Similarity measurement-Clustering

Image segmentation and Feature Extraction

Similarity measurement-Clustering

The classification problem described above is to construct a classifier using samples of known classes. Its training set samples are known categories, so they are also called supervised learning. A single sample to be tested is classified under the guidance of samples of known classes. The clustering problem is different. It does not know the category of each sample in a batch or other prior knowledge in advance, and the unique classification is based on the characteristics of the sample. A classifier is constructed based on the characteristics of samples. A classifier is called unsupervised classifier or cluster. Clustering Analysis is a tool for classification and analysis of the test data. Many disciplines classify the data based on the measured or perceived similarity, and classify the test data into various aggregation classes, the patterns in the same aggregation class are more similar than those in different aggregation classes, so as to estimate the relationship between patterns. Clustering Analysis results can be used to propose initial assumptions for data, classify new data, and compress the same type of test data. Clustering Algorithms focus on finding aggregation classes with similar features. Humans are the two-dimensional optimal classifier. However, most practical problems involve high-dimensional clustering. It is very difficult to intuitively explain the data in a high-dimensional space. In addition, data does not follow the ideal structure of rules, which is why a large number of clustering algorithms appear in the literature. Because clustering analysis is performed in an image, an image contains multiple objects, which must be separated and identified by different objects. To classify different objects, you must master the following content. 1. Basic concepts of clustering 2. Divide the image, identify each object, and identify the object. 3. Measure each object, such as the area and perimeter, and extract the features of each object. 4. Based on these features, use the clustering algorithm for similarity analysis and then classify them. 1. Definition of clustering evertt proposes that an aggregation class is a set of similar entities, and the entities of different aggregation classes are not similar. The distance between two points in an aggregation class is smaller than the distance between any point in the class and another point not in the class. An aggregation class can be described as a continuous area with a higher density point and a region with a lower density point in the D dimension space, while a region with a lower density point separates the areas with other higher density points. In mode space S, if n samples X1, x2... xn: Find the corresponding region R1, R2... rk, which belongs to any Xi class and does not belong to the two types of selection clustering at the same time, should be based on an ideal clustering concept. However, if the data does not meet the assumptions made by the clustering technology, the algorithm does not discover the real structure, but adds a structure to the data. The clustering criterion has n samples of unknown classes,