Face Hallucination using Eigen Transformation in Transform Domain

摘要

Faces often appear very small in surveillance imagery because of the wide fields of view that are typically used and the relatively large distance between the cameras and the scene. In applications like face recognition, face detection etc. resolution enhancement techniques are therefore generally essential. Super resolution is the process of determining and adding missing high frequency information in the image to improve the resolution. It is highly useful in the areas of recognition, identification, compression, etc. Face hallucination is a subset of super resolution. This work is intended to enhance the visual quality and resolution of a facial image. It focuses on the eigen transform based face super resolution techniques in transform domain. Advantage of eigen transformation based technique is that, it does not require iterative optimization techniques and hence comparatively faster. Eigen transform is performed in wavelet transform and discrete cosine transform domains and the results are presented. The results establish the fact that the eigen transform is efficient in transform domain also and thus it can be directly applied with slight modifications on the compressed images.