Automatic view selection: An application to image mining

Abstract

In this paper we discuss an image mining application of Egeria detection. Egeria is a type of weed found in various lands and water regions over San Joaquin and Sacramento deltas. The challenge is to find a view to accurately detect the weeds in new images. Our solution contributes two new aspects to image mining. (1) Application of view selection to image mining: View selection is appropriate when a specific learning task is to be learned. For example, to look for an object in a set of images, it is useful to select the appropriate views (a view is a set of features and their assigned values). (2) Automatic view selection for accurate detection: Usually classification problems rely on user-defined views. But in this work we use association rule mining to automatically select the best view. Results show that the selected view outperforms other views including the full view. © Springer-Verlag Berlin Heidelberg 2005.