Evaluation of Object-Oriented and Pixel Based Classification Methods for extracting changes in urban area

This study focuses on the comparison between three image classifications of remote sensing imagery to estimate change detection in urban area (Tabriz, Iran) by Post Classification Comparison (PCC) technique. In order to investigate an appropriate method for extracting changes, pixel-based image classifiers such as: Maximum likelihood Classifier (MLC), Neural Network Classification (NN) and an Object-Oriented (OO) image Classifier were tasted and compered by using Landsat TM and ETM+ image respectively belong to 1990 and 2010. A priori defined five land cover classes in classification scheme were built-up, vegetated area, bare areas, water bodies and roads.

The accuracy of each method was assessed using reference dataset from high resolution satellite image and aerial photograph. The results shows that the MLC method has achieved an overall accuracy of 69% with kappa coefficient of 0.67, NN 92% overall accuracy with a kappa coefficient of 0.90 and object oriented with 94% overall accuracy and 0.92 coefficient kappa. The PCC technique reliably identified area of change was used as the change detection technique for evaluating the three classification method. Different classification methods have their own advantages and disadvantages. Quantitative results shows that, the MLC method is found to be unable to differentiate urban areas such as Built-up, Roads and Barren, the Object-Oriented Classifier has superior performance in classifying Vegetation and Water areas and the Neural Net Classifier also has the best performance in classifying Built-up, Road and Barren areas.