中文摘要：

针对光场描述符丢失三维模型空间信息以及全景视图描述符投影视图较少的问题,提出一种结合深度图像的三维模型检索算法.该算法引入深度图像对光场描述符加以改进,得到投影视图后,分别提取其离散小波变换特征和Zernike矩特征;然后对深度图像进行聚类去掉冗余信息,并通过随机游走算法来确定每一类的权重,以更好地反映类间关系;最后设计改进全景视图相似距离计算方法,用于进一步的三维模型检索.在普林斯顿模型库上的实验结果表明,该算法有效利用了三维模型空间信息,提高了检索精确度.

英文摘要：

Due to the problem of light field descriptor lacking spatial information of 3D model,and the problem of little projection of PANORAMA descriptor,a 3D model retrieval algorithm combining depth image was proposed. The depth images are used to improve light field descriptor. Discrete wavelet transform features and Zernike moments are extracted from projection views. Then the depth images are clustered to remove redundancy. The random walk algorithm is employed to determine the weight of each cluster. Finally,the retrieval method of PANORAMA is optimized to calculate similarity distance between two 3D models. Results on princeton shape benchmark show that the proposed algorithm can make good use of spatial information of 3D model and improve the accuracy of retrieval.