Error Estimation by Regression Model and Eigentransformation for Face Hallucination

Abstract

In this paper, we propose a new face hallucination using Eigen transformation with error regression model. Normally in the traditional methods, a high-resolution (HR) face image is reconstructed only from low-resolution (LR) face image. Nevertheless, none of these works interested to take advantage of reconstruction error. Therefore, the error information is included in our framework to correct the final result. In this way, the error estimation can be used from the existing LR feature in eigen space to be found by regression analysis, in order to improve the performance of facial image reconstruction. Our framework consists of two-phase series. In the first phase, learning process is from the mistakes in reconstruct face images of the training dataset by Eigen transformation, then finding the relationship between input and error by regression analysis. In the second phase, hallucinating process uses normal method by Eigen transformation, after that the result is corrected with the error estimation. Experimental results on the well-known face databases show that the resolution and quality of the hallucinated face images are greatly enhanced over the traditional Eigen transformation method, which is very helpful for face recognition.