| Popis: |
Face detection is one of the classical computational challenges in computer vision. Detecting faces in an image has many applications in the field of security surveillance, marketing, biometric authentication, social media, photography and many more. Scanning window is most common technique used in face detection. In this method, an image is divided into multiple regions and each region is processed to detect presence of human face. To process an image of size 640 × 480 pixels, 178000 image regions must be processed and at-least 24 frames must be processed within a second, the number of regions to be processed grows significantly for images with higher dimensions and frame rates. With the advent of high resolution video capturing devices capable of recording at higher frame rates, real-time face detection is a challenge as it is computationally intensive process. In this paper, we present a framework which uses scanning-window technique, integral image and Local Binary Pattern (LBP) implemented on GPGPU to process images in short duration with high accuracy. In our experiments, we have achieved processing speed up to 287 frames per second on average with image dimensions of 640 × 480 pixels. |
