Machine Learning and Friends Lunch

home
past talks
resources
conferences

Using All Available Information: Combining Feature Vectors and Bags of Features for Object Class Recognition.

Abstract

Many classification techniques expect class instances to be represented as feature vectors, i.e. points in a feature space. In computer vision classification problems, it is often possible to generate an informative feature vector representation of an image, for example using global texture or shape descriptors. However, in other cases, it may be beneficial to treat images as variable size unordered sets or bags of features, in which each feature represents a salient image structure or patch. We show that combining feature vectors and bags of features is beneficial in an application where rough segmentations of objects are available. We present a method for classification with bags of features using non-parametric density estimation. Subsequently, we present a methods for utilizing both representations for image classification via ``stacking'', an ensemble technique. Experimental results show the superior performance of this method over the classifiers using either representation exlusively, with a reduction of over 20% in the error rate on a challenging marine science application.