Focusing attention on objects of interest using multiple matched filters

In order to be of use to scientists, large image databases need to be analy zed to create a catalog of the objects of interest. One approach is to appl y a multiple tiered search algorithm that uses reduction techniques of incr easing computational complexity to select the desired objects from the data base. The first tier of this type of algorithm, often called a focus of att ention (FOA) algorithm, selects candidate regions from the image data and p asses them to the next tier of the algorithm, In this paper we present a ne w approach to FOA that employs multiple matched filters (MMF), one for each object prototype, to detect the regions of interest. The MMFs are formed u sing K-means clustering on a set of image patches identified by domain expe rts as positive examples of objects of interest. An innovation of the appro ach is to radically reduce the dimensionality of the feature space, used by the k-means algorithm, by taking block averages (spoiling) the sample imag e patches. The process of spoiling is analyzed and its applicability to oth er domains is discussed. Combination of the output of the MMFs is achieved through the projection of the detections back into an empty image and then thresholding, This research was motivated by the need to detect small volca nos in the Magellan probe data from Venus. An empirical evaluation of the a pproach illustrates that a combination of the MMF plus the average filter r esults in a higher likelihood of 100% detection of the objects of interest at a lower false positive rate than a single matched filter alone.