Efficient 3D Object Retrieval Based on Compact Views and Hamming Embedding

View-based 3-D object retrieval techniques have been increasingly important in various applications of computer vision. In this paper, we present a novel framework for view-based 3-D object retrieval. First, we exclude the background of views to avoid the disturbance of background noise. Then for these views, we extract the domain-size pooled SIFT descriptor features and encode them using approximate K-means algorithm. After quantizing each object with the approximate near neighbor, the hamming embedding is applied to refine the descriptors by adding binary signatures. Finally, we use the hamming matching to measure the similarity between two 3-D objects. A large number of experiments are performed on the ETH-80 benchmark. Compared with the state-of-art methods, the proposed method is demonstrated to be effective and robust.