Histogram Curve Matching Approaches for Object-based Image Classification of Land Cover and Land Use

Area of Science:

• Remote Sensing
• Geographic Information Systems (GIS)
• Image Analysis

Background:

• Traditional image object classification relies on statistical measures like mean or standard deviation.
• Parametric statistical measures may not fully capture the spectral characteristics of image objects.

Purpose of the Study:

• To analyze digital number histograms of image objects.
• To evaluate classification measures that utilize histogram signatures.
• To compare histogram matching classifiers against a standard nearest neighbor to mean classifier.

Main Methods:

• Utilized an ADS40 airborne multispectral image of San Diego, California.
• Applied two histogram matching classifiers and a standard nearest neighbor to mean classifier.
• Performed classifications using spatial resolutions of 0.5 m, 2.5 m, and 5 m within a Geographic Object-Based Image Analysis (GEOBIA) framework.

Main Results:

• Digital number histograms proved to be reliable features for class characterization.
• Both histogram matching classifiers demonstrated superior performance compared to the standard nearest neighbor to mean classifier.
• The highest classification accuracies were achieved using images with 2.5 m spatial resolution.

Conclusions:

• Histogram matching is a robust method for image object classification.
• The choice of spatial resolution significantly impacts classification accuracy.
• Histogram-based features offer advantages for characterizing spectral signatures in GEOBIA.