Digital Image Processing & Integration with GIS
Idea in plain words
Pre-Processing: Radiometric correction (sensor calibration, atmospheric haze removal) and Geometric correction (orthorectification using DEM).
Exam tip
High-yield exam topic (Very High).
Explanation
Raw satellite data contains distortions caused by Earth rotation, topographic relief displacement, and atmospheric aerosols. Orthorectification corrects relief displacement on pixel-by-pixel bases using a DEM, converting perspective satellite scenes into planimetrically true map projections. Digital image classification assigns land cover identities to pixels. In Supervised Classification using Maximum Likelihood, pixels are assigned to the class for which they have the highest posterior probability based on multivariate Gaussian training statistics.
Key points
- Image Enhancement: Contrast stretching (linear, histogram equalization) and Spatial filtering (low-pass smoothing, high-pass edge detection).
- Supervised Classification: Analyst selects representative training sites for each land cover class; algorithms: Maximum Likelihood, Random Forest, Support Vector Machines.
- Unsupervised Classification: Algorithm clusters spectral data into statistical groups without prior training data; algorithms: K-Means, ISODATA.
- Accuracy Assessment: Error matrix (Confusion Matrix) calculating Overall Accuracy and Cohen's Kappa coefficient (kappa > 0.8 indicates strong agreement).
- GIS Integration: Exporting classified raster grids into polygon vector layers for overlay with cadastral parcels and administrative boundaries.
Example
Example
Applying a 3x3 Sobel high-pass filter over satellite imagery to extract linear fault lines, road networks, and geological structural lineaments.
Code examples
/*
Classification Error Matrix (Confusion Matrix):
Reference (Ground Truth) Data
Class | Forest | Water | Urban | Row Total | User's Accuracy
Forest | 90 | 2 | 8 | 100 | 90/100 = 90%
Water | 1 | 95 | 4 | 100 | 95/100 = 95%
Urban | 5 | 3 | 92 | 100 | 92/100 = 92%
Column Total | 96 | 100 | 104 | Total=300 |
Producer Acc | 90/96 | 95/100| 92/104| Overall Accuracy = (90+95+92)/300 = 92.3%
*/
Exam questions
- [10 Marks] Explain the steps of Supervised Image Classification. How is classification accuracy evaluated using a Confusion Matrix and Kappa coefficient?
- [8 Marks] Differentiate between Supervised and Unsupervised classification (ISODATA vs Maximum Likelihood).