (variance) present in the original image bands; PCA reduces data redundancy • First PC accounts for the maximum proportion of the variance, each succeeding PC accounts for the maximum proportion of the remaining variance • Reduce dimensionality (i. e. , # of bands need to be analyzed) 51

band #2 values band #1 values 1 st component is the longest axis (AB); minimizes the squared distance from each point to the line § 2 nd component is the 2 nd longest axis (CD); it’s “orthagonal”, or completely uncorrelated with the first axis § Original image values are converted based on the equation defining the axis line §

Landsat ETM+ (6 bands, excluding thermal & pan) 1 2 3 4 5 7

Principal Components Analysis (PCA) PCA seeks to generate uncorrelated images to reduce data redundancy. pixel 1 pixel 2 pixel 3 band 1 40 60 100 band 2 60 80 120 band 3 30 50 90 band 4 200 30 50 band 5 120 230 20 band 7 10 80 255 • The pixels in bands 1 through 3 are perfectly correlated • Band 2 = Band 1 + 20; Band 3 = Band 1 -10 • Bands 4, 5 & 7 are less correlated to Band 1. They contain more unique information to contribute to the first PC 54

Tasseled cap transformation • Transforms a multi-band image into a series of images optimized for vegetation studies using coefficients specific to a particular sensor • Images represent the “brightness”, “greenness”, and “wetness” • Vegetation studies: § brightness is used to identify and measure soil § greenness is used to identify and measure vegetation § wetness is used to measure soli/vegetation moisture content 55

Tasseled cap transformation Healthy dense vegetation Water Micale and Marrs 2006 Bare soil

57

Indices • Create new images by mathematically combining the pixel values from multiple image bands • Most often ratios of band values 58

Common uses of indices § Mineral exploration § Reduce radiometric differences § Minimize shadow effects § Vegetation analysis 59

Normalized Difference Vegetation Index (NDVI) • A ratio of the red visible and near infrared bands • Used widely as a measure of both the presence and health of vegetation • Values range from -1 to +1 60

Normalized Difference Vegetation Index (NDVI) Based upon findings that the chlorophyll in plant leaves strongly absorbs red visible light (from 0. 6 to 0. 7 µm), while the cell structure of the leaves strongly reflects near-infrared light (from 0. 7 to 1. 1 µm) 61

Normalized Difference Vegetation Index where: NIR is the near-infrared response of pixel p R is the visible red response of pixel p § example: NIR=100, R=50 (0. 333) § NDVI is positive when NIR > R, negative when NIR < R § Larger NDVI values result from larger differences between the NIR and Red bands § Note that the software may scale the -1 to +1 NDVI values to 8 -bit (0 to 255) 63

Normalized Difference Vegetation Index (NDVI) The main difference between green and dry vegetation is the amount of red visible absorbed

Normalized Difference Vegetation Index (NDVI) greyscale NDVI pseudocolor NDVI 65

NASA MODIS global NDVI