Three-step-approach with validation for face hallucination

In this paper, we propose a novel face hallucination framework using validation based on three-step-approach. In order to improve the performance of facial image reconstruction, we included validation in our framework for correcting error. That is the final result should be more accurate than the result before the validation process. In this paper, the 2D framework is applied that means the image can directly process without requiring the vectorization. Moreover, the spatial information can be preserved. Our framework is based on a three-step-approach. In the first step, error of face image reconstruction is learnt from training data set by Bilateral Two Dimensional Principal Component Analysis (B2DPCA). In this step, the validation is obtained from the error of Low-Resolution (LR) and error of High-Resolution (HR). In the second step, the global image is reconstructed by using Maximum a Posteriori (MAP) estimator and the final step is using Regression Model for Tensor (RM-T) to learn from samples data set by applying error regression analysis. The experimental results on a well-known face database demonstrate that the proposed methods can improve the face reconstruction. The results show that our method enhances the resolution and improves the quality of the face hallucination in comparison with the conventional method.