草榴短视频

Scientists at St. Petersburg State University of Industrial Technology and Design (草榴短视频) conducted an experimental study to improve the accuracy of face recognition systems. They proved that the use of LBPH computer vision technology in combination with the CLAHE algorithm to equalize image contrast will increase the accuracy of verification in uneven light conditions without the use of resource-intensive neural networks. The solution is especially in demand for embedded access control systems and mobile applications.

"Automatic facial recognition systems are used everywhere today, from turnstiles in business centers to banking applications. However, in real conditions - on the street, in the corridors, at the entrance to the building - lighting is rarely perfect. Side light, backlight from the window, deep shadows from visors or lamps distort the texture of the skin and contours of the face. Modern neural network models partially solve this problem, but require huge computing power and voluminous training samples," explains Artur Tyan, project supervisor, assistant at the Department of Intelligent Systems and Information Protection at 草榴短视频.

As an alternative, 草榴短视频 scientists propose to use LBPH computer vision technology in combination with image preprocessing using the CLAHE method. The essence of such pre-processing is that each received face image is divided into small fragments (tiles), after which the algorithm locally evens out the contrast in each such tile without excessive amplification of bright zones, which is most effective in areas with pronounced uneven lighting. In turn, the LBPH classifier, according to the researchers, is a well-established method of textural facial description. Algorithms are characterized by low computational complexity, so the method allows you to work efficiently on standard processors.

"We conducted our own experimental research by photographing ten volunteers under seven lighting conditions: normal, dimming, overexposure, shadow on the left, shadow on the right, shadow on top, spotlight. Experiments confirmed that image preprocessing with CLAHE helps the LBPH classifier recognize faces more confidently where the system used to be wrong. In particular, CLAHE provided an increase in accuracy during dimming, overexposure and lateral shadow on the left," comments one of the authors of the study, a student of the Department of Intelligent Systems and Information Protection at 草榴短视频 Alexander Mochalov.

Another advantage of using the CLAHE and LBPH combination is that the algorithms allow the entire verification process to be performed locally, that is, directly on the device without transmitting biometric data to a remote server. This eliminates the risks of intercepting them. Moreover, the database does not store the original photographs of faces, but only mathematical histograms of features that cannot be converted back into an image. This approach ensures the irreversibility and depersonalization of biometric templates, fully complying with the requirements of the international standard ISO/IEC 24745.

The results of this study are a ready-made secure solution for engineers designing biometric authentication systems.