Demographic analysis of facial landmarks
Inventors
Assignees
University of North Carolina at Wilmington
Publication Number
US-9177230-B2
Publication Date
2015-11-03
Expiration Date
2031-09-07
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
A facial image may be annotated with the plurality of facial landmarks. These facial landmarks may be points or regions of the face that are indicative, either alone or in combination with other facial landmarks, of at least one demographic characteristic. Demographic characteristics include, for example, age, race, and/or gender. Based on the demographic characteristic being analyzed, one or more of these facial landmarks may be selected and arranged into an input vector. Then, the input vector may be compared to one or more of the training vectors. An outcome of this comparison may involve in the given facial image being classified into a category germane to the analyzed demographic characteristic (e.g., an age range or age, a racial category, and/or a gender).
Core Innovation
The invention presents methods for demographic analysis of facial landmark data in digital images. Facial images are annotated with a plurality of facial landmarks, which are points or regions indicative of demographic characteristics such as age, race, and gender. These selected landmarks are arranged into an input vector, which is then compared against a set of training vectors derived from facial images with known demographic data.
The primary objective addressed is improving the accuracy of demographic estimations—such as age, race, or gender—from digital facial images. Previous methods often lacked sufficient precision in determining these characteristics. The described method solves this problem by using machine learning techniques, especially support vector machines (SVM) for classification and support vector regression (SVR) for estimation, applied to vectors of selected facial landmarks.
Depending on the demographic characteristic analyzed, different subsets of facial landmarks may be employed. The chosen subset is formulated into an input vector, which is classified into one of several predefined demographic categories, such as age range, racial identity, or gender. Further, after initial classification, a second regression-based comparison may be performed to provide a finer estimation, such as determining a specific age within an age range. The invention thus enables accurate, automated classification and estimation of demographic attributes from facial images.
Claims Coverage
The patent contains several independent claims disclosing five main inventive features.
Method for demographic classification using facial landmark vectors
A method comprising: - Obtaining an input vector of facial landmarks from a given facial image, with landmarks selected for their ability to represent anthropometric characteristics. - Performing a weighted comparison between the input vector and a plurality of training vectors, where each training vector represents facial landmarks from different images and is associated with an age range. - Classifying the facial image into one of several categories based on the comparison outcome. - Performing a second comparison of the input vector with a subset of training vectors mapped to the identified category. - Estimating the age of the individual from whom the facial image was derived, based on this second comparison.
Mapping training vectors to multidimensional regions for classification
A method wherein each training vector also maps to a region in an m-dimensional space, assigning one region per category. The comparison of input vectors involves mapping them to these regions, and classification is based on the region(s) to which the input vector maps.
Categorization of facial images by race or gender using landmark vectors
A method in which each category of the plurality of categories comprises a different human racial identity or gender. Some facial landmarks in the vector represent racial or gender characteristics of the facial image, and classification is performed accordingly.
Application-driven selection of facial landmark subsets for demographic vector formation
A method involving determining a relevant application (such as age, gender, or race estimation), then selecting a subset of facial landmarks from the original vector to form a second input vector for comparison to training vectors and classification.
Computer-readable medium with instructions for demographic estimation from facial landmark data
An article of manufacture (non-transitory computer-readable medium) storing software instructions that cause a computing device to: - Identify an input vector of facial landmarks from an image. - Perform a weighted comparison to training vectors, classify the image into a category, and perform a further comparison with a subset of training vectors mapped to that category. - Estimate an age of the individual based on the outcome.
The inventive features collectively establish a method and system for demographic classification and estimation from facial landmarks in images, covering applications for age, race, and gender using machine-based learning and vector comparison techniques.
Stated Advantages
Enables more accurate estimation of demographic characteristics (such as age, race, and gender) from facial images compared to previous methods.
Reduces the number of images to be reviewed in identification and verification scenarios by automating demographic classification.
Achieves a lower mean absolute error in age estimation than other publicly available results, yielding improved precision.
Allows rapid processing of large numbers of digitally-represented facial images using computing devices.
Documented Applications
Supporting public security functions, such as enabling law enforcement agencies to narrow digital photograph searches by estimating age, race, and gender from facial images.
Facilitating age verification in settings like casinos, bars, youth sports organizations, and professional sports organizations to determine qualification or legal status based on age.
Interested in licensing this patent?