Abstract: This study introduces an efficient view based algorithm for 3D non-rigid object retrieval that accepts all the models. The algorithm is based on describing the 3D Model using multi-scale local visual features. To reduce the cost of distance computation and features storage, the extracted features integrated into histogram using Bag of Features (BoF) approach. The proposed algorithm used only 42 depth images to extract the features which reduced the runtime. The proposed Enhanced Ray Tracing algorithm was used for generating best quality depth images with great reduction in runtime also. The codebook for the vector quantization is learned via. our proposed Modifed Extremely Randomized Clustering trees (MERC-trees). In comparison with other algorithms, the proposed algorithm achieved high performance on SHREC’11 and 15 datasets; the well-known benchmarks. Experimental results show that, the proposed algorithm is very fast, achieved high retrieval result and robust against 11 types of different transformations with different levels of strength when tested on SHREC’11-Robust dataset. The proposed algorithm achieved the 6th best performance among 37 methods participated in SHREC’15 contest which none of them was view based and all used topological based features that are easier to discriminate the non-rigid objects.
H. Ahmed, H. Shdeed, S. Hamad and D. Hegazy, 2017. A New View Based Algorithm for Non-Rigid 3D Object Retrieval. Asian Journal of Information Technology, 16: 576-587.