This work aims to develop a method for automated segmentation of thigh muscle groups in three-dimensional (3D) thigh magnetic resonance images (MRI). Thigh muscle groups lie anatomically adjacent to each other, rendering their manual delineation a challenging and time-consuming task. The proposed technique employs image and label fusion of multi- atlas segmentation (MAS) with 3D deep learning model-based segmentation. It leverages the generalizability of MAS and the representation power of deep networks for accurate segmentation of individual muscle groups, which is critical for accurate assessment of their volume and fat content. We evaluated the performance of the proposed framework and baseline methods on MRIs of 15 healthy subjects by threefold cross-validation and further tested it on four patients. The experimental results indicated that fusion of anatomical mappings from multiple MAS techniques produces accurate muscle group segmentation. Additional fusion of MAS with deep network segmentation, further improves muscle delineation. The proposed approach can enable accurate and timely measurement of muscle structure and composition on large scale clinical studies, and lead to better understanding of body composition changes in metabolic and age-related diseases.