Method for reconstructing human facial image super-resolution based on similarity of facial characteristic organs

Abstract

The invention discloses a method for reconstructing human facial image super-resolution based on the similarity of facial characteristic organs. The method comprises the following steps of: 1, establishing a high-resolution front human facial image library and a high-resolution characteristic organ image library by utilizing a gray scale projection method according to a preset ideal high-resolution human facial image; 2, extracting a low-resolution characteristic organ image from a low-resolution target human facial image; 3, performing bicubic interpolation on the low-resolution target humanfacial image and the low-resolution characteristic organ image to acquire a training image set of the low-resolution image; 4, constructing characteristic space corresponding to the training image set by the training image set to reconstruct projection vectors of a corresponding high-resolution integral human facial image and a corresponding high-resolution organ image; and 5, fusing the high-resolution integral human facial image and the high-resolution characteristic organ image into a high-resolution target human facial image. The method has the characteristics of less preprocessing time, high retrieval accuracy of training images, high trueness of the acquired human facial images and the like.

Description

technical field [0001] The invention relates to a face image super-resolution reconstruction method based on the similarity of facial feature organs. Background technique [0002] The purpose of image super-resolution reconstruction technology is to effectively improve the resolution of a given low-resolution image through digital image processing technology, so as to obtain high-quality high-resolution images. Face image super-resolution reconstruction (also known as phantom face technology) is to reconstruct a high-resolution face image containing sufficient effective information from a given low-resolution frontal face image. This technology is currently widely used in video surveillance, criminal Security fields such as investigation, and the above fields have high requirements on the quality of high-resolution face images recovered by this technology. The currently widely used image super-resolution reconstruction techniques are basically divided into three categories:...

AI Technical Summary

Problems solved by technology

[0004] The learning-based face image super-resolution reconstruction method mainly has the following two defects: First, it requires complex preprocessing operations

The preprocessing process refers to establishing an effective image training set for a target image. Generally, multi-step complex operations are required, including image retrieval, scaling, image registration, and brightness normalization. Simple preprocessing is difficult to obtain ideal Effect, high-precision preprocessing methods are very time-consuming (such as optical flow method, etc.)

If the preprocessing is insufficient, the effect of the reconstructed high-resolution target image will be greatly affected, especially the method based on feature subspace is the most affected; second, it is not suitable for super-resolution reconstruction of small target images

Except for method six, the role of the training images in the above methods is to compensate for high-frequency information, that is, to retrieve high-frequency information of images or local image blocks similar to the target image in the training image set, while the low-frequency information of the training image set It only played a role in helping retrieval, and did not participate in the actual image super-resolution reconstruction process, resulting in a waste of resources

When the size of the target image is small, the target image itself basically does not contain effective high-frequency information, and the high-frequency information obtained only by relying on low-frequency image similarity retrieval cannot effectively compensate, and sometimes even lead to the opposite effect.

Images