A Comparison of Pixel-Based and Object-Based Classification Approaches in Arid and Semi-Arid Areas of Dead Sea Region Using Landsat Imagery

Abstract

In this study, land cover types in Dead Sea area were analyzed on the basis of the classification results acquired using the pixel based and object-oriented image analysis approaches. A subset of Landsat TM satellite data was used for a comparison between pixel-based and object-based classification approaches. Ground truth data were collected from the available maps, aerial photographs, and personal knowledge. In pixel-based supervised classification, the spectral information for each pixel is utilized as the basis of categorization, the result shows that there are some small areas of anomalous pixels (salt and pepper) representing the noise in the data within the same class. On the other hand, object-oriented image analysis was evaluated based on object characteristics. This approach classifies not single pixels but groups of pixels that represent already existing objects in the field represent the n dimensional feature space for the classification. The result indicated an overall accuracy of 81.6% for pixel-based approach and 80.7% for object-based approach. The accuracy of the vegetation class is improved from 69% (for pixel-based) to 78.5% (for object-based). Outcome from the classification show that the object-oriented approach shows more accurate boundary results especially for agricultural land cover classes than those achieved by pixel-based classification algorithms.