With the progressive advancement in Artificial Intelligence, Deep learning techniques have stood out with its remarkable performance in various computer vision applications like image classification, object detection, segmentation. A feature fusion methodology which bridges the gap between the machine and deep learning technique, captivating the benefits of both the field is proposed. Machine level features – color, texture, shape which describes an image are fused with the higher level semantic features extracted by state-of-the-art technique. Convolutional Neural Network (CNN) based Vgg16 architecture is used to as feature extractor which are later fused with machine level features. The performance of the algorithm is measured on Caltech-101and Caltech-256 object category datasets using various machine learning classifiers. The proposed methodology with its powerful fused descriptor outperforms the state of the art complicated models.

Feature Fusion, Vgg16, Machine Learning, Caltech101, Caltech256.