Color and Texture Features Based Object Recognition Using Machine Learning Methods

Abstract

Object recognition in images is a simple task for human but it is a complex and challenging task for machines due to different factors such as occlusion, lightening, object size, scaling etc. A Robust and automatic image processing system is thus critically required to make such a sampling approach practical. In this paper, we propose a method for classification of objects in images by incorporating global descriptors of the image such as color moment features along with wavelet Packet Entropy, which yields translation, rotation, and scaling invariant recognition. To recognize the objects, we have evaluated three machine learning methods-Classification and Regression Tree (CART), Support Vector Machines (SVM), and AdaBoost and compared their performances. We evaluate them on the publicly available WANG dataset of 500 images. K-fold (for K=10) cross-validation method is used to train and test the model. From experimental results, we see that SVM shows promising performance, however, the training time is extremely slow in SVM. CART is particularly attractive in terms of computational speed but shows less accuracy than SVM. Ensemble method, AdaBoost is superior in accuracy with promising computational speed than other methods used here.