Article Preview
TopIntroduction
Facial beard and facial hair play an important role in the individual expressions of the human. Hair and faces are essential parts of human virtual characters because virtual characters (computer-generated characters) are providing naturalism or realism, video film, and gaming industry. Hair is an essential piece to build digital human characters and fulfil the respective requirements. Currently, face-capture technology does not easily accommodate facial-hair features. Facial hair is an important step in the 3D reconstruction of real facial geometry and cyber-physical systems.
Currently, the advancement in field of machine vision and computer graphics plays an important role in the cyber-physical systems like autonomous mobile systems and robotics systems. 3D face reconstruction is an important task for solving security problem in the departmental stores, supermarket, crowded and tourist places. 3D face reconstruction from single photo addresses various challenges like profile, wild, occluded, and facial beard poses. Traditional approaches are used to obtain the 3D morphable model (3DMM) coefficients and render the respective 3D faces from a single photo. These methods are time consuming, need high complexity and not efficient. Recent approaches are using convolution neural networks (CNNs) and they are more efficient and have achieved remarkable success in 3D facial reconstruction from single photo. 3D face reconstruction from a single image faces many issues and problems such as lack of diversity in facial expression, occlusions, facial beard poses and environment conditions. Some algorithms are suggested and proposed that is going to solve the issues of large poses, wild poses, occluded poses, robustness, and detailed reconstruction. All foundational 3D face reconstruction approaches are mostly focus on profile view, occlusion and robustness but not focus on facial beard pose. The chapter emphasizes on 3D face reconstruction and the facial hair reconstruction approaches.
There are numerous approaches for 3D face shape reconstruction that reconstructs the 3D model only for frontal poses, medium poses, and uniform illumination while many challenges come when there is large pose, and occlusion occurred. The various approaches that are using single photo as an input cannot get the better result as compared to multiple images on input. The computational cost of 3D face reconstruction from multiple images is high. 3D face reconstruction from a single image is still ill-posed problem and researchers have done research for solving different issues.
The earliest methods are used 3D reference faces, landmark-based facial model and shape from shading techniques for recovering 3D facial geometry. These days, the 3D single image facial reconstruction is seldom used for modern facial recognition applications. 3DMM and CNNs are frequently used in the 3D single image facial model. CNNs applications include natural language processing, occluded pedestrian detection, image classification, decoding facial recognition and image reconstruction etc.
3D hair reconstruction plays an essential part in facial recognition and reconstruction of facial geometry. Some approaches are required user interactions, depth information for annotation the head region and hair (Chai, 2015) (Chai, 2013). In some approaches, 3D hair is reconstructed via structured light pattern (Chen, 2018). Recently, deep-learning approaches (Chao, 2016) (Zhou, 2018) are contributing to the field of 3D hair reconstruction. There are very few achievements that was reported over 3D facial hair reconstruction.