This thesis investigates kinship recognition in real-world settings. The thesis first provides a survey on existing kinship verification methods and datasets, highlighting promising areas for future research. Based on the challenges, Chapter 3 proposes a novel kinship identification approach based on joint training of kinship verification ensembles and classification modules. Chapter 4 introduces a unified approach to kinship verification for child-adult pairs to solve the side effects of aging variances among facial images. Given the challenges of kinship recognition in open-set scenarios, Chapter 5 proposes a more general task called Open-set Kinship Similarity Measurement. A pairwise-based method is developed using hierarchical information. In Chapter 6, the thesis explores the use of off-the-shelf knowledge from pretrained facial networks to improve kinship verification performance when working with limited kinship datasets. The findings of this thesis have the potential to facilitate the development of more advanced algorithms for kinship recognition in real-world scenarios.