Evaluation of Texture Energies for Classification of Facade Images

Texture is one of the most important features for object detection and recognition. In many applications, it is derived from the responses of texture filters. In this paper, we evaluate the potential of seven texture filter banks for the pixel-based classification of terrestrial facade images. Particularly, we analyze features from Gabor, Haar, Root Filter Set, and Walsh filters as well as filters that have been proposed by Laws (1980), Leung and Malik (2001), and Schmid (2001). We determine texture energies similar to the approach of Laws (1980) using the proposed filter banks, and then we classify the derived feature vectors by three different methods: maximum a posteriori probabilities (MAP), linear discriminant analysis (LDA) and random forest (RF), respectively. In all three cases, we obtained best classification results with Haar, Laws, and Walsh filters.