Pattern Recognition (a) schema definition

Related concepts:
mode :
defined patterns :
In order for the machine to perform the task and complete the identification, classification must be carried out to identify the object of scientific abstraction, the establishment of the mathematical model to describe and identify objects in place, this is the description of the object model.
Manifestations mode:
feature vector symbol string diagram, the relation
concept of pattern recognition:
the feature or attribute of the object of study, the use of certain analysis algorithm finds its category, and the classification result as the real conform.
Pattern recognition applied:
with a visually robot, biometrics, remote sensing image interpretation and other machinery, automobile autopilot system;
the process of pattern recognition system :

three core issues of pattern recognition: feature extraction and selection, learning and training, classification and recognition
feature extraction and selection:
feature extracting features of the object type of the nature of the research and measurement results of numerical or symbolic objects and decompose to form a feature vector, or a symbol string graph, the representative object generation mode. Wherein the selective conditions in correct classification rate, according to some criterion try to use a large classification feature for proper action to be accomplished with less features classification task.
(1) acquisition mode
imaging, sound recording, a digital camera, a television, infrared, laser, sonar, radar, remote sensing, A / D conversion.
In the pretreatment mode acquisition and generally use analog to digital (A / D) conversion. A / D conversion must pay attention to two problems:
a sample rate must satisfy the sampling theorem;.
B quantization level, depending on the accuracy requirements;.
(2) pre-treating
a denoising: eliminate or reduce noise in the acquisition mode and the other. interference and improve the signal to noise ratio;
. deblurring B: eliminate or reduce the image blur data (including motion blur) and geometric distortion and improve clarity;
C mode conversion structure: for example non-linear model into linear mode, in order to facilitate subsequent processing.
Methods of pretreatment are: filtering, transforming, encoding, and the like normalization.
(3) feature extraction / selection
purposes: to reduce the number of dimensions, reducing the processing load of the classification error is relatively small. The most beneficial amount selected from a pattern classification model as a feature space, the dimension of the compressed mode, in order to facilitate handling, to reduce consumption.
Feature extraction: the general classification in a certain decision rules used as a criterion. The extracted features in the classification of certain criteria a minimum of errors. This requires consideration statistical relationships between features, choose the appropriate orthogonal transformation in order to extract the most effective feature.
Feature selection: the need for some kind of classification criteria, choose a larger contribution to the classification of features in this criterion, remove the smaller contribution features.
Learning and training:
the machine has a classification function, it is subject training, the recognition of human knowledge and methods as well as knowledge about the classification of objects into the machine, resulting in classification rules and analysis program.
Classification:
the knowledge machine match classification and object to be identified, the more rational use of knowledge, the stronger the system of recognition accuracy rate is higher.
Category: The feature space is divided into class space. The sample of unknown class attribute space class is determined as a certain type. Factors affecting the classification error rate:
A classification.
B classifier design.
C extracted features.
D sample quality.

Pattern recognition technology is mainstream :

Statistical pattern recognition : the direct use of various distribution, or implied use of probability density function, posterior probability concepts identified. The basic technology has cluster analysis, discriminant interface class field of algebraic methods, statistical decision method, nearest neighbor method;
Structure pattern recognition : the object into a plurality of basic units, i.e. cell; structural relationships which may be represented by strings or FIG, i.e. sentences; syntactic analysis performed by the sentence, the grammar is determined according to its category;
fuzzy pattern recognition : the mode or pattern type set as the mode, which was converted to the membership attribute, using the membership functions for fuzzy inference, or fuzzy relation classification;
artificial neural network : a large number of simple basic units, i.e., from neurons connected to each other nonlinear dynamic system, self-learning, self-organizing, fault tolerance capabilities and strong Lenovo, Lenovo can be used to identify and decision-making.
Method AI : how to make the machine and method having a theoretical brain function;
subspace method : generating a subspace corresponding to various types of pattern features from the original spatial correlation matrix in accordance with various types of training samples by a linear transformation, each with each subspace category-one correspondence;
the basic principles of pattern recognition: