15: Images & Image Processing Flashcards
WHAT IS IMAGE PROCESSING
collection of methods helping us improve images and understanding of what the image is
Example: make island boundary more distinct
PRE AND POST PROCESSING STEPS
Preprocessing aim: correct distortion / degraded image data to create a more faithful representation of original scene (on raw data) types: radiometric correction geometric correction
Postprocessing aim: improve image scene quality and help data interpretation, derive information from the data types: image enhancement image classification change detection
LOW PASS FILTERS
- “blurring”, “smoothing” or “moving average” filter - averages out rapid changes in intensity
- simplest: calculates average of a pixel and all of its 8 immediate neighbours (3x3 kernel with averaging filter), then replaces original pixel value, process repeated for every image pixel
- uses a kernel with an averaging filter, 3x3, 5x5, or 7x7, the larger the kernel the greater the averaging effect
HIGH PASS FILTERS
- does the opposite: sharpens appearance of fine detail in an image
- first applies a low-pass filter to an image and then subtracts the result from the original, leaving behind only high spatial frequency information
- directional, or edge detection filters: designed to highlight linear features e.g. roads or field boundaries, useful for vectorisation process
- can also be designed to enhance features oriented in specific directions and are useful in applications, e.g. geology, for detection of linear features or structures
NDVI
Normalised Difference Vegetation Index
Uses red and near infrared bands
NDVI = NIR - Red / NIR + Red
Normalised: to account for different time of day/year in which satellite imagery may have been acquired, therefore relative level of reflectivity is always changing, mitigate it by dividing it by the sum
- shows plant ‘greenness’ / photosynthetic activity
- commonly used vegetation indices
- range -1 to +1
- very low (<0.1): barren areas of rock, sand or snow
- moderate. (0.2-0.3): shrub and grassland
- high (0.6-0.8): temperate and tropical forest
NDVI PROBLEMS
atmospheric effects:
actual composition of atmosphere, especially water vapour and aerosols, can affect measurements
clouds:
sensor can’t see through clouds + shadows/small clouds affect measurements
soil effects:
wet soils are darker, reflectance a direct function of water content
anisotropic effects:
all surfaces reflect differently in different directions
spectral effects:
each sensor has its own wavelength bands NDVI yields different values with different instruments
Radiometric correction
correcting data for sensor irregularities and unwanted sensor / atmospheric noise; converting data so they accurately represent the reflected / emitted radiation measured by sensor
important because:
different images may need to be compared from different sensors / different times of day
Geometric corrections
correcting for geometric distortions due to sensor-Earth geometry variations and conversion of data to real world coordinates (latitude and longitude) on Earth’s surface
important because:
earth’s surface is curved and some satellites don’t look straight down as they are restricted by orbit and will instead angle the sensor
Spatial filtering
enhances or suppresses specific spatial patterns in an image
this technique explores distribution of pixels of varying brightness over an image and can detect and sharpen boundary discontinuities
types: -low pass -high pass -edge enhancers and detectors (all use histograms either for classification or filtering to determine clusters of pixel types)
what is the result of the low pass filter?
- reduces overall variability of an image
- lowers contrast
- reduces noise
- blurs appearance of an image
why?
-emphasises large homogeneous areas of similar tone and reduce small detail in an image
- smooth appearance of an image, remove random/periodic noise and reveal background pattern
- average and median filters, often used for radar imagery
high pass filter kernel
if there is no change in intensity, nothing happens. if one pixel is brighter than its immediate neighbours, it gets boosted
the sum of all ‘weights’ in the kernel is zero, the centre pixel is enhanced (has highest weight)
Post processing: image classification
- classifies pixels of raster image to certain categories
- makes use of multispectral images, the more bands the more accurate the classification
- unsupervised: computer based approach that assigns pixels to classes automatically, after the process you have to identify what the categories are (normally good if you can’t access the area/undertake ground-truthing)
- supervised: user defines the training areas and tell the computer, then run classification which places areas into the classes and categories of interest
