Frequent Itemset Mining and Association Rules in Visual Data

Till Quack
Eidgenössische TH Zürich-Hönggerberg

We present an application of the traditional itemset mining techniques in the non-traditional domain of visual data. By mining frequent configurations of local visual features, we are able to collect evidence for the presence of objects or object classes. We first demonstrate how this technique can be used to mine frequently occurring objects in video data.
The second part of the talk describes an extension of this technique to mine configurations of features which are characteristic for a class of objects, such as cars, bikes or giraffes. Frequent feature configurations are mined from training data for each object class. Based on the mined configurations we develop a method to select features which have high probability of lying on previously unseen instances of the object class. The technique is meant as an intermediate processing layer to filter the large amount of clutter features returned by lowlevel feature extraction, and hence to facilitate the tasks of higher-level processing stages such as object detection.

This is joint work with Vittorio Ferrari and Luc Van Gool.

Back to Workshops II: Numerical Tools and Fast Algorithms for Massive Data Mining, Search Engines and Applications