COMPARISON OF OBJECT BASED FEATURE EXTRACTION ALGORITHM WITH PIXEL BASED ALGORITHM IN CLASSIFICATION OF A COMPLEX URBAN ENVIRONMENT.

Abstract

Most studies on land use land cover uses the traditional pixel-based classification which analyses pixels using multivariate statistics such as the mean, minimum -maximum, covariance. the pixels are then grouped using their spectral values. Urban areas are complex and have high degree of heterogeneity leading to mixed problems while using pixel-based classification. Object based classification differs from pixel-based classification by including both segmentation and the use of vector layers for image analysis. segmentation partitions remote sensing imagery into meaningful homogeneous objects which delimits patterns observed with the naked eye. The aim of my proposed project is to assess the accuracy of the object-based analysis of high-resolution satellite imagery in relation to complex urban environment and to compare this method to standard pixel-based approaches. Use of vector layers in performing object-based classification of high-resolution satellite imagery will also be studied. The project will also seek to assess the suitability of object-based satellite image analysis for landcover studies in a complex urban environment. The study will focus on Imara Daima ward, Nairobi city. Ecognition software will be used for object-based analysis of high resolution imagery together with QGIS for GIS analysis.