Face key point detection method based on attention mechanism

Abstract

The invention discloses a face key point detection method based on an attention mechanism. The method comprises the following steps: establishing a multi-channel feature extraction network structure oriented to multi-resolution input; designing an attention module, introducing an attention mechanism by utilizing a feature fusion network with an hourglass structure, and selecting a key feature related to a task; and enabling the feature map output by the attention module to pass through a residual module, and outputting key point position information. According to the multi-channel feature extraction network oriented to multi-resolution input, relay supervision is introduced, so that the detection precision is ensured while the network depth is reduced; according to the attention module, the features are continuously focused to the region of interest by introducing an attention mechanism, and the face key point detection precision under shielding and large postures is improved. Experiments show that the method can effectively overcome the influence of large attitude and shielding on the precision, and solves the problem of network depth and detection speed equalization.

Description

technical field [0001] The invention relates to a human face key point detection method based on an attention mechanism, and belongs to the technical field of image processing. Background technique [0002] Face key point detection, also known as face key point positioning or face alignment, refers to marking the eyebrows, eyes, nose, mouth, and contour areas of a face based on a given face image. In scientific research and practical applications have received widespread attention. For example, face posture correction, posture recognition, expression recognition, fatigue monitoring, mouth shape recognition, face beauty, etc. There are many publicly available face datasets, such as WFLW (98 points), Helen (194 points), 300W (68 points), IBUG (68 points), LFPW (29 points), AFLW (21 points), etc. , which can be directly used in the research of face key point detection algorithm. [0003] Considering that the convolutional neural network has a strong feature expression abilit...

AI Technical Summary

Problems solved by technology

The above algorithm can improve the detection accuracy, but it will increase the complexity of the network accordingly and affect the speed of the detection algorithm

In addition, existing algorithms do not perform well in occlusion, large-pose face keypoint detection

[0004] Therefore, there are still the following problems in the field of face feature point detection on these data sets: 1) cannot effectively solve the detection of face key points under large poses, occlusions, and low resolution; 2) the detection accuracy of face key points Can't take care of speed

Images