A binary classification and online vision

Abstract

In this study we consider vision as a binary classification problem, where an ensemble of decision tree based classifiers is trained on-line, new images are continuously added and the recognition decision is made without delay. The ensemble of decision tree classifier is combined into a forest classifier using averaging, generate on-line random forest (RF) classifier. First we employ object descriptor model based on bag of covariance matrices, to represent an object visual features then run our online RF learner to select object descriptors and to learn an object classifier. Validation of the method with empirical studies in the domain of the GRAZ02 dataset shows its superior performance over those of histograms based, subsequently yields in object recognition performance comparable to the state-of-art standard RF, AdaBoost, and SVM classifiers, even when only 10% training examples are used.