A new framework to enhance Evolution-COnstructed object recognition method

Abstract

This paper presents a new approach for reinforcing a recent method named Evolution-Constructed (ECO). ECO is an ensemble learning approach which combines predictions of multiple weak classifiers. Our framework includes two enhancements. First, we utilize different linear/nonlinear base classifiers. Second, we employ different ensemble types such as bagging and boosting to combine these classifiers. It is our hypothesis that there is no single combination of base classifier and ensemble type used on given datasets that leads to obtain maximum accuracy. Furthermore, we implement our proposed framework to process classifiers concurrently to enhance the running time. Experimental results showed that the proposed method has a competitive performance over contemporary methods for different object datasets. Keywords: Object recognition, ECO features, Adaboost, Random Forest, Feature descriptor