Extended local binary pattern features for improving settlement type classification of quickbird images

Abstract

Despite the fact that image texture features extracted from high-resolution remotely sensed images over urban areas have demonstrated their ability to distinguish different classes, they are still far from being ideal. Multiresolution grayscale and rotation invariant texture classification with Local Binary Patterns (LBPs) have proven to be a very powerful texture feature. In this paper we perform a study aiming to improve the performance of the automated classification of settlement type in high resolution imagery over urban areas. That is, we combined the LBP method based on recognising certain patterns, termed “uniform patterns” with the rotational invariant variance measure that characterises the contrast of the local image texture, then combined multiple operators for multiresolution analysis. The results showed that the joint distribution of these orthogonal measures improve performance over urban settlement type classification. This shows that variance measure (contrast) is an important property when classifying settlement types in urban areas.