The interpolation of face/license-plate images using pyramid-based hallucination

Human faces and license plates are the most important targets for many digital surveillance systems. The image quality is too poor to recognize the targets due to the uncontrollable effects such as poor light conditions or far distances from the concerned objects. The cameras are zoomed in to make the targets to be discernible. Image interpolation and super-resolution are two popular techniques for removing the blurry effects of the zoomed images. In this paper, the super-resolution approach is applied on the reconstructions of facial and license plate images. Training samples are collected and their pyramidal edge images are built. The intuition idea for enhancing image quality is to preserve the edge data of the high resolutional images. Face and license plate hallucination are constructed. Some experimental results are conducted to show the effectiveness of the proposed approach. Some conclusions and future works are given.