A Comparison of Pixel Based and Object Based approach for Land Use/Land Cover Classification: A Case Study in Mediterranean Region

Gürkan Aysel
2017 Journal of Scientific and Engineering Research   unpublished
This paper aims to compare between application of pixel-based and object-oriented classifications of Mediterranean region in Turkey. In pixel-based classification a supervised Maximum Likelihood Classification (MLC) algorithm was utilized; in object oriented classification, a soft nearest neighbour classifier were used. The classification data was a high resolution with 0.50 meter of WorldView-2 2014 imagery. Classification results were compared in order to evaluate the suitability of the two
more » ... bility of the two classification techniques. The study comprised 3 stages; first stage is pre-processing of imagery, second stage is classification, third stage is accuracy assessment of these classifications. The object-based classifier achieved a high overall accuracy (90.24%), whereas maximum likelihood classifier produced a lower overall accuracy (85.58%). This study demonstrates that the object-based classifier is a significantly better approach than the pixel based classifiers. Introduction For more than 40 years, satellite images and aerial photographs have formed a strong basis for land cover classifications. Pixel-based land cover classification methods, such as maximum likelihood classification, use the spectral information contained in individual pixels to generate land cover classes. This method has been shown to perform accurately for the classification of certain land use/cover classes and has proven accurate in change detection analysis [1-2]. In particular, object-based image analysis (OBIA) techniques enable connecting information from the image with database information [3-4]. Object-oriented classification does not operate directly on single pixels, but objects consisting of many pixels that have been grouped together in a certain way by image segmentation [5-6]. Object-based image analysis is quickly gaining acceptance among remote sensors, and has demonstrated great potential for classification and change detection, compared to pixel-based approach [7-9]. The advantage of the object-based approach is that it offers new possibilities for image analysis because image objects can be characterised by features of different origin incorporating spectral values, texture, shape, context relationships and thematic or continuous information supplied by ancillary data. Integration of additional knowledge is a valuable means to distinguish ecologically meaningful habitat types that don't have