Comparative Study on Super-Resolution of Images

Abstract

Super-resolution of images has become a very important research topic nowadays. There are many algorithms that have been developed to enhance the resolution of images. In this paper, we undertake a study for evaluating and comparing three of these algorithms. These three algorithms are: neural network algorithm, wavelet extrema extrapolation algorithm, and hallucinating faces algorithm. Our study indicated that: the better performance comes at the expense of higher complexity, large database, and more computational time. The hallucinating faces algorithm gives the largest peak signal to noise ratio (PSNR) when magnifying low dimensional faces and gives better output when the database contains larger number of images. The neural network algorithm gives better results for high dimensional faces, but it needs long time for training. The wavelet extrema extrapolation algorithm gives better results for high dimensional faces than for low dimensional faces. The performance of these three algorithms gets better as the dimension of input faces gets higher and only the hallucinating faces can give good results for lower dimensional faces such as 64times48 pixels