Chapter 9 - Object Categorization Using Adaptive Graph-Based Semi-supervised Learning

Abstract

In this chapter, we introduce a new efficient graph construction method that can be used by many automatic learning tasks. Unlike the mainstream, our proposed method simultaneously estimates the graph structure and its edge weights through sample coding via data self-representativeness. Compared with the recent ℓ1 graph based on sparse coding, the proposed coding scheme has an analytical solution and thus is more efficient. The chapter has two main contributions. Firstly, it introduces a principled two-phase weighted regularized least square graph construction method that exploits a collection of images. Secondly, the obtained data graph is used, in a semi-supervised context, in order to categorize detected objects in outdoor and indoor scenes. The graphs are built by using Local Binary Patterns (LBPs) as image descriptors. The experiments show that the proposed method can outperform competing methods including the recent sparse graphs.