What's that Style? A CNN-based Approach for Classification and Retrieval of Building Images

• Published in: 2017 Ninth International Conference on Advances in Pattern Recognition (ICAPR) pp. 1 - 6
• Main Authors: Meltser, Rachel D., Banerji, Sugata, Sinha, Atreyee
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2017
• Subjects: Buildings; Eigenvalues and eigenfunctions; Feature extraction; Principal component analysis; Support vector machines; Task analysis; Visualization
• DOI: 10.1109/ICAPR.2017.8593206

Image classification and content-based image retrieval (CBIR) are important problems in the field of computer vision. In recent years, convolutional neural networks (CNNs) have become the tool of choice for building state-of-the-art image classification systems. In this paper, we propose novel mid-level representations involving the use of a pre-trained CNN for feature extraction and use them to solve both the classification and the retrieval problems on a dataset of building images with different architectural styles. We experimentally establish our intuitive understanding of the CNN features from different layers, and also combine the proposed representations with several different pre-processing and classification techniques to form a novel architectural image classification and retrieval system.