Automatic Wrinkles Detection Algorithm toAssess The uniqueness of Facial Wrinkles

Abstract

Facial recognition considered as an important applicationsdue to increasing needs to it in several fields in last decades. But facial recognition applications may effected by several conditions such as poor resolution, occlusion, blur, illumination and pose variation. This is why latest researches in face recognition have focused on new areas of using facial features rather than the typical features like: eyes, nose, mouth, chin, and ears. The new features are called facial micro-features or facial soft biometric such as scars, freckles, moles, and wrinkles. Facial wrinkles considered as natural features appear as people get older. Existing wrinkles detection algorithm focusing on forehead horizontal lines’ detection, while, it is better to detect wrinkles (vertical and horizontal) for entire face rather than just forehead wrinkles. The primary aim of this research is to develop a new wrinkle detection method for all facial regions and construct new method to investigate If the uniqueness of facial wrinkles. Therefore, the performance of wrinkle detection algorithms on entire face has been evaluated, and proposes to use an enhancement technique to improve the performance. The methods have been evaluatedusing selected images from Face Recognition Technology (FERET) and selected images fromnew Sudanese dataset. In the experiment phase, the selected images were manually annotated by researcher to construct ground truth. The experiments showed that; the proposed methodperformed better in detecting facial wrinkles when compared to the state-of-the-art methods. When evaluated on FERET dataset, the average Jaccardsimilarity index (JSI) are 56.17% for proposed method , 31.69% and 15.87% for Hybrid Hessian Filter and Gabor Filter, respectively. Moreover, a new method to investigate the uniqueness of facial wrinkles is proposed using Modified Hausdorff Distance (MHD). We presented experiments on selected images from FERET and Sudanese dataset using manually and automatically detected facial wrinkles. The experiments showed that the wrinkles for same subject achieve the lowest Mean Absolute Error (MAE) of 0.4, while the different subjects achieve Mean Absolute Error (MAE) of 4.0.