Image alignment method based on novel end-to-end face super-resolution network

Abstract

The invention provides an image alignment method based on a novel end-to-end face super-resolution network. In the image alignment method, a face super-resolution network and a face super-resolution generative adversarial network are mainly involved; the image alignment method comprises the following process: constructing a coarse-grained super-resolution (SR) network for recovering a coarse-grained high-resolution (HR) image; sending the coarse-grained HR image to a fine-grained SR encoder and a prior information estimating network respectively, wherein the fine-grained SR encoder is used forextracting image features and the prior information estimating network is used for estimating a critical point heat map; restoring the HR image by using the image features and prior information in adecoder, wherein in order to further generate a realistic face, the face super-resolution generative adversarial network is provided and adversarial loss is brought into the face super-resolution network. A network consisting of the face super-resolution network and the face super-resolution generative adversarial network is used for image alignment, so that the imaging quality can be effectivelyimproved and the image alignment method is simplified.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an image alignment method based on a novel end-to-end face super-resolution network. Background technique [0002] Image alignment is a basic problem in image processing, which is widely used in many fields such as public security, military investigation, and medical image processing. Specifically, in the field of public security, the collected surveillance images can be processed in case investigation, the feature points of the two surveillance images can be extracted, and the feature point sets of the two images can be aligned to achieve image stitching and target recognition, effectively identifying Suspicious personnel in monitoring, further enhancing security capabilities. In the field of military reconnaissance, in the face of complex military environments, such as field environment operations and aerial UAV reconnaissance, using image alignment can improve the recognition ...

AI Technical Summary

Problems solved by technology

Although there are many studies on face super-resolution, due to the complexity of the face, most of the existing methods use multi-stage instead of end-to-end training, which makes the method too cumbersome and complicated, so it is necessary to ensure the super-resolution image On the premise of quality, there are still some challenges in further simplifying research methods

Images