Historic Chinese architectures image retrieval by SVM and pyramid histogram of oriented gradients features

Content-Based Image Retrieval (CBIR) of historic Chinese architecture images is an important area of research bridging society, culture and information technology. Most of the image features used in previous content-based image retrieval systems such as colour, texture and some simple shape descriptors are not effective in describing building images due to high variability in the heterogeneous architectural image collections. This study investigates content-based architectural image retrieval mainly by shape features. The recently proposed shape descriptor, Pyramid Histogram of Oriented Gradients (PHOG) features, counts occurrences of gradient orientation in localized portions of an image and has been proved as an efficient tool for providing spatial distribution of edges. Many existing image retrieval systems attempt to compare the query image with every target image in the database to find the top matching images, resulting in an essentially linear search which is prohibitive when the database is large. To solve the problem, it propose to introduce classification as the first stage in the retrieval system. Based on the PHOG features, it apply the Support Vector Machine (SVM) to automatically classify the ancient Chinese architecture images. Cross-validation test results indicate that the generalization performance of the SVM was over 60% compared to neural network's accuracy of 30% and kNN's accuracy 50%.