The effects of viewing- and illumination geometry on settlement type classification of quickbird images

Image texture features extracted from high-resolution remotely sensed images over urban areas have shown promise in their ability to distinguish different settlement classes. Without any explicit mechanism to counter the effects of variable illumination- and viewing geometries, these features may not generalize well in multi-date applications such as change detection. This paper presents the results of a small study of the effects of unwanted variability on low-income settlement classification performance in the Soweto residential area of the city of Johannesburg, South Africa. Somewhat surprisingly, the Gray-Level Co-occurrence Matrix (GLCM) features were found to perform better than Local Binary Pattern (LBP) features on combined spatial and temporal generalization tasks, although the LBP features offered better performance on spatial-only generalization problems.