Non-negative adaptive feature extraction-based human face identification method, device and equipment

Abstract

The embodiment of the invention discloses a non-negative adaptive feature extraction-based human face identification method, a non-negative adaptive feature extraction-based human face identificationdevice, non-negative adaptive feature extraction-based human face identification equipment and a computer storage medium. The method comprises the following steps: integrating non-negative matrix decomposition, feature extraction and adaptive neighborhood learning into a uniform frame, acquiring local reconstruction data by a non-negative decomposition technology, minimizing reconstruction error in a non-negative reconstruction space and a feature embedding space at the same time, performing weight adaptive construction and label propagation learning on the reconstruction data, and performingminimizing learning by utilizing a projection-based feature approximate error item; alternately optimizing and learning a human face identification model to acquire an adaptive weight coefficient matrix for maintaining neighbor information, a projection matrix for extracting features and a neighbor maintaining non-negative decomposition matrix; and extracting the identification feature of a humanface test sample set by utilizing the projection matrix and realizing human face identification by utilizing the human face identification model and according to the identification features. Accordingto the technical scheme provided by the invention, the accurate rate of human face identification is improved.

Description

technical field [0001] Embodiments of the present invention relate to the technical fields of computer vision and image recognition, and in particular to a face recognition method, device, device and computer storage medium based on non-negative adaptive feature extraction. Background technique [0002] With the continuous development of Internet and communication technologies, the explosive growth of high-dimensional data has brought great challenges to effective data representation. The high-dimensional attributes, noise, and unfavorable features contained in most real data may directly reduce the representation and classification capabilities of later data. In order to solve this problem, a dimensionality reduction method based on feature extraction emerges at the historic moment, which converts the original high-dimensional real data into representative low-dimensional compact features. [0003] Data representation extraction techniques can be divided into three categor...

AI Technical Summary

Problems solved by technology

Secondly, most of the current feature learning methods use independent steps to obtain the weight coefficient matrix, which cannot ensure whether the pre-calculated weight coefficient is the best for the later data representation.

Finally, existing learning models usually perform feature learning in the original high-dimensional input space, and high-dimensional data usually contains redundant information, unfavorable features and noise, etc., which will directly reduce the learning performance

Images