Land use classification for urban areas using spatial properties

The cover frequency algorithm (CFA) has been applied to prepare land use maps of two cities, Lucknow (India) and Assuit (Egypt). The overall classification accuracy improved by 10% and 15% in case of training and test area respectively for the well-planned Assuit city. The overall and individual class accuracies of most of the classes reach 85% or higher suggesting usefulness of CFA. On the other hand CFA does not show any advantage in classifying Lucknow which is relatively poorly planned. A comparison of classification using CFA to that using a combination of three spectral and one texture features computed from grey level co-occurrence matrix (GLCM) algorithm indicated that for the well-planned Assuit city, the improvement in accuracy using both approaches were similar. However, CFA is computationally simple and unlike texture algorithms it takes shorter time and requires optimisation of only one parameter i.e. window size. However, for a heterogeneous and poorly planned city like Lucknow, one should use combined spectral and textural features rather than CFA to improve classification accuracy