A method for constructing low-level features of human face

Abstract

The invention discloses a method for constructing underlying features of a human face, which belongs to the technical field of pattern recognition and feature extraction. It is characterized in that: comprising the following steps, 1.1, reading the reference face image in the face database and the multi-pose face image with posture change; 1.2, positioning the reference face and the position coordinates of the eyes of the posture face, and determining the multi-pose The left and right sway angle of the face; 1.3. According to the obtained left and right sway angle, calculate the weight value of the swayed face; 1.4. Calculate the new underlying feature image of the face according to the calculated weight value. The present invention provides a new method for constructing the underlying features of human faces, which can more accurately reflect the different influences of different faces. The new underlying features better integrate the comprehensive information of multiple facial images, saving storage space, It reduces the computational complexity, weakens the noise interference in a single image, and provides a new feature with simple calculation and remarkable effect for the later deep feature extraction and classification recognition.

Description

technical field [0001] The invention discloses a new method for constructing the underlying feature of human face, which belongs to the technical field of pattern recognition and feature extraction. Background technique [0002] Extracting good features has always been the essence of pattern recognition. According to the different ways of feature generation, image features are divided into two categories: natural features and artificial features. Artificial features are features constructed through calculations, such as histograms, spectrograms, chain codes, etc.; natural features are inherent features of the image itself, which can be directly obtained through the human visual perception system, such as image edges, textures, shapes, points, line etc. In face recognition, by measuring the face, a description of the face can be obtained, that is, a point in the measurement space is used to represent a face. [0003] At present, the research on algebraic feature extraction...

AI Technical Summary

Problems solved by technology

However, the construction of this image is simply the average superposition of images, and the energy map obtained by construction cannot reflect the influence of faces of different qualities or different degrees of importance.

Images