Apple emphasized in the patent application that this faceprint automatic identification system is mainly aimed at public figures or iconic graphics, not users' friends and family. The invention needs the support of digital image management application to analyze the content in a given photo, create facial lines, and then identify people in other photos through facial recognition.
The patent obtained by Apple also describes the function of grouping multiple texture data by reliability score. For example, the device has stored various photos of someone for decades, and now his appearance must have changed a lot from 40 years ago. The system can carry out statistical analysis on the photos with large span in these years and classify them into one group.
It is pointed out in the patent document that facial lines are a subsystem of feature vectors and can identify objects, so facial lines can actually be used to identify things that are not faces, such as buildings.
In order to realize face recognition in selected digital photos, face detection/recognition software generates a set of features or facial lines, which can describe the personalized details of faces in detail.
Then compare the generated texture with other textures to determine whether they match (or are similar enough). If the compared facial prints are found to match each other, the face detection/recognition software determines that the person in the current photo is the person whose facial print information already exists in the data.
In the process of photo analysis, the management application compares the texture with other textures that have been generated and stored on the local or remote server. If the two data are found to match each other, the software will label the person.
The reliability of facial pattern matching mainly depends on face recognition technology, and there will be corresponding scores in the matching process. If the score is lower than the critical value, it is determined that they do not match.
Users can also give feedback in the process of recognition and intervene in this process to improve the recognition efficiency. Click the green button to indicate that it is correct, and click the red "X" button to indicate that there is a problem with recognition.
This patent also takes into account that devices like iPhone can create and store texture data locally, or use offline cache for identification. In addition, faceprint content can also be stored in the cloud and synchronized with devices. In addition, photos can be read in advance, so that the whole recognition process is faster.