List

The problem of open-set face recognition determines whether the subject’s picture presented to the recognition system belongs to a known individual (person in the gallery). This problem is of great importance to find suspects in public environment. For instance, an airport or a train station have a list of wanted subjects (watch-list) that should be recognized when they walk in the monitored space. However, at the same time the face recognition system cannot mistakenly recognize an unwanted person as a wanted one (false alarms). This is the focus of our paper entitled Towards Open-Set Face Recognition using Hashing Functions that was presented in The International Joint Conference on Biometrics (IJCB 2017) and received the Best Paper Runner-up Award.

Open-set face recognition has a large room for improvement since only few researchers have focused on that subject. In fact, a real-world recognition system has to cope with several unseen individuals at real time and determine whether a given face image is associated with a subject registered in a gallery of known individuals. In this work, we combine hashing functions and classification methods to find out when probe samples are known.

Hypothesis vote list histograms proceed differently whether we present probe face images whose identity are enrolled in the gallery or whether we examine “unseen” individuals.

The next figure illustrates two different queries: one corresponds to querying an enrolled individual and the other searches for an unknown subject. Notice that there is a highlighted bin when the query image has a matching identity. It probably represents the probe sample’s identity. On the contrary, several bins are incremented when no subject from the gallery correlates to the probe sample.

Approach subject samples are partitioned into positive and negative sets and different classification models are learned containing different subjects in each subset, generating hash functions. Then, the probe sample is compared to all hashing functions and their response values are used to increment a vote list. If the ratio of the top scorer to the remaining subjects satisfies a threshold, it is considered a known individual.

Reference:

Rafael Vareto, Samira Silva, Filipe Costa, William Robson Schwartz. Towards Open-Set Face Recognition using Hashing Functions. International Joint Conference on Biometrics, 2017.

  Posts

August 26th, 2017

Face Verification based on Relational Disparity Features and Partial Least Squares Models

Face recognition has been one of the most important tasks in computer vision due to the wide range of applications. One of the face recognition tasks, the face verification is responsible for  determining whether two facial images belong to the same subject or are from different subjects. In this paper, that will be presented in the Conference on Graphics, Patterns and Images (SIBGRAPI 2017), we exploit the differences between pairs of images and the diversity generated by multiple classifiers to perform face verification.

July 20th, 2017

Towards Open-Set Face Recognition using Hashing Functions (Best Paper Runner-up Award IJCB 2017)

The problem of open-set face recognition determines whether the subject’s picture presented to the recognition system belongs to a known individual (person in the gallery). This problem is of great importance to find suspects in public environment. In this paper, that will be presented in The International Joint Conference on Biometrics (IJCB 2017), we combine hashing functions and classification methods to find out when probe samples are known.

June 10th, 2017

R&D Project – Video Analytics Solutions

The Smart Surveillance Interest Group started early this month the execution of a R&D project as a partnership with Maxtrack, a Minas Gerais based company leader in vehicle tracking and telemetry, through EMBRAPII DCC/UFMG.

October 10th, 2016

2nd DeepEyes Workshop

The 2nd DeepEyes Workshop was held at UFMG Campus in October 03, 2016. Counting researchers such as Larry S. Davis (University of Maryland), François Bremond (INRIA), Eduardo Valle (Unicamp), Sandra Ávila, Filipe Costa, William Robson Schwartz (UFMG), were presented and discussed advances achieved in the previous year of the project.