COMPARISON OF PIXEL-BASED AND OBJECT-ORIENTED CLASSIFICATION METHODS

ABSTRACT: In remote sensing, different classification methods are performed to satellite imagery according to the value of accuracy required for the user. This study aims to compare the satellite images of Istanbul, Sarıyer which have different resolution and varying dates after pixel-based and object-oriented classification methods are performed. For this purpose, three different satellite images of Sarıyer area from the years 2004, 2010 and 2011,from Ikonos2, Spot 5 and Landsat 5 (TM) satellites were utilized. Pixel-based supervised classification process was performed on the all satellite images and object-oriented classification process was performed on only the Ikonos 2 satellite image. For testing to increase the accuracy of classification, supervised classification with samples obtained from object-based classification process was performed on all of the images. After classification processes, accuracy analysis was performed. Pixel-based and object-oriented classifications were performed by using Erdas-9.1 and eCognitiondeveloper-8 software packages and four different classes were achieved. These classes are; structures, green areas, wetlands, and empty fields. Subsequent the classification, total class accuracy was determined for all images. Various classification methods have been performed and after an accuracy analysis results were compared