Texture-based Fuzzy System for Rotation-invariant Classification of Aerial Orthoimage Regions

Orthoimages are aerial images where feature displacements and scale variations have been removed. This type of images is widely used to calculate areas, determine land cover and land use, among others. This paper introduces a rotation-invariant classification model for three common orthoimage regions: city, sea and forest areas, using only texture information (without color information). Our classification model analyzes small sub-images (for example, of 20x20 pixels) to determine their region classes. Our model is based on a Fuzzy Inference System (FIS) constructed over a set of new rotation-invariant texture features. The features are extracted using two rotation-invariant versions of the well-known grayscale co-occurrence matrix (GLCM). Rotation-invariance is a desirable property of orthoimage classification systems, because the aerial images can be taken from different angles. We executed tests on samples from the three regions, including several rotated versions. These experiments show that our system reaches 100% of correct classification rate for our image test database. This correct classification rate is far superior to the rate obtained using the classical GLCM without the rotation-invariant property. Our classifier is robust to images that contain small areas that do not belong to the overall region type. The results demonstrate that our model offers a reliable rotation-invariant orthoimage region classification. This work was financially supported by CAPES. * Corresponding author.