Pixel-Level Decisions based Robust Face Image Recognition
Alex Pappachen James · InTech eBooks · 2010
Face Recognition 66to provide deeper insight.In majority of Marr's work (Marr, 1982), he assumed and believed vision in humans to be nothing more that a natural information processing mechanism, that can be modelled in a computer.The various levels for such a task would be: (1) computational model, (2) a specific algorithm for that model, and (3) a physical implementation.It is logical in this method to treat each of these level as independent components and is a way to mimic the biological vision in robots.Marr attempted to set out a computational theory for vision in a complete holistic approach.He applied the principle of modularity to argue visual processing stages, with every module having a function.Philosophically, this is one of the most elegant approach proposed in the last century that can suit both the paradigms of software and hardware implementations.Gibson on the other hand had an "ecological" approach to studying vision.His view was that vision should be understood as a tool that enables animals to achieve the basic tasks required for life: avoid obstacles, identify food or predators, approach a goal and so on.Although his explanations on brain perception were unclear and seemed very similar to what Marr explained as algorithmic level, there has been a continued interest in the rule-based modeling which advocates knowledge as a prime requirement for visual processing and perception.Both these approaches have a significant impact in the way in which we understand the visual systems today.We use this understanding by applying the principles of modularity and hierarchy to focus on three major concepts: (1) spatial intensity changes in images, (2) similarity measures for comparison, and (3) decision making using thresholds.We use the following steps as essential for forming a baseline framework for the method presented in this chapter: Step 1 Feature selection of the image data: In this step the faces are detected and localized.Spatial change detection is applied as a way to normalize the intensity features without reducing the image dimensionality. How to referenceIn order to correctly reference this scholarly work, feel