The basic concept of pattern recognition §1.1
First, the broad definition
1, the pattern: a description of objective things, a perfect example to emulate available.
2, pattern recognition: by definition philosophy is a process of "external information reaches the sensory organs and converted into meaningful sensory experience" of.
Example: water identification, writing, etc.
Second, the narrow definition
1, mode: the description of the object of interest in some quantitative or structure. Class is a collection pattern having certain common features mode.
2, Pattern Recognition: Study on an automatic technique, relying on this technique, the computer will automatically (or interfere as little as possible human) to be respectively assigned to the respective pattern recognition model to class. Note: narrow concept "mode" - is a description of the object, whether the object to be recognized, or known objects. Generalized "model" concept - means "a perfect example to follow
Third, related computer technology
1, based on the current computer system over Neumann.
1946: Hungarian-American mathematician von Neumann computer out of the basic assumptions about the composition and working methods: digital computer binary number system; computer program order execution, that is, the concept of "program memory" of.
1949: developed the first von Neumann computer.
1956: The first AI (artificial intelligence) seminar held in the United States.
2, the fifth-generation artificial intelligence computer
The essential difference: The main function is to rise from the information processing knowledge processing (learning, association, reasoning, interpretation), the computer has certain human intelligence. Development work since the 1980s, has yet to form a unanimous conclusion.
Several possible directions: neural network computer - simulated human brain. Biological computer - use of biological engineering technology, protein molecules for the chip. Optical computer - light as an information carrier, to complete the processing of information by the processing light.
The purpose Fourth, research and development pattern recognition
Raise awareness of the computer, thus greatly to develop computer applications.
§1.2 a pattern recognition system,
Simple example: the establishment of perceptions
To identify cancer cells, for example, to understand the whole process of machine recognition.
1, the information input and data acquisition
The cells convert the image into digital microscopic image of cells, reflecting the size of the optical density value of the pixel, also known as gray-scale digital image. Original digital cell image analysis computer data base.
2, the digitized image of cells pretreated with zoning
Pretreatment of purpose:
(1) removing obtain noise and interference introduced when the data.
(2) removing all secondary inclusions on the background image, the projection image of the main cell to be recognized.
Example: smoothing digital image processing and image enhancement techniques.
Zoning purposes: to identify boundaries, divided into three regions, in preparation for the feature extraction.
3, the feature extraction cells, selection and extraction
Objective: To establish a mathematical model various features for classification. ① extraction features: original data collection, first hand, features large volumes of data. Feature selection and extraction basis.
Example: extracting features of a cell 33, the establishment of a 33-dimensional space X, every cell by a 33 dimensional random vector representation, i.e., to a physical entity "cell" into a mathematical model "33-dimensional random vector" , i.e. 1:33 dimensional space.
② Feature Selection: Select some of the main features of the original features as a feature based on the use of discrimination. Feature extraction ③: using a transformation technique, less comprehensive than the original characteristics as the classification number of stars, called the characteristic dimension of the compression, also known as feature extraction is customary.
§1.3 pattern recognition profiles
I. Introduction Development Pattern Recognition
The 1950's, the rise of the sixties and the rapid development of the early seventies laid the theoretical foundation.
Second, pattern recognition and classification
1, theoretically classification ① statistical pattern recognition mode set to class probability distributions in the feature space density function is based on the general characteristics of the study. Including the decision function method and cluster analysis.
② structure pattern recognition (syntactic pattern recognition)
The complex pattern of differentiation of the sub patterns even simpler primitives, the relationship between the various levels is described by the "structure method", corresponding to the language grammar. Small and simple primitives to describe the grammar rules large and complex patterns.
③ fuzzy pattern recognition
With membership based on the use of "relationship" concept and operation of fuzzy mathematics classification. Membership reflects a certain degree of elements belonging to a set.
④ Intelligent Pattern Recognition
Artificial intelligence and pattern recognition product of the combination, has two branches: a) and artificial neural networks: near physiologic simulation, to achieve the simulation of the image thinking. b) knowledge-based logical reasoning: from the analog to human logical thinking, is the category of abstract thinking.
2, the implementation of points: ① supervised (Supervised) identification: classified and identified using the discriminant function. You need to have enough prior knowledge. ② unsupervised (unsupervised) Recognition: for the absence of prior knowledge, the use of cluster analysis.
§1.4 feature selection and feature extraction
I. Overview: two data measurement situation
① Because of considerations may be implemented or economic restrictions on the measurement, the obtained measurement values few. The nature of the measurement ② can get a lot of value. If all directly as classification characteristic, the machine consuming, and not very well classified. Some call it "feature dimension disaster."
The purpose of feature extraction and selection: After selecting transformed or composition identified features retain as classification information, the classification accuracy guaranteed under the premise of reducing the characteristic dimension of the classifier i.e. fast and accurate operation.
§1.5 pattern recognition applications
First, the car toll system on the highway: by car charging methods
1, the vehicle outline extraction process and analyze the geometrical parameters, to classify. The video detection method, an infrared detection method.
2, other physical parameters (noise, vibration, pressure, etc. weight) measured to achieve a vehicle classification. The WIM, electromagnetic induction.
3, to directly identify the vehicle identification methods to classify. The electronic tag video license plate recognition.
Second, biometrics
Identification techniques according to characteristics unique to each person can be sampled and measured biological characteristics (physical characteristics) and behavior.
1 fingerprint recognition: the earliest and most mature recognition technology.
Wherein the magnitude of a few points of the pattern line studies (gradation value), the ratio of line length and the line corresponding to the angles and the like: 2 palmprint identification.
3 Face Recognition:
4 retina recognition:
5 based on the kinetic characteristics identified: