Method and device for extracting image features based on semi-supervised learning

Abstract

The embodiment of the invention discloses a method and device for extracting image features based on semi-supervised learning. The method comprises steps of: initializing a model parameter and preprocessing image data to obtain image samples, and dividing the image samples into training samples including tagged samples and untagged samples, and test sample just including the untagged samples; according to a pairing constraint condition, determining a constraint set corresponding to the tagged samples in the training samples; by using a neighbor search algorithm, constructing neighbor graphs corresponding to all the training samples and calculating a weight matrix; by minimizing a feature approximate error, subjecting the training samples to low-dimensional manifold feature processing to obtain low-dimensional manifold features and a linear projection matrix; and extracting the image features of the training samples and the test samples by using the linear projection matrix. The methodand device can simultaneously maintain the global and local structure information of the sample data, improve the distinguishability of the features, quickly map new test data to low dimension and improve the performance of image feature extraction.

Description

technical field [0001] The invention relates to the technical field of computer vision and image recognition, in particular to a method and device for extracting image features based on semi-supervised learning. Background technique [0002] In a large number of practical applications, real data can be described by high-dimensional attributes or features. However, the dimension of the original feature may be very large, or the sample is in a very high-dimensional space, and the high-dimensional data can be transformed into a low-dimensional space through the method of feature mapping or feature transformation. [0003] Extracting the most effective features for classification from high-dimensional features has always been one of the very important and difficult research topics in the research fields of computer vision and image recognition. The isometric mapping algorithm (Isomap) is a classic nonlinear manifold learning method, and its main idea is to find the subspace tha...

AI Technical Summary

Problems solved by technology

However, MMD-Isomap is a transductive method that cannot quickly solve the mapping problem of external test data. In addition, in practical applications, labeled sample data is scarce, and the process of manually calibrating data will consume a lot of manpower and time.

Images