Basic concepts of 1-1 Pattern Recognition

The purpose of pattern recognition research is to use computers to Classify physical objects. The recognition results should be consistent with objective objects as much as possible with the minimum error probability. The most basic way for a machine to identify things is to calculate the degree of similarity between things to be analyzed by computers and standard templates. Therefore, the difference between different things must be seen from the measurement to distinguish the current things to be identified.

1. Mode description

In pattern recognition technology, each object observed becomes a sample. For each sample, some identification-related factors must be identified as the basis for the study, and each factor becomes a feature. The pattern is the description of the features of the sample. Pattern feature sets can be used to represent feature vectors in the same feature space. Each element of a feature vector is called a feature vector.

If a sampleXIf n features existXAs an n-dimensional column vector, this vectorXIt is called a feature vector. Pattern recognition is based onXN features to identify the patternXBelongW1, W2 ,... WmClass. Different modes to be recognized are investigated in the same feature space. Different pattern classes vary in their feature value range due to their different properties, it appears in different areas of the feature space.

Therefore, the goal of the pattern recognition system is to find a ing between the feature space and the interpretation space. A feature space is a space consisting of measurements, attributes, or elements that are useful for classification obtained from a pattern.MSet of categories.

2. Pattern Recognition System

A typical pattern recognition system consists of data acquisition, preprocessing, feature extraction, classification decision making, and classifier design. It is generally divided into the upper part and lower part: the upper part completes the classification of the unknown category mode; the lower part belongs to the training process of classifier design. It uses samples for training, determines the specific parameters of the classifier, and completes the classifier design. Classification decision-making takes effect in the recognition process and makes classification decisions for the identified samples.

1) Feature Extraction and selection

Transform the original data to obtain the features that best reflect the nature of classification. Convert a measurement space with a higher dimension (space composed of raw data) to a feature space with a lower dimension (space on which classification is based ).

2) classification decision

In feature space, pattern recognition is used to classify the identified object into a certain category.

3) classifier design

The basic practice is to determine the discriminant function based on the sample training set and improve the discriminant function and error test.

3. Main Problems of Statistical Pattern Recognition

1) feature selection and Optimization

There are two basic methods to optimize the feature space. One is feature selection. If the selected feature space can tighten the distribution of similar objects, it can provide a good foundation for classifier design. Otherwise, if different types of samples are mixed in the feature space, the optimal design method cannot improve the accuracy of the classifier. The other is feature combination optimization. The original feature space is transformed through a ing transformation to construct a new simplified feature space.

2) Classification

The classification and features of several samples are known. For example, the identification of handwritten Arabic numbers is a classification problem of 10 classes. The machine must first know the shape and features of each handwritten number, for the same number, different people have different ways of writing, and machines must be informed of which type they belong. Therefore, a sample library is required for classification issues. Establish a discriminant classification function based on these sample libraries. This process is implemented by machines and becomes a learning process. Then, analyze the features of an unknown new object to determine which type of object it belongs. This is a method of supervised learning.

3) Cluster Identification

Some objects and their features are transplanted, but they do not know which category each object belongs to, and they do not know in advance how many classes they are divided into. They are measured using a similarity measure method, and "things are clustered, and people are grouped by groups ", classify the same features into one type. This is an unsupervised learning method.

Basic concepts of 1-1 Pattern Recognition