A Study on Ensemble feature learning for material recognition

Recognition of variety of materials that is present in our surroundings plays an important role in wide variety of applications like robotics, material-based product search, image editing for interior designing etc. Hence it is necessary to propose a system to automatically recognize the material recognition. The different category of materials includes all kinds of fabric, foliage, glass, leather, metal, paper, plastic, stone, water and wood. This paper studies and compares knowledge-based classifiers with ensemble classifier to achieve automatic material recognition. The comparison study is carried on open datasets such as MINC-2500 dataset. Initially, the pre-processed input images from the training set are then fed to VGG16 and ResNet50 network. Machine learning models such as linear SVM, logistic regression and random forest are trained on top of these extracted features to obtain a classifier that can recognize the classes of materials. The experimental results exhibits that the ensemble feature learning of SVM and logistic regression through ensemble voting classifier achieves higher recognition accuracy of 80.05% than the traditional knowledge-based classifiers.