Averaging Dimensionality Reduction and Feature Level Fusion for Post-Processed Morphed Face Image Attack Detection

Mary Ogbuka Kenneth, SULAIMON ADEBAYO BASHIR

Abstract:

Facial morphing detection is critical when applying for a new passport and using the passport for identity verification due to the limited ability of face recognition algorithms and people to detect morphed photographs. As a result of face recognition systems' vulnerability to morphing attacks, the value of detecting fake passports at the ABC gate is undeniable. Nonetheless, identifying morphed images after they have been altered using image operations like sharpening, compression, blurring, print-scan and resizing is a significant concern in Morphing Attack Detection (MAD). These image operations can be used to conceal the morphing artefacts, which makes MAD difficult. Several researchers have carried out MAD for print-scan images; few researchers have done MAD for compressed images; however, just one paper has considered image sharpening operation. Hence, this paper proposes a MAD technique to perform MAD even after image sharpening operation using averaging dimensionality reduction and feature level fusion of Histogram of Oriented Gradient (HOG) 8 × 8 and 16 × 16 cell size. The 8 × 8 pixels cell size was used to capture small-scale spatial information from the images, while 16 × 16 pixels cell size was
used to capture large-scale spatial details from the pictures. The proposed technique achieved a better accuracy of 95.71% compared with the previous work, which reached an accuracy of 85% when used for MAD on sharpened image sources. This result showed that the proposed technique is effective for MAD on sharpened post-processed images.