A Novel Sketch-based 3d Model Retrieval Approach by Integrating Pca-daisy Descriptor and Fisher Coding Algorithm

Since the local feature-based methods recognize interior content and utilize the rich information of the sketches and 2D views of 3D model, this paper proposes a novel sketch-based 3D model retrieval algorithm which utilizes local feature descriptors to describe the object shape. Firstly, we propose PCA-DAISY descriptor to create the efficient local feature descriptors. The DAISY descriptor consists of a vector from the convolved orientation maps with Gaussian filters. The Principal Component Analysis (PCA) technique is applied to reduce the dimensionality of DAISY descriptor, which is not only useful in reducing the dimensionality, but also beneficial in reducing the error rate. Then, our Fisher coding algorithm quantizes the PCA-DAISY descriptor by using the Gaussian Mixture Model (GMM), which can be understood as a “probabilistic visual vocabulary” and brings large improvements in accuracy. The experimental results demonstrate that our approach which combines these two algorithms significantly outperforms several latest sketch-based retrieval approaches.