Video super-resolution reconstruction method based on deep learning and self-similarity

Abstract

The invention discloses a video super-resolution reconstruction method based on deep learning and self-similarity, which belongs to the field of video processing technologies. The video super-resolution reconstruction method comprises the steps such as video frame extraction, preliminary estimation, preliminary reconstruction, secondary reconstruction and video frame integration. The video super-resolution reconstruction method makes use of apriori constraint conditions provided by an external training set and the internal self-similarity comprehensively, and has good reconstruction effect on some smooth regions, irregular structure information which seldom occurs in a video frame sequence, and some situations such as unique and singular characteristics which rarely occur in the external training set but repeatedly occur in the video frame sequence; in addition, the video super-resolution reconstruction method does not rely on precise sub pixel motion estimation, thus can adapt to a complicated motion scene, and achieves large multiples of super-resolution reconstruction.

Description

technical field [0001] The invention relates to the technical field of video processing, in particular to a video super-resolution reconstruction method based on deep learning and self-similarity. Background technique [0002] Super-resolution reconstruction refers to a technology that uses a computer to process a low-resolution (Low Resolution, LR) image or video to obtain a high-resolution (High Resolution, HR) image or video. Super-resolution reconstruction can provide more detailed information than traditional interpolation methods, which can greatly improve the quality of images or videos. [0003] The current super-resolution reconstruction methods mainly include reconstruction methods based on learning mechanism and reconstruction methods based on self-similarity. [0004] The super-resolution method based on the learning mechanism can adapt to a large super-resolution multiple, but because it relies on a large-scale external training set, it cannot guarantee that an...

AI Technical Summary

Problems solved by technology

The self-similarity feature provides internal instances that are highly correlated with the low-resolution input, and super-resolution methods based on this internal similarity do not require additional training sets and long training time, but in the case of insufficient internal similarity blocks , tends to cause some visual artifacts due to the mismatch of the inner instance

[0006] In short, the video super-resolution reconstruction methods in the prior art have disadvantages such as poor reconstruction effect and low reconstruction magnification.

Images