Product image classification using Eigen Colour feature with ensemble machine learning

The plethora of e-commerce products within the last few years has become a serious challenge for shoppers when searching for relevant product information. This has consequently led to the emergence of a recommendation assistant technology that has the capability to discover relevant shopping products that meet the preferences of a user. Classification is a machine learning technique that could assist in creating dynamic user profiles, increase scalability and ultimately improve recommendation accuracy. However, heterogeneity, limited content analysis and high dimensionality of available e-commerce datasets make product classification a difficult problem. In this present study, we propose an enhanced product image classification architecture which has data acquisition pre-processing, feature extraction, dimensionality reduction and ensemble of machine learning methods as components. Core amongst these components is the Eigenvector based fusion algorithm that is meant to obtain dimensionality reduced Eigen Colour feature from the histogram of oriented gradient based colour image representative features. The ensembles of Artificial neural network and Support vector machine were trained with the Eigen Colour feature to classify product images acquired from the PI100 corpus into 100 classes and their classification accuracies were compared. We have obtained a state-of-the-art classification accuracy of 87.2% with the artificial neural network ensemble which is an impressive result when compared to existing results reported by other authors who have utilised the PI100 corpus.