Performance comparison of Support Vector Regression and Relevance Vector Regression for facial expression recognition

This paper compares the performance of Relevance Vector Regression and Support Vector Regression for the purpose of facial expression recognition. The Support Vector Machine (SVM) is a state-of-the-art technique for regression and classification, but lacks the probabilistic treatment which is overcome by Relevance Vector Machine (RVM). Though SVM's have a good generalization performance, but their results are in general less sparse. This sometimes results in almost all of the training data to be used as Support Vectors. Comparing with RVM, the results obtained are relatively more sparse than SVM which results in lesser number of Relevance Vectors ultimately leading to lesser computation overhead. The above models are compared for facial expression recognition on Cohn Kanade database. Local Binary Pattern features are extracted from facial images. These are preprocessed for illumination and size, and also for dimensionality reduction before being used for training the RVM and SVM models. The paper concludes with a comparison of the SVM and RVM on the basis of test results.