Face Image Super Resolution based on Smooth Regression with Local Structure Prior: A Comparative Study

R. Gokulakrishnan,Dr. G. Zayaraz,r. S. Ravi,R. Bhavya

In recent years, face image super-resolution (SR) is identified as one of the major research areas in digital image processing field. face image super-resolution is also known as Face hallucination. Due to its importance in the research field, it is better to learn about super-resolution techniques which, have several applications. This paper provides a comprehensive comparative study on face image super resolution with help of three smooth regression with local structure priors. All the three methods used the interpolation methods for determining the missing intensity values in the output images. The first method is based on a smooth regression with local structure prior using two freely available face databases and some real-world images proving its efficiency and effectiveness for face image super resolution. The second method introduced the fused smooth constraint and locality-based smooth constraint for obtaining a stable reconstruction weights using FEI face database and CMU+MIT face databases method producing much better results in presence of strong noise in the input loe resolution images. The third method presented a new face SR method using Tikhonov regularized neighbor representation with the help of a threshold for searching the best similar patches from the training images and discarding patches that are dissimilar to the input patch. Even though several novel algorithms have been proposed in the past decades, face image super resolution approaches using smooth regression achieves comparatively better performance than other methods. But the problem with these SR-based methods is that they produce better results only if the input images are noiseless or with very lesser noise. If the input image contains more noise, the reconstruction coefficients of the input low-resolution patches using SR-based methods will result in a poor reconstruction performance. The extensive experimental results produce much better performance for face recognition in terms of recognition accuracy.

Volume 11 | 05-Special Issue

Pages: 2116-2125