Automatic extraction of buildings and trees using fuzzy K-means classification on high-resolution satellite imagery and LiDAR data

Clustering methods play an important role in automatic land cover classification issues when little previous information about baseline data is available. Particularly, K-means algorithm tends to generate good results in classification when data of different nature are is available. It shows some strength as these good results are achieved in many different applications in the multivariate analysis. However, a limitation noticed in the classifications made with this algorithm is not to produce sets with homogeneous texture, despite deliver acceptable ratings, there is a presence of noise in the resulting image. The current research compares this method with its variant in the field of fuzzy logic (Fuzzy K-means). The comparison have been applied for automatic extraction of buildings and trees over high resolution satellite and LiDAR data acquired over an area of Alcalá de Henares city. A fuzzy set has a relaxed membership function. This assumption is what led us think that if the classifier considers clusters as fuzzy sets instead of classic sets, as the algorithm K-means does, noise will be decreased in the resulting image without adversely affecting the ability of the classifier to work consistently. To evaluate the obtained results a ground truth from aerial imagery and field visits have been created in order to determine which classifier achieves the best results.