This paper introduces a novel method for detecting real faces in images by distinguishing live captures from screen-based images. The technique utilizes key metrics such as variance of Laplacian (sharpness), brightness spot ratio, and color deviation. A critical innovation is the variance-to-distance ratio, which effectively differentiates real persons from screen captures. This method enhances security in applications requiring facial verification, such as biometric authentication and online banking. Experimental results demonstrate its efficacy under various conditions.
For more details, read the full paper here.