الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 78 |  |  |
| | | from | 78 | | |
Abstract
Kinship recognition is becoming an important path in the evolution of biometric recognition research. Over the past decade, notable progress has been achieved in this emerging field attracting more researchers to exploit new methods for its branches. The most common use case is facial kinship recognition with its two variants: verification and identification. Kinship identification is not as well explored as kinship verification. In this thesis, we propose three novel kinship verification networks based on unified channelspatial attention models and residual blocks to expand the fields-of-view through the attention stack for producing more discriminative features. The methods represented belong to the two different schemes that form the attention to channel and spatial features in either coupled or decoupled modes. In addition, we extend our work to build a kinship identification network based on joint learning of unified verification ensembles. Experimental results on benchmark datasets are demonstrated to show the effectiveness of our proposed schemes when compared to state-of-the-art solutions on the kinship identification track.