Machine learning to determine facial measurements via captured images
Techniques for automated facial measurement are provided. A set of coordinates for a set of landmarks on a face of a user are extracted by processing an image using a machine learning model. An orientation of the face of the user is determined. It is determined that impedance conditions are not present in the images, and a reference distance on the face of the user is estimated based on the image, where the image depicts the user facing towards the imaging sensor. A nose depth of the user is estimated based on a second image based at least in part on the reference distance, where the second image depicts the user facing at an angle relative to the imaging sensor. A facial mask is selected for the user based on the nose depth.
1. A method, comprising:
identifying a set of images, from a plurality of images captured by an imaging sensor, that satisfy defined orientation criteria, comprising, for each respective image in the set of images:
extracting, by processing the respective image using one or more landmark-detection machine learning models, a respective set of coordinate locations for a set of facial landmarks on a face of a user depicted in the respective image; and
determining a respective orientation of the face of the user depicted in the respective image based on the respective set of coordinate locations for the set of facial landmarks, wherein at least one image of the plurality of images is discarded for failing to satisfy the defined orientation criteria;
determining that one or more impedance conditions are not present in the set of images;
estimating a reference distance on the face of the user based on a first image of the set of images, wherein the first image depicts the user facing towards the imaging sensor and the reference distance indicates an absolute measurement of at least one of the set of facial landmarks;
estimating a nose depth of the user based on a second image of the set of images based at least in part on the reference distance, wherein the second image depicts the user facing at a predefined angle relative to the imaging sensor; and
selecting a facial mask for the user based at least in part on the nose depth.
2. The method of claim 1 , wherein determining that the one or more impedance conditions are not present comprises:
determining, by processing each image of the set of images using a glasses-detection machine learning model, that the user is not wearing glasses in the set of images; and
prior to identifying the set of images, determining, by processing at least one of the plurality of images using the glasses-detection machine learning model, that the user is not wearing glasses.
3. The method of claim 1 , wherein determining that the one or more impedance conditions are not present comprises:
determining, based on the respective set of coordinate locations for each respective image of the set of images, that a mouth of the user is closed in the set of images.
4. The method of claim 3 , wherein determining that the mouth of the user is closed in the set of images comprises, for the first image:
determining a mouth width of the user in the first image based on the respective set of coordinate locations;
determining a mouth height of the user in the first image based on the respective set of coordinate locations; and
determining that a ratio of mouth height to mouth width is above a defined threshold.
5. The method of claim 1 , wherein determining that the one or more impedance conditions are not present comprises:
determining, based on the respective set of coordinate locations for the first image, that at least one eye of the user is open in the first image.
6. The method of claim 5 , wherein determining that at least one eye of the user is open in the first image comprises:
determining an eye width of the user in the first image based on the respective set of coordinate locations;
determining an eye height of the user in the first image based on the respective set of coordinate locations; and
determining that a ratio of eye height to eye width is below a defined threshold.
7. The method of claim 1 , wherein:
estimating the reference distance comprises determining a width of an iris of the user by processing the first image using an iris-detection machine learning model,
estimating the reference distance comprises determining a scaling factor based on the width of the iris, and
the scaling factor indicates a number of pixels in the first image per millimeter on the face of the user.
8. The method of claim 1 , wherein determining the nose depth of the user comprises:
determining a relative scale for the second image based on a face height of the user depicted in the first image; and
determining the nose depth based at least in part on the relative scale.
9. The method of claim 1 , further comprising determining a nose width of the user depicted in the first image based at least in part on the reference distance, wherein selecting the facial mask is performed based further on the nose width, and wherein determining the nose depth comprises:
determining a distance between a tip of a nose of the user and an alar-facial groove of the user based on the second image; and
computing the nose depth based on the predefined angle in the second image, the nose width, and the distance between the tip of the nose and the alar-facial groove.
10. The method of claim 9 , wherein:
the second image depicts the user facing at the predefined angle towards a first side of the imaging sensor, and
the nose depth of the user is further determined based on a third image, from the set of images, wherein the third image depicts the user facing towards an opposite side of the imaging sensor, as compared to the second image.
11. A system, comprising:
a memory comprising computer-executable instructions; and
one or more processors configured to execute the computer-executable instructions and cause the system to perform an operation comprising:
identifying a set of images, from a plurality of images captured by an imaging sensor, that satisfy defined orientation criteria, comprising, for each respective image in the set of images:
extracting, by processing the respective image using one or more landmark-detection machine learning models, a respective set of coordinate locations for a set of facial landmarks on a face of a user depicted in the respective image; and
determining a respective orientation of the face of the user depicted in the respective image based on the respective set of coordinate locations for the set of facial landmarks, wherein at least one image of the plurality of images is discarded for failing to satisfy the defined orientation criteria;
determining that one or more impedance conditions are not present in the set of images;
estimating a reference distance on the face of the user based on a first image of the set of images, wherein the first image depicts the user facing towards the imaging sensor and the reference distance indicates an absolute measurement of at least one of the set of facial landmarks;
estimating a nose depth of the user based on a second image of the set of images based at least in part on the reference distance, wherein the second image depicts the user facing at a predefined angle relative to the imaging sensor; and
selecting a facial mask for the user based at least in part on the nose depth.