Tangential and Exponential Kernel Weighted Regression Model for Multi-View Face Video Super Resolution

Abstract

The necessity of recognizing a person from the low resolution non-frontal image is a challenging problem in video surveillance. For alleviating the problem of recognition in images with low resolution, literature presents different techniques for face recognition after converting the images of low resolution into images of high resolution. Accordingly, this paper presents a technique for multi-view face video super resolution using the tangential and exponential kernel weighted regression model. In this paper, a novel hybrid kernel is presented to perform non-parametric kernel regression model for estimating neighbor pixels in the super-resolution after the face detection is performed using Viola-Jones algorithm. The experimentation is performed with the UCSD face video databases, and the quantitative results are analyzed using the SDME with the existing techniques. From the resulting outcome, we prove that the maximum SDME of 77.3 dB is obtained by the proposed technique while comparing with the nearest interpolation, bicubic interpolation, and bilinear interpolation techniques.