Method and apparatus for labeling face images

Abstract

The invention provides a method and an apparatus for labeling face images. The method comprises: acquiring a face image set which corresponds to a specific personal name and comprises a plurality of face images, wherein each face image only comprises one face; establishing a whole object assumption that both random face images in the face image set belong to the same person, and verifying whether the whole object assumption is tenable by a predetermined face verification model; and if verifying the whole object assumption is tenable, labeling the plurality of face images in the face image set as a result that the face images in the face image set belong to the specific personal name. According to the scheme, a great number of images crawled from the Internet can be rapidly discriminated and screened, error images are efficiently removed, labor and time cost of labeling the face images is greatly saved, and a powerful guarantee is provided for a subsequent face identification model based on automatic labeling, which has high face image training accuracy; and meanwhile, according to the scheme, an existing data monopoly in the industry can be broken through, which is convenient for popularization of a face identification technology, thereby promoting the progress of the technology.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular, the present invention relates to a method and a device for marking human face pictures. Background technique [0002] Face recognition technology is a biometric technology for identification based on human facial feature information. The face recognition process mainly includes four steps, namely: face image acquisition and detection, face image preprocessing, face image feature extraction, and matching and recognition; among them, the steps of matching and recognition are usually based on pre-trained human face recognition model. In the existing face recognition model training technology, human-labeled face pictures are generally used to train the face recognition model. One way is to crawl a large number of pictures from the Internet based on the names of people, and then use manual methods to perform manual processing on the massive pictures corresponding to each name...

AI Technical Summary

Problems solved by technology

[0003] Since the existing technologies are all done manually, which requires a lot of manpower and material resources and a long time cost, each organization protects the face picture data obtained at a relatively large cost as a commercial secret, thus to a certain extent As a result of data monopoly, other research institutions that have demand for face image data either need to repeat similar work, resulting in a great waste of resources, or are unable to carry out corresponding research, which may hinder technological progress

At the same time, due to manual long-term review of massive pictures, it will cause fatigue and face blindness, which will lead to a decrease in the accuracy of manual labeling pictures in the later stage.

Images