Facial image recognition based on fractal image encoding

Fractal image encoding uses self-similarity to encode an image. Utilizing this feature, a new method of facial image recognition based on fractal image encoding is examined. Fractal image encoding can approximate any given image by capturing intrinsic self-similarity within the image. This method has some invariance to rotation, scaling, translation, and luminance. The method encodes an image of record through the use of an encoding dictionary composed of images in a face image database. The fractal codes are then placed in a file, and the fractal codes are iterated multiple times to obtain decoding images. By comparing the peak signal-to-noise ratio of the image of record to all the decoding images, the image in the image database whose decoding image minimizes this norm is the recognized image. By experiment, the recognition performance achieved by this new method posted an average recognition rate of 94.54 percent on the publicly available database of faces hosted by the Cambridge University Computer Laboratory Digital Technology Group, when facial images at a side profile orientation are within 15 degrees of a full frontal image.