Posture-spanning colored image facial expression recognition of direct push type migration group sparse discriminant analysis

Abstract

The invention discloses a posture-spanning colored image facial expression recognition method of a direct push type migration group sparse discriminant analysis (TTGSLDA). Training and testing human face images are photographed under two different face visual angles; one group of auxiliary unmarked human face images are selected from target human face postures and are integrated into a marked training image set in source human face postures; and a label of an auxiliary image is a parameter to be optimized by a direct push type linear discriminant analysis (TTLDA) method. After an obtained class label of an auxiliary image set is learnt, a support vector machine (SVM) is trained based on the parameter so that the classification of the testing human face images is finished. In order to sufficiently utilize face information of a colored image to improve the accuracy of the facial expression recognition, characteristics of the human face images are represented by colored scale invariant feature transform (SIFT). The posture-spanning colored image facial expression recognition method has a facial expression recognition effect with relatively high recognition rate and relatively good robustness.

Description

technical field [0001] The invention belongs to the technical field of image processing and pattern recognition, and in particular relates to a cross-posture color image expression recognition method of transductive migration group sparse discriminant analysis. Background technique [0002] Cross-pose facial expression recognition (FER) has become a research hotspot in the field of facial expression recognition research. Compared with the FER problem of frontal or near-frontal perspective, due to the significant difference in the distribution of expression features between training face images and test images, cross-pose expression recognition faces many difficulties. For example, for a face image of a frontal or near-frontal perspective, the right half and the left half of the face are symmetrical or nearly symmetrical. However, for non-frontal face images, there is a large difference between the two parts. [0003] When dealing with the cross-pose FER problem, most metho...

AI Technical Summary

Problems solved by technology

However, in many real scenes, it is difficult to obtain images with the same class information as the test face image pose

Therefore, the original non-positive FER method cannot achieve better recognition results in this case.

[0004] Another major problem of FER is the extraction of facial features

Images