Probabilistic Latent Semantic Analysis for Sketch-based 3D Model Retrieval

Sketch-based 3D model retrieval has recently attracted a lot of interest in the content-based 3D model retrieval community. We propose a novel method to solve this problem by using a latent topic model probabilistic Latent Semantic Analysis (PLSA). PLSA is an unsupervised learning method used to learn the latent topics from text documents. Although pLSA is first proposed in the statistical text analysis and retrieval field, it has been demonstrated to be an effective topic model widely used in computer vision research. In this paper, a bag-of-features (BOF) model based on local features of sketches is employed to obtain the visual word vocabulary, and then the pLSA model is used to learn the latent semantic topic representations for sketches based on their visual word representations. A query sketch is matched with a 3D model within the latent semantic topic space to alleviate the semantic gap and decrease the matching time. We conduct our experiments based on the most popular local shape features in order to have a comprehensive study of this topic model for sketch-based 3D model retrieval. The experimental results on the common sketch-based watertight model benchmark show that our approach significantly outperforms the original word-occurrence statistic methods.