Structured face feature extraction algorithm based on elastic embedding
Junhong Tong · 2020
Traditional linear feature extraction methods, such as PCA and LDA, are based on global structure, but human faces may exist approximately on a nonlinear low-dimensional submanifold in high-dimensional data space. Aiming at the problems of high spatial complexity and poor real-time performance of traditional elastic graph matching algorithm for face recognition, an improved elastic graph matching algorithm is proposed. Firstly, Gabor is used to detect and locate face features (such as eyes, nose, mouth, etc.), and a set of Gabor wavelet coefficient vectors are extracted; Then, under the constraint of cost function, elastic graph matching is performed for each feature. The improved method proposed in this paper overcomes the problems of inaccurate face subregion, missing feature information in the process of face roughness extraction and poor real-time performance of the algorithm. The improved algorithm obviously improves the efficiency and accuracy of gender recognition.