Building detection with loosely-coupled hybrid feature descriptors

The paper presents a hybrid approach that ultilizes multiple low-level feature descriptors for performing building detection in 2D images. The proposed method is a symbiosis of two feature descriptors, namely Color and Edge Directivity Descriptor (CEDD) and Fuzzy Color and Texture Histrogram (FCTH) . The use of edge detection, texture and color combined features using fuzzy technique in encoding low-level visual information from images are embedded in the hybridization. First, multiple locations from a target image are chosen in the feature extraction process. Then, a hybridized vector index is proposed for measuring the low-level visual features distance between the target natural images with the training images, allowing a building content to be detected. Size and resolution of the source of images are not restricted in the proposed model and thus it can enhance the computational effectiveness. The empirical assessment, in term of the accuracy in detecting building objects in a set of images, validates the feasibility and potentiality of the proposed techniques.