How the human brain perceives images?—— 10 cognitive theory of Pattern Recognition guide

In many sci-fi movies, we can see that robots are legs, looking for and discovering enemies, far more than we humans see, accurate, powerful and wise, always saving people in the most dangerous time. In fact, these are still just sci-fi, not in the real world. One reason is that we can't "teach" robots to recognize landscapes like we do. Why don't we teach robots to recognize things as humans do? Not wanting to teach, but we have not yet figured out the working mechanism of human vision system, let alone the transformation of this mechanism into a computable cognitive model applied to robot vision. People in the cognitive scene, often looking for it and other things in the same and different, according to the purpose of classification, the brain's thinking ability constitutes a pattern and the ability to recognize. The so-called pattern, refers to a number of elements or components in a certain relationship to form a certain stimulus structure, it can be said that the pattern is a combination of stimuli. When one is able to identify what a pattern he is aware of, it distinguishes it from other patterns, which is pattern recognition. For example, some people want to divide a large number of images into characters, animals, landscapes, buildings, and other five types of storage, the five types are five categories, that is, five different patterns, the process of classification is called pattern recognition. Patterns are simple and complex, complex patterns are often composed of multiple sub-patterns. Cognitive psychologist Simon said: "People in solving mathematical problems, most of them through pattern recognition to solve, first of all to identify the current problem belongs to which category, and then as the index in the memory storage of the corresponding knowledge, this is the pattern recognition." ” The reason why we care about pattern recognition theory is that it is the source of thought that establishes the mathematical model of image (scenery) comprehension. For example: The traditional pattern recognition theory is divided into five categories: template matching mode, prototype matching mode, feature analysis mode, structure description mode and Fourier model. Nowadays, the main mathematical processing methods in image comprehension are almost originated from five traditional pattern recognition theories, or based on their deformation. In recent more than 20 years, a new pattern recognition theory has been put forward, which is divided into five kinds: visual computing theory, attention feature integration theory, component recognition theory, interaction activation Theory and visual topological theory. Among them, Marr's visual computing theory is the mainstream theory of current computer (robot) vision. Other theories are also the source of innovation for many explorers. However, regardless of the above-mentioned pattern recognition theory, there is more or less one-sidedness, so far has not formed a more persuasive, universally recognized theory of pattern recognition. This is the fundamental problem that restricts the development of the mathematical model of image recognition (computer vision). Here we introduce the various pattern recognition theories separately. (Author: Liu Jianzhong Http://hi.baidu.com/liujianz)