Method of extracting classification information from high dimensional asymmetric data

Abstract

The present invention relates to the signal and image processing field, and provides a method of extracting classification information from high dimensional asymmetric data for solving the problems that a conventional relevant classification information extraction method is not suitable for the sample asymmetric data or is high in calculation complexity, and the calculated amount is easy to overflow when the high dimensional data is processed. The method comprises the steps of obtaining the high dimensional asymmetric data; giving new weights to sigma o and sigma c, forming a new covariance matrix sigma a to substitute sigma t to carry out characteristic decomposition, and solving the characteristic values and the characteristic vectors; combining to obtain a dimensionality reduction matrix, and projecting the high dimensional asymmetric data via the dimensionality reduction matrix to obtain the classification information after dimensionality reduction. The technical scheme provided by the present invention is low in calculation complexity, high in accuracy, fast in operation speed and good in stability.

Description

technical field [0001] The invention relates to the field of signal and image processing, in particular to a method for extracting classification information from high-dimensional asymmetric data. Background technique [0002] The method of extracting classification information from two types of sample data has very important practical application value. For example, the extracted classification information is used to distinguish face and non-face images, to distinguish disease samples from non-disease samples, and to distinguish useful information from useless information, etc. With the increasingly advanced technology and means of obtaining information, the dimensions of the two types of data that need to be classified are becoming larger and larger. In addition, the number of samples obtained from the two types is usually unbalanced, which makes the traditional classification method of the two types of samples subject to greater limitations. Therefore, there is an urgent...

AI Technical Summary

Problems solved by technology

In many practical applications, the initial dimension of the data is relatively large. For example, an image with a size of 200×200 has 40,000 pixels, that is, 40,000 dimensions. Therefore, APCA calculates the eigenvalues ​​of the high-dimensional data covariance matrix Sometimes it is easy to cause computer memory overflow and subsequent calculations cannot be continued. Even if it can be calculated, such a huge matrix dimension will inevitably bring extremely high computational complexity, and the calculation time and error will increase significantly.

[0004] It can be seen that the existing related classification information extraction methods are either not suitable for data with asymmetric samples, or the calculation is complex and high, and it is prone to calculation overflow when dealing with high-dimensional data.

Images