Visual Object Detection using Frequent Pattern Mining

Object search in a visual scene is a highly challenging and computationally intensive task. Most of the current object detection techniques extract features from images for classification. From the results of these techniques it can be observed that the feature extraction approach works well for single images but are not sufficient for generalizing over a variety of object instances of the same class. In this work we try to address this problem by using a well known machine learning technique, namely, frequent pattern mining. The approachwe use here is to find frequently occurring patterns of visual properties across the whole set of images in the class. The frequent patterns thus foundwould potentially represent those features which bind together the images of that class. Shape, color, texture, spatial orientation etc., or any combinations of these can be used as the visual properties. During the testing phase the object presence is detected by analyzing the images for the presence of these learned patterns. The proposed framework has been tested with Caltech 101-object dataset and the results are presented.