Local Feature Binary Coding for Approximate Nearest Neighbor Search

The potential value of hashing techniques has led to it becoming one of the most active research areas in computer vision and multimedia. However, most existing hashing methods for image search and retrieval are based on global representations, e.g., GIST [3], which lack the analysis of the intrinsic geometric property of local features and heavily limit the effectiveness of the hash code. In this paper, we propose an supervised local feature hashing framework, i.e., Local Feature Binary Coding (LFBC), for visual similarity search, in which the feature-to-feature (F2F) and image-to-class (I2C) structures are successfully preserved and combined together. Specifically, the F2F structure considers the pairwise relationship between local features in the original feature space. While, from a higher-level aspect, I2C structure reflects the connection between images and their corresponding classes, which is derived from [1]. The outline of the proposed method is illustrated in Fig. 1. It is worthwhile to highlight several properties of the proposed method: (1) Different with global representation based hashing, LFBC directly learns hashing function from local features and simultaneously preserves pairwise F2F and I2C structure, which is proved to be more effective for accurate retrieval. (2) Inspired by [2, 4], bilinear projection based hashing function is adopted in our method. Thus, the complexity of the eigen-decomposition, which is the cubic form of the dimensionality, will be significantly reduced. The corresponding integrated LFBC algorithm is depicted in Algorithm 1.