NNIR-DLRF: neural network–based image reconstruction for darker low-resolution faces

Face image reconstruction (FIR) has emerged as a compelling area of research, where high-resolution face images are reconstructed from low-resolution ones. However, current state-of-the-art FIR algorithms struggle when test faces are affected by darkness problems caused due to uneven lighting in the surroundings. Therefore, we propose a new neural network–based image reconstruction for darker low-resolution faces (named NNIR-DLRF) algorithm. In NNIR-DLRF, an artificial neural network–based darkness level estimator network is designed to estimate the level of darkness from the test images. Furthermore, the estimated value is used in the least square representation–based objective function to neglect the low-light effect from the reconstruction process. Experimental results conducted on benchmark face datasets and real-world images validate that the above-mentioned contributions allow the proposed algorithm to achieve better performance over other compared methods.