Review of Related Literature
Related literatures refer to the list of reference related to the study being conducted. This will serve as a guide.
According to: Anil K. Jain, Brendan Klare, and Unsang Park of Michigan State University.
Face recognition is the task of recognizing a person using digital face images. A FRS is typically designed to output a measure of similarity between two face images. Automated FRSs typically involve finding key facial landmarks (such as the center of the eyes) for alignment, normalizing the face’s appearance, choosing a suitable feature representation, learning discriminative feature combinations, and developing accurate and scalable matching schemes. Two decades of vigorous research has yielded face-recognition systems that are highly accurate in constrained environments. However, the face-recognition community has recognized four key factors that significantly compromise recognition accuracy: pose, illumination, expression, and aging. Face images in government-issued identification documents (such as driver’s licenses and passports) and mug shots are two scenarios that offer such constraints, which has led to success in the de-duplication (that is, matching process to detect ID cards enrolled under different names but belonging to the same subject) of identification cards and prevention of false prisoner releases.
Paradigm for Forensic Face Recognition
In forensic identification, investigators must use any available information to facilitate subject identification. Typically, the sources of face images are surveillance cameras, mobile device cameras, forensic sketches, and images from social media sites. These face images are difficult to match because they are often captured under non-ideal conditions. Non-forensic, fully automated scenarios are not severely impacted by these performance degrading factors. As a result, forensic face recognition often requires a preprocessing stage of image enhancement or a specialized matcher to perform recognition. Another important aspect in face recognition in forensics is the continuously increasing size of face databases or galleries. For example, the mug shot database at the Pinellas County Sheriff’s Office in Florida contains more than 7.5 million face images. Most Departments of Motor Vehicles (DMV) in the US (34 states) utilize FRSs. The US Department of State hosts one of the largest face databases in the world, with a gallery of approximately 100 million images, which are being used for de-duplication of passport and visa applicants. We can state the problem of forensic face recognition as follows. A low-quality query (or probe) image of an unidentified subject is available from a source such as a surveillance camera or a forensic sketch. An expansive database (or gallery) of high-quality face images (such as mug shots) exists that might contain the subject. To boost the recognition accuracy in this difficult matching scenario, a modified matching paradigm with a human in the loop is necessary Although mainstream face-recognition research does not often consider this semi-automated face recognition, it is necessary to include a human in the recognition loop to boost the accuracy and confidence in forensic scenarios. The role of man and machine can vary in this scenario, with two expectations: the machine is used to return a similarity score from some probe image for each image in the gallery, and the human examines the top-K matches (as opposed to only returning the closest match). The article discusses additional scenarios in which a man or machine can be used to improve the prospects of a successful face identification. We separate such methods into two main categories. The first approach uses preprocessing methods to improve the quality of a face image prior to submission to a COTS FRS. These methods do not require any changes to Face detection Image normalization Researchers have developed...