Chapter 4 - Supervised Dictionary Learning via Semantic Embedding

Abstract

The visual dictionary and its indexing structure serves as the core of visual local representation systems. In Section 4.2, learning-based local interest-point detectors are investigated. Subsequently, Section 4.2.1 investigates the problem of unsupervised visual dictionary learning. However, as stated before, the building of a human visual system typically involves heavy learning procedures, either on the evolution of mankind or during growth of individual people. In other words, environmental supervision performs an important role in the visual representation system of a human being. In this chapter, similar ideas are discussed to investigate the possibility of learning-based visual representation.