A study of factors that influence the accuracy of content-based geospatial ranking systems

Visual patterns found in geospatial images are complex, dynamic and difficult to be articulated by human analysts; let alone building a computational model to understand the intertwining semantics in the images. Advancements in image collection and pre-processing have led to a need for identifying the factors that affect content-based geospatial retrieval systems. In this article, we study the factors that influence the semantic assignment precision when varying semantic space complexity and training set size. We test their influence using different data mining algorithms. Our findings provide some new insights for future research in training image retrieval systems under various conditions related to semantic mixture, feature space overlapping and size of training data set.