Face detection and alignment method and device based on scale estimation, and storage medium

Abstract

The invention discloses a face detection and alignment method based on scale estimation. The face detection and alignment method comprises the steps of: inputting a picture into a scale estimation network, and outputting a scale with a scale probability vector greater than a preset threshold; when the scale estimation network is trained, allocating attention weights to faces in the picture in advance according to the face scales, wherein a loss function of the scale estimation network during training comprises binary classification loss of the face attention images; scaling a to-be-detected image through using the scale obtained by the scale estimation network to obtain a plurality of scale images; inputting the plurality of scale images into an anchor Pnet to obtain a plurality of candidate boxes, and removing non-face candidate boxes through adopting a non-maximum suppression algorithm to obtain a pre-processed candidate box; and cutting the preprocessing candidate box on the original image, zooming the preprocessing candidate box to a preset size, inputting the zoomed preprocessing candidate box into an anchor Rnet, removing a redundant box by using the non-maximum suppression algorithm to obtain a detection box, and extracting corresponding face feature points according to the detection box. The face detection and alignment method has the advantages of being high in adaptability and capable of detecting small-scale faces more efficiently.

Description

technical field [0001] The present invention relates to the technical field of video monitoring and image processing, and in particular to a face detection and alignment method, device and storage medium based on scale estimation. Background technique [0002] With the rapid development of science and technology, computer vision is becoming more and more popular in social life. Face detection and alignment technology is one of the research hotspots. It has many applications in real life, such as facial recognition access control, mobile phone unlocking, security monitoring, identity Inspection, etc., the application of face detection and alignment technology has brought greater convenience to daily life. In an actual scene, an image may contain faces of different scales at the same time, such as small-scale and large-scale faces. In order to detect faces of different scales at the same time, the existing approach is to use uniformly distributed image pyramids. Detect on den...

AI Technical Summary

Problems solved by technology

However, these methods have the disadvantage of high computational complexity

In addition, in order to reduce the number of pyramids, some detection techniques use the scale estimation method, but this method tends to ignore small-scale faces when there are multi-scale faces in the image, resulting in the error of missing face detection, giving people The application of face detection brings a lot of inconvenience

Images