S-GWO-FH: sparsity-based grey wolf optimization algorithm for face hallucination.

In recent years, face hallucination (FH) techniques are developed to generate the high-resolution (HR) version of the captured blurry low-resolution face images. The popular FH techniques transform the face hallucination problem as a least square representation (LSR) formulation. The performance of these FH techniques entirely depends on how optimally the LSR problem is minimized. Hence, in this paper, a new FH model using a sparsity-based grey wolf optimization algorithm (named S-GWO-FH) is developed which optimizes the LSR problem more efficiently. The concept of sparsity with GWO helps the proposed FH model in ignoring the dislike training images; consequently, reconstructed images have better personal characteristics. It also helps in minimizing computational overhead as it reduces the population size. Moreover, a domain-specific prior is incorporated to initialize the positions of the grey wolves, which helps the GWO algorithm to converse with the more appropriate solution. Several state-of-the-art nature-inspired optimization algorithms and conventional super-resolution techniques are considered for performance comparison. The experiments results show that the proposed S-GWO-FH model is superior to competitive algorithms in terms of reconstruction capability as well as computation time. [ABSTRACT FROM AUTHOR]