No label defined (Q4780247)

This booklet is the reprint of a thesis. It treats of object representation by means of set of views of its different aspects. The main application being here, face tracking or recognition. The thesis is divided in seven chapters: After a short introduction chapter 2 presents the notion of elastic graph matching based on local image description, which is the core of the developed methods. Chapter 3 is devoted to show how the model can be constructed with little knowledges about the object at hand (namely faces), what are the local transforms relevant to both tracking and recognition? Chapter 4 is devoted to two applications of the schemes developed in chapter 3, namely, landmark finding on faces and finger tracking for imitation learning. Chapter 5 to 7 consider the problem of creating an object representation without assumptions on the observer-object relation. The proposed solution integrate local representation in a view ``manifold''.