1. What does face recognition tell you?
What can face recognition tell you?
AI extracts facial features in an image as a sequence of numbers, compares those numbers against registered actress data, and presents candidates in order of closeness. Results present the closest candidates and do not definitively establish identity. When the leading candidates have very close scores, they cannot be narrowed down to one person, so also check the title pages for each candidate.
Site search covers Japanese-language content; enter queries in Japanese.
2. What is face recognition (AI recognition)?
Here, face recognition means using the face in an image as a clue to compare against pre-registered actress data and present candidates. It is not a fortune-telling-style assessment of who someone looks like, but a mechanical calculation of how closely the face matches existing actress data.
The results therefore present the candidates with the closest features, rather than definitively establishing the person's identity. This characteristic is directly relevant to how you should read the similarity scores described below.
3. How the recognition process works
Uploaded images are processed in roughly the following order.
- Locate faces in the image (recognition cannot be performed on an image in which no face can be detected).
- Use AI to extract the detected face's features as a sequence of numbers (representing the spatial relationships between the eyes, nose, facial outline, and other features).
- Compare those numbers against the registered actress data and calculate their closeness as a similarity score.
- Display candidates in descending order of similarity.
Because the comparison uses sequences of numbers, the images themselves do not need to match. Images used for recognition are discarded after the results are returned and are not stored (see the Privacy Policy for details).
4. How to read similarity scores
Instead of narrowing the results down to a single person, the system lists multiple candidates in descending order of similarity. Keep the following points in mind when interpreting them.
- If the top-ranked candidate has a notably high similarity score with a large gap over the second and lower-ranked candidates, she is a strong candidate.
- If the leading candidates have very close scores, it is likely that several actresses with similar facial features are registered, so you cannot definitively identify a single person.
- Similarity indicates relative closeness within the registered data; actresses who are not registered will not appear as candidates.
When scores are close, a reliable way to narrow down the candidates is to check their titles from each candidate's actress page, considering the characteristics of the titles and their studios as well.
5. When it is useful for amateur and themed titles
In amateur and themed titles where performers' names are not disclosed, the title name and cover information do not let you trace the performers. Face recognition compares only facial features without using any name information, making it particularly useful for titles with no clues to a performer's name.
On the other hand, if a product code is printed on the cover, looking it up by product code may be more reliable and faster. Choose a method according to the clues available.
6. Possible reasons why no recognition results appear
- No face could be detected because it is too small, turned sideways, obscured by mosaic censoring, or similarly difficult to see.
- The image quality is too low to retain enough information to extract features (for example, a low-resolution screenshot).
- The relevant actress is not yet included in the registered data.
The Actress Recognition AI User Guide explains how to choose images and specific ways to address problems. For an overall comparison of methods for searching from images, see AV image search methods.
7. Frequently asked questions
What does AV actress face recognition do?
AI extracts the facial features in an image as a sequence of numbers, compares those numbers against registered actress data, and presents candidates in order of closeness. It is not a fortune-telling-style assessment of who someone looks like, but a mechanical calculation of how closely the face matches existing data.
Does a high similarity score definitively establish the person's identity?
No. If the top candidate's similarity score stands out with a large gap over the second and lower-ranked candidates, she is a strong candidate. However, if the leading candidates have very close scores, it is likely that several actresses with similar facial features are registered, so you cannot narrow it down to one person.
Can the system recognize actresses who are not registered?
No. Similarity is relative closeness within the registered data, so actresses who are not yet registered will not be displayed as candidates.
Can face recognition be used for amateur and themed titles?
It is particularly useful for amateur and themed titles where performers' names are not disclosed. Face recognition uses no name information and compares only facial features, so it can present candidates even for titles with no clues to a performer's name.
When might recognition return no results?
Possible reasons include the face not being detected because it is too small, turned sideways, or obscured by mosaic censoring; image quality being too low to extract features; or the relevant actress not yet being included in the registered data.