Bookmark and Share

Computer Vision Metrics: Chapter Four (Part G)

Register or sign in to access the Embedded Vision Academy's free technical training content.

The training materials provided by the Embedded Vision Academy are offered free of charge to everyone. All we ask in return is that you register, and tell us a little about yourself so that we can understand a bit about our audience. As detailed in our Privacy Policy, we will not share your registration information, nor contact you, except with your consent.

Registration is free and takes less than one minute. Click here to register, and get full access to the Embedded Vision Academy's unique technical training content.

If you've already registered, click here to sign in.

See a sample of this page's content below:


For Part F of Chapter Four, please click here.

Bibliography references are set off with brackets, i.e. "[XXX]". For the corresponding bibliography entries, please click here.


Kernel Machines

In machine learning, a kernel machine [362] is a framework allowing a set of methods for statistically clustering, ranking, correlating, and classifying patterns or features to be automated. One common example of a kernel machine is the support vector machine (SVM) [341].

The framework for a kernel machine maps descriptor data into a feature space, where each coordinate in the feature space corresponds to a descriptor. Within the feature space, feature matching and feature space reductions can be efficiently carried out using kernel functions. Various kernel functions are used within the kernel machine framework, including RBF kernels, Fisher kernels, various polynomial kernels, and graph kernels.

Once the feature descriptors are transformed into the feature space, comparisons, reductions, and clustering may be employed. The key advantage of a kernel machine is that the kernel methods are interchangeable, allowing for many different kernels to be evaluated against the same feature data. There is an active kernel machine community (see kernel-machines.org).

Boosting, Weighting

Boosting [381] is a machine learning concept that allows a set of classifiers to be used together, organized into combinatorial networks, pipelines, or cascades, and with learned weights applied to each classifier. This results in a higher, synergistic prediction and recognition capability using the combined weighted classifiers. Boosting is analogous to the weighting factors used for neural network inputs; however, boosting methods go further to combine networks of classifiers to create a single, strong classifier.

We will illustrate boosting from the Viola Jones method [146,186] also discussed in Chapter 6, which uses the ADA-BOOST training method to create a cascaded pattern matching and classification network by generating strong classifiers from many weak learners. This is done through dynamic weighting factors determined in a training phase, and the method of using weighting factors is called boosting.

The idea of boosting is to first start out by equally weighting the detected features— in this case, HAAR wavelets—and then matching the detected features against the set of expected features; for example, those features detected for a specific face. Each set of weighted features is a classifier. Classifiers that fail to match correctly are called...