Object recognition from remotely sensed imagery is an unsolved problem that is, in part, difficult because objects are typically observed in a variety of contexts. Of particular interest are general purpose methods for identifying important image structure robust to changes in context. In this respect, areas not related to the task at hand can be ignored, or de-emphasized, when features are extracted and subsequently used for classification. In this article, we utilize an unsupervised differential morphological profile (DMP) technique for the identification of important image structure of varying scales. The DMP levels are fused using a Choquet fuzzy integral with a bias towards identifying a general class of scale related objects via density selection. Fused DMP opening and closing results are compared, an entropic filter is used to further accentuate important regions and an eigen-based image chip alignment method is performed. Next, we propose importance map weighted cell-structured local binary pattern (LBP) and histogram of oriented gradient (HOG) multi-scale pyramid descriptors. The final descriptor is used in support vector machine-based classification. Cross validation and left out performance for three common aerial object classes at multiple geographical sites is discussed. Advantages of this work include higher positive detection rates, lowered false alarm rates and an improvement in robustness of object recognition.
展开▼
