3D Object retrieval based on viewpoint segmentation

Abstract

© Springer-Verlag Berlin Heidelberg 2015 including computer vision, pattern recognition, and computer-aided design, demonstrating its vitality and importance. The initial attempts of this domain try to retrieve 3D models via textual annotations, which is however, hampered by two aspects. First, annotations may be incomplete and ambiguous due to the provider's language, culture, career habits, and other aspects. Second, the model features are difficult to describe in only a few words in most cases. Therefore, content-based 3D model retrieval becomes a reasonable solution, which concentrates on the model's low-level characteristics.Recently, a great number of algorithms have been proposed in the field of 3D object retrieval [9, 10, 12, 13, 15, 16, 23-28, 33-35, 68]. For the model-based methods, different features of 3D models are presented, including geometric moment [45], surface distribution [42], volumetric-based information [7], shape descriptors (e.g., Zernike moments [20] and Fourier descriptors [58]). However, expensive computation of obtaining the features and unsatisfactory retrieval performance limit their applications. On the contrary, view-based methods overcome these disadvantages, and show their better retrieval performance than model-based ones. There are many existing view-based methods, such as lighting field descriptorLFD) [4], elevation descriptor (ED) [49], adaptive view clustering (AVC) [2], Panorama [44], compact multi-view descriptors (CMVD) [5], and camera constraint-free viewbased method (CCFV) [11]. LFD represents a 3D object using 20 views from the vertexes of a dodecahedron, and the features are Zernike moments and Fourier descriptors extracted from the views [4]. ED describes the altitude information of a 3D object by obtaining six elevations from the front, top, right, rear, bottom and left views [49].Later, AVC, CMVD and CCFV consider the similarities between different views and represent a 3D object by the Abstract In the last decades, extensive efforts have been dedicated to develop better 3D object retrieval methods. View-based methods have attracted a significant amount of attention, not only because of their state-of-the-art performance, but also they merely require some of a 3D object's 2D view images. However, most recent approaches only deal with the images' content difference without the discrepancy of view relative positions. In this paper, we propose a normal method for view segmentation, based on Markov random field (MRF) model, which consider not only the difference between the content of views but also the relative locations. Each view is obtained by projecting at certain viewpoints and angels, therefore, these locations can be applied to depict each view, with content of views. We use the MRF to implement view segmentation and choose the representative views. Finally, we present a framework based on the proposed view segmentation method for 3D object retrieval and the experimental results demonstrate that the proposed method can achieve better retrieval effectiveness t...