CN-122027758-A - Super-resolution method and system for reconstructing high-quality 4K video through low-image-quality description

CN122027758ACN 122027758 ACN122027758 ACN 122027758ACN-122027758-A

Abstract

The invention discloses a super-resolution method and a system for reconstructing high-quality 4K video through low-image-quality description, which relate to the technical field of video image processing and comprise the steps of obtaining continuous frames of a video to be processed, constructing a time sequence frame set, respectively extracting compression block characteristics, ringing characteristics and noise characteristics from a target frame, and generating a space degradation graph; the method comprises the steps of identifying a continuous degradation region according to a space degradation graph and generating a space weight graph, inputting a time sequence frame set into a space-time reconstruction network, carrying out structural reconstruction to obtain a basic reconstruction image and generating a high-frequency residual image, carrying out space weighting adjustment on the high-frequency residual image according to the space weight graph, calculating a global degradation index of a target frame according to the space degradation graph to determine a time sequence consistency threshold, restraining residual variation between adjacent frames, and synthesizing the high-frequency residual image and the basic reconstruction image to obtain a 4K super-resolution video frame. Compression artifacts are suppressed and inter-video flicker is reduced while 4K video super-resolution reconstruction is completed.

Inventors

WU HAO
SHI DONG
MA CHENYANG
GU GUOYING
SUN BO
YAN WENWEN
GU WEI
DU WEI
LI RUOXUAN

Assignees

江苏省广播电视总台

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. A super-resolution method for reconstructing high-quality 4K video from a low-quality description, comprising: acquiring continuous frames of a video to be processed, and constructing a time sequence frame set, wherein the time sequence frame set comprises a target frame and adjacent frames; Extracting compression block characteristics, ringing characteristics and noise characteristics of a target frame, and generating a space degradation graph to serve as a low-image-quality description of the target frame; Identifying a continuous degradation region according to the spatial degradation map, and generating a spatial weight map based on the continuous degradation region; Inputting the time sequence frame set into a space-time reconstruction network to generate a basic reconstruction image and a high-frequency residual image corresponding to a target frame; Carrying out space weighting adjustment on the high-frequency residual image according to the space weight graph; calculating a global degradation index of a target frame according to the space degradation graph, and determining a time sequence consistency threshold according to the global degradation index; And constraining residual variation between adjacent frames according to the time sequence consistency threshold, and synthesizing the processed high-frequency residual image and the basic reconstruction image to obtain a 4K super-resolution video frame, thereby completing reconstruction of high-quality content.
2. The method for super-resolution of high-quality 4K video reconstructed from low-quality image description of claim 1, wherein said constructing a set of time-ordered frames comprises reading consecutive video frames from the video to be processed in time order, sequentially selecting a current frame from said consecutive video frames as a target frame, selecting a predetermined number of video frames before and after the target frame as neighboring frames according to the position of said target frame in the sequence of consecutive video frames, and combining said target frame with said neighboring frames in time order to form said set of time-ordered frames.
3. The method for reconstructing high-quality 4K video with low-quality description according to claim 2, wherein the generating a spatial degradation map comprises extracting a luminance channel of a target frame to obtain a target luminance image, detecting a luminance change at a preset block boundary based on the target luminance image to generate a compressed block feature map, detecting luminance oscillation information of an edge neighborhood based on the target luminance image to generate a ringing feature map, performing high-pass filtering based on the target luminance image, and performing local statistical analysis on a filtering result to generate a noise feature map; And normalizing the compressed block feature map, the ringing feature map and the noise feature map, and fusing the normalized feature maps to generate the space degradation map.
4. The method for super-resolution of high-quality 4K video reconstructed from a low-quality image representation of claim 3, wherein said generating a spatial weight map comprises performing statistics on pixel degradation values in said spatial degradation map to determine a degradation decision threshold; Comparing the degradation value of each pixel in the spatial degradation map with the degradation judgment threshold value, marking the pixel with the degradation value reaching the degradation judgment threshold value as a candidate degradation pixel, and generating a candidate degradation marking map; Establishing a neighborhood analysis window by taking the candidate degenerate pixels as the center, counting the number of the candidate degenerate pixels in the neighborhood, determining a continuous degenerate region and generating a continuous degenerate region mark graph when the number reaches a preset supporting threshold, distributing weights to pixels in a space degenerate graph according to the continuous degenerate region mark graph to generate an initial space weight graph, and carrying out neighborhood smoothing on pixels positioned at the boundary of the continuous degenerate region in the initial space weight graph to generate the space weight graph.
5. The method for super-resolution of high quality 4K video reconstructed from a low image quality description of claim 4, wherein said determining a degradation decision threshold comprises counting all pixel degradation values in a spatial degradation map, calculating a global map average degradation value and a degradation discrete level; The method comprises the steps of establishing a neighborhood analysis window by taking the candidate degenerate pixels as the center, counting the number of the candidate degenerate pixels in the neighborhood analysis window, and determining the corresponding pixels as continuous degenerate region pixels when the number of the candidate degenerate pixels reaches a preset supporting threshold.
6. The method for super-resolution of high-quality 4K video reconstructed by low-quality description of claim 5, wherein generating a basic reconstructed image and a high-frequency residual image corresponding to a target frame comprises respectively inputting the target frame and an adjacent frame in the time sequence frame set into a shared encoder to perform convolution feature extraction to obtain a target frame feature and an adjacent frame feature; The method comprises the steps of carrying out channel splicing on target frame features and adjacent frame features after alignment treatment, obtaining space-time fusion features through convolution fusion treatment, carrying out structural reconstruction treatment on the space-time fusion features, outputting basic reconstruction images, retaining intermediate structural features in the structural reconstruction process, carrying out difference calculation on the space-time fusion features and the intermediate structural features, extracting detail residual information, carrying out detail reconstruction treatment according to the detail residual information, and generating the high-frequency residual image.
7. The method for reconstructing high-quality 4K video with low-quality description according to claim 6, wherein the performing spatial weighting adjustment on the high-frequency residual image according to the spatial weighting map comprises expanding the spatial weighting map into weighting maps with the same number as the channels of the high-frequency residual image, multiplying the expanded weighting map by the high-frequency residual image position by position according to pixel positions, performing weighting adjustment on residual values of pixels in the high-frequency residual image, so that high-frequency residual in a degradation region is suppressed, and high-frequency residual in a non-degradation region maintains detail information to obtain a spatially weighted high-frequency residual image; The method comprises the steps of calculating global degradation indexes of a target frame, calculating a global degradation index of the target frame, wherein the calculation comprises the steps of reading all pixel degradation values in a space degradation graph, carrying out statistical calculation on all pixel degradation values to obtain the global degradation index reflecting the overall degradation degree of the target frame, determining a corresponding time sequence consistency threshold according to the position of the global degradation index in a preset degradation range, obtaining a 4K super-resolution video frame, wherein the obtaining comprises the steps of obtaining a spatial weighted high-frequency residual image of a previous frame, carrying out inter-frame alignment processing on the high-frequency residual image of the previous frame, and calculating residual change between the spatial weighted high-frequency residual image of a current frame and the aligned high-frequency residual image of the previous frame; And carrying out limiting treatment on residual variation according to the time sequence consistency threshold value to inhibit detail flicker between adjacent frames to obtain a high-frequency residual image after time sequence constraint, and carrying out position-by-position synthesis on the high-frequency residual image after time sequence constraint and the basic reconstruction image to obtain a 4K super-resolution video frame.
8. A super-resolution system for reconstructing high-quality 4K video from low-quality descriptions, based on the super-resolution method for reconstructing high-quality 4K video from low-quality descriptions according to any one of claims 1 to 7, characterized in that: The time sequence frame construction module is used for acquiring continuous frames of the video to be processed and constructing a time sequence frame set; The degradation analysis module extracts compression block characteristics, ringing characteristics and noise characteristics of the target frame and generates a space degradation map as low-image-quality description of the target frame; The time-space reconstruction module inputs the time sequence frame set into a time-space reconstruction network to generate a basic reconstruction image and a high-frequency residual image corresponding to the target frame; The residual modulation and synthesis module is used for carrying out space weighting adjustment on the high-frequency residual image according to the space weight graph; And constraining residual variation between adjacent frames according to the time sequence consistency threshold, and synthesizing the processed high-frequency residual image with the basic reconstruction image to obtain a 4K super-resolution video frame so as to finish reconstruction of high-quality content.
9. A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the computer program is characterized in that the processor executes the computer program to realize the steps of the super-division method for reconstructing high-quality 4K video through low-quality description according to any one of claims 1-7.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the superdivision method for reconstructing high quality 4K video from a low image quality description as claimed in any one of claims 1 to 7.

Description

Super-resolution method and system for reconstructing high-quality 4K video through low-image-quality description Technical Field The invention relates to the technical field of video image processing, in particular to a super-resolution method and a system for reconstructing high-quality 4K video through low-quality description. Background In the actual video production and distribution process, 4K video is usually stored or transmitted after compression encoding. In order to control the code rate, the encoder often adopts higher compression strength, which is easy to generate phenomena such as compression blocks, ringing, random noise and the like in the decoded picture under a low code rate scene. These degradation features, which are generated by compression and transmission, objectively form a low-quality description of the video frame, and degradation is usually concentrated near regions with complex textures or edges with high contrast, so that the boundaries of the picture become rough, the detail textures are weakened, and the degradation is particularly obvious when super-resolution enlarged display is performed. The conventional video superdivision method mostly adopts a deep learning model, and the detail information is recovered by carrying out feature extraction and reconstruction on a single-frame image or a multi-frame sequence. Some methods make use of redundant information between adjacent frames to supplement the missing texture details of the current frame through inter-frame alignment and feature fusion, thereby desiring to achieve the goal of reconstructing high quality content. However, in actual video, the degradation distribution tends to have a significant spatial difference, and the degree of degradation varies greatly from region to region. The existing superdivision network often lacks of accurate extraction and utilization of low-image-quality description of input video, if a unified detail restoration strategy is adopted for the whole image in the superdivision processing process, original distortion is easily amplified in a region with obvious compression artifacts, and meanwhile, the detail structure of a region which is originally clear can be possibly changed. This results in the prior art being unable to realize spatially adaptive control, and thus it is difficult to stably and truly accomplish reconstruction of high-quality contents. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides a super-resolution method and a system for reconstructing high-quality 4K video through low-quality description, which solve the problem that the prior video image quality enhancement technology is difficult to identify a degradation region and simultaneously carry out space control on detail enhancement and maintain inter-frame detail consistency. In order to solve the technical problems, the invention provides the following technical scheme: in a first aspect, the invention provides a super-resolution method and system for reconstructing high-quality 4K video through low-quality description, comprising the steps of obtaining continuous frames of video to be processed and constructing a time sequence frame set, wherein the time sequence frame set comprises a target frame and adjacent frames; Extracting compression block characteristics, ringing characteristics and noise characteristics of a target frame, and generating a space degradation graph to serve as a low-image-quality description of the target frame; Identifying a continuous degradation region according to the spatial degradation map, and generating a spatial weight map based on the continuous degradation region; Inputting the time sequence frame set into a space-time reconstruction network to generate a basic reconstruction image and a high-frequency residual image corresponding to a target frame; Carrying out space weighting adjustment on the high-frequency residual image according to the space weight graph; calculating a global degradation index of a target frame according to the space degradation graph, and determining a time sequence consistency threshold according to the global degradation index; And constraining residual variation between adjacent frames according to the time sequence consistency threshold, and synthesizing the processed high-frequency residual image and the basic reconstruction image to obtain a 4K super-resolution video frame, thereby completing reconstruction of high-quality content. The super-division method for reconstructing high-quality 4K video through low-image-quality description is characterized in that the constructing time sequence frame set comprises the steps of reading continuous video frames from video to be processed according to time sequence, sequentially selecting current frames from the continuous video frames as target frames, selecting a preset number of video frames before an