An AI algorithm-based video picture resolution intelligent repair system and device

Abstract

The application provides a video picture resolution intelligent repairing system and device based on an AI algorithm.The video picture resolution intelligent repairing system based on the AI algorithm comprises a preprocessing module, the preprocessing module is used for decoding, correcting, blocking and normalizing original video materials; and an AI video repairing module, the AI video repairing module is constructed based on a deep learning algorithm of a convolutional neural network, bidirectional time series propagation and optical flow alignment to construct an AI video super-resolution model.The video picture resolution intelligent repairing system based on the AI algorithm provided by the application completes a processing link through the construction of a preprocessing module, an AI video repairing module, an auxiliary repairing module, an encoding output module and a checking module, and multiple modules are cooperated to realize video resolution improvement and intelligent repairing of a face, text and compression distortion, the preprocessing module reduces hardware load through blocking reasoning to improve processing efficiency, the quality checking module is matched, the repairing effect is stable and uniform, and the application scenarios are wide.

Description

Technical Field

[0001] This invention relates to the field of AI video restoration, and in particular to a video resolution intelligent restoration system and device based on AI algorithms. Background Technology

[0002] With the continued growth in industry demands for digital archiving of video archives, renovation of old film and television resources, and review of surveillance video quality, a large number of old film videos, tape recordings, and old surveillance videos shot and stored in the early days generally suffer from problems such as low resolution, multiple image defects, and degraded audio and video quality. Specifically, these problems manifest as insufficient image resolution, dense grain and snow noise, color fading and color cast, distortion from transcoding and compression blocks, and blurry details of faces and subtitles. They are also accompanied by defects such as video frame jitter, timing flicker, and audio noise interference, which greatly reduce the video viewing experience and the value of archive reuse.

[0003] Currently, mainstream video resolution restoration solutions are mainly divided into two categories: traditional image processing restoration solutions and single AI super-resolution network restoration solutions. Among them, traditional image processing algorithms rely on fixed filtering, interpolation and other methods to improve video resolution. They can only achieve simple noise reduction and image enlargement, but cannot intelligently restore key area details such as faces and text. They are difficult to adapt to complex and diverse old video damage scenarios, and the restored image details are severely lost and the visual appearance is harsh.

[0004] Current mainstream AI video super-resolution restoration models can improve video resolution by relying on neural networks, but they still have many inherent technical defects: First, they lack a standardized pre-processing link, resulting in poor compatibility with multiple source file formats such as MP4, MKV, old AVI, and H.264 / H.265 surveillance videos. Direct inference of ultra-high resolution videos is prone to problems such as memory overflow and inter-frame motion misalignment. Second, the models lack versatility. Currently, the models mainly complete resolution upscaling restoration, and are insufficient in repairing defects specific to old videos, such as distorted facial features, overlapping subtitle strokes, compression artifacts, and secondary problems such as screen flickering. Furthermore, they lack standardized post-processing optimization and automated image quality verification, resulting in poor color uniformity and inconsistent inter-frame timing in the restored videos, and there is no objective quantitative detection or manual color accuracy verification mechanism.

[0005] Therefore, existing video resolution restoration technologies still have room for improvement in terms of video preprocessing and the ability to repair specific defects.

[0006] Therefore, it is necessary to provide an AI-based intelligent video resolution restoration system to solve the above-mentioned technical problems. Summary of the Invention

[0007] This invention provides an intelligent video resolution restoration system based on AI algorithms, which solves the problem that existing video resolution restoration technologies still have room for improvement in video preprocessing and specific defect repair capabilities.

[0008] To address the aforementioned technical problems, the present invention provides a video resolution intelligent restoration system based on AI algorithms, comprising:

[0009] The preprocessing module is used to decode, correct, segment, and normalize the original video footage.

[0010] The AI ​​video restoration module is based on a deep learning algorithm that uses convolutional neural networks, bidirectional temporal propagation, and optical flow alignment to build an AI video super-resolution model. The AI ​​video super-resolution model is then used to process and restore pre-processed video materials.

[0011] The auxiliary repair module includes an AI face repair module, a subtitle enhancement module, and a distortion removal module. The AI ​​face repair module is used to locate the face area in the image and reconstruct the face details, eliminating face distortion. The subtitle enhancement module is used to locate the text area within the video frame, enhance the outline of the text strokes, and eliminate text blurring, adhesion, and edge artifacts. The distortion removal module is used to eliminate compression artifacts caused by multi-layer transcoding.

[0012] An encoding output module, which is used to encapsulate the repaired image sequence into a standard video file;

[0013] The verification module is used to detect the image quality indicators of the repaired video.

[0014] The auxiliary repair module also includes an old video repair module and a stabilization module. The old video repair module is used to eliminate the snow noise, physical scratches and color fading problems inherent in tape and film materials. The stabilization module is used to calculate the global motion offset between video frames and correct the image shake and frame misalignment caused by shooting shake.

[0015] The preprocessing module includes a decapsulation unit, a timing alignment unit, a color space conversion unit, and a block processing unit. The decapsulation unit is used to decapsulate the original video file and decode the bitstream to output the original image frames. The timing alignment unit extracts video frames and binds timing markers according to a set standard frame rate, constraining the correspondence between frames. The color space conversion unit performs mutual conversion between RGB and YUV color spaces and unifies them into a model-adaptive data format, simultaneously cropping invalid black border areas and blocking watermark-occluded image blocks. The block processing unit performs image block splitting on the video to distribute the single-frame video memory load.

[0016] The AI-based intelligent video resolution restoration system also includes an audio restoration module, which is used to separate the video audio track, filter out audio noise, equalize the volume, and correct audio-visual timing deviations.

[0017] The present invention also provides a device for a video resolution intelligent repair system based on AI algorithm, used in the correction module of the video resolution intelligent repair system based on AI algorithm, including: a color accuracy display and a colorimeter, wherein the color accuracy display includes a display screen, a light shield, an L-shaped plate and a positioning component;

[0018] The display screen includes an IPS LCD panel and a Mini-LED backlight structure;

[0019] The light shield includes a top cover and two side covers. The top cover is mounted on the top of the display screen, and the two side covers are mounted on the top cover and located on both sides of the display screen.

[0020] The top cover has a detection hole. One end of the L-shaped plate is rotatably installed in the detection hole via a rotating shaft. The other end of the L-shaped plate is detachably connected to the top cover. The L-shaped plate is used to seal the detection hole.

[0021] The positioning component includes a positioning block and a positioning pin. The positioning block is mounted on the L-shaped plate, and the positioning pin passes through the positioning block and is threadedly connected to the positioning block. The axis of the positioning pin is perpendicular to the display screen.

[0022] The colorimeter is used to collect color data from the display screen and generate a calibration file to correct color deviation.

[0023] Preferably, the device for the AI-based intelligent video resolution restoration system further includes a wiping structure. A frame is provided on the top of the top cover and outside the detection hole. The wiping structure is located inside the frame and on the L-shaped plate. The wiping structure includes a water storage box, a wiping part, and a valve. The wiping part includes a base plate and a wiping cotton. The base plate is detachably mounted on the water storage box, and the wiping cotton is detachably mounted on the base plate. A wetting hole is provided on the base plate. A connecting hole is provided on the top of the water storage box. The wetting hole communicates with the connecting hole. The valve is used to control the communication state of the connecting hole.

[0024] Preferably, the wiping structure further includes a drying sponge, which is installed on the side of the water storage box.

[0025] Preferably, the water storage box has mounting grooves at both ends, and the wiping part further includes two inserts. The two inserts are symmetrically installed on the bottom of the base plate, and the inserts are inserted into the mounting grooves. The inserts are detachably connected to the water storage box through connectors.

[0026] Preferably, the connector includes a connecting pin, a knob, and a nut. The knob is sleeved and fixed on the connecting pin, the nut is installed at the end of the water storage box, and one end of the connecting pin passes through the nut, the end of the water storage box, and a plug in sequence, and is threadedly connected to the nut.

[0027] Limiting holes and positioning holes are respectively provided on the L-shaped plate and the frame, and the center line connecting the limiting holes and the positioning holes is collinear with the axis of the connecting pin;

[0028] When the other end of the connecting pin passes through the limiting hole and the positioning hole in sequence, one end of the connecting pin separates from the insertion hole on the insertion block.

[0029] Preferably, the valve includes an inverted U-shaped plate and a connecting rod. The inverted U-shaped plate is slidably disposed in the water storage box. A valve hole is provided on the inverted U-shaped plate. One end of the connecting rod is connected to the inverted U-shaped plate, and the other end is rotatably connected to the connecting pin through a through hole. A strip hole is connected to the bottom of the insertion hole.

[0030] When the connecting pin passes through the insertion hole, the valve hole communicates with the connecting hole.

[0031] Compared with related technologies, the AI-based intelligent video resolution restoration system provided by this invention has the following advantages:

[0032] This invention provides an AI-based intelligent video resolution restoration system. The system constructs a complete processing chain consisting of a preprocessing module, an AI video restoration module, an auxiliary restoration module, an encoding output module, and a verification module. The multiple modules work together to achieve intelligent restoration of video resolution, including the removal of facial and text distortions and compression loss. The preprocessing module reduces hardware load and improves processing efficiency through block-based inference. Combined with the image quality verification module, the restoration effect is stable and consistent, and the system is applicable to a wide range of scenarios. Attached Figure Description

[0033] Figure 1 This invention provides a principle block diagram of an AI-based intelligent video resolution restoration system.

[0034] Figure 2 A block diagram illustrating the components of the auxiliary repair module provided by this invention;

[0035] Figure 3 A block diagram illustrating the composition of the preprocessing module provided by this invention;

[0036] Figure 4 A schematic diagram of the device for the AI-based intelligent video resolution restoration system provided by the present invention;

[0037] Figure 5 A partial structural schematic diagram of the light shield provided by the present invention;

[0038] Figure 6 A partial cross-sectional view of the light shield provided by the present invention;

[0039] Figure 7 This is a schematic diagram illustrating the usage state of the L-shaped plate provided by the present invention, wherein... Figure 7 (a) Schematic diagram of the unlocked state of the L-shaped panel and the surrounding frame. Figure 7 (b) is a schematic diagram showing the state where the wiping structure is separated from the L-shaped plate;

[0040] Figure 8 A schematic diagram illustrating the usage state of the colorimeter provided by this invention;

[0041] Figure 9 A schematic diagram of the wiping structure provided by the present invention;

[0042] Figure 10 A schematic diagram of the wiping structure provided by the present invention after the wiping part has been removed;

[0043] Figure 11 A side view of the wiping part provided by the present invention.

[0044] Numbering on the map:

[0045] 1. Display screen; 11. Base;

[0046] 2. Sunshade; 21. Top cover; 22. Side cover; 211. Enclosure; 201. Inspection hole; 202. Positioning hole;

[0047] 3. L-shaped plate; 31. Rotating shaft; 32. Limiting hole;

[0048] 4. Positioning component; 41. Positioning block; 42. Positioning pin;

[0049] 5. Wiping structure; 51. Water collection box; 52. Wiping part; 53. Drying cotton; 54. Connecting parts; 55. Valve parts; 510. Connecting hole; 511. Mounting groove; 512. Through hole; 513. Water inlet hole;

[0050] 521. Base plate; 522. Wiping cotton; 523. Insert block; 520. Wetting hole;

[0051] 541. Connecting pin; 542. Knob; 543. Nut; 501. Insertion hole; 502. Strip hole; 551. Inverted U-shaped plate; 552. Connecting rod; 550. Valve hole;

[0052] 6. Colorimeter. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0054] This invention provides a video resolution intelligent restoration system based on AI algorithms.

[0055] Please refer to the following: Figure 1 and Figure 2 In one embodiment of the present invention, the AI-based intelligent video resolution restoration system includes:

[0056] The preprocessing module is used to decode, correct, segment, and normalize the original video footage.

[0057] The AI ​​video restoration module is based on a deep learning algorithm that uses convolutional neural networks, bidirectional temporal propagation, and optical flow alignment to build an AI video super-resolution model. The AI ​​video super-resolution model is then used to process and restore pre-processed video materials.

[0058] The auxiliary repair module includes an AI face repair module, a subtitle enhancement module, and a distortion removal module. The AI ​​face repair module is used to locate the face area in the image and reconstruct the face details, eliminating face distortion. The subtitle enhancement module is used to locate the text area within the video frame, enhance the outline of the text strokes, and eliminate text blurring, adhesion, and edge artifacts. The distortion removal module is used to eliminate compression artifacts caused by multi-layer transcoding.

[0059] An encoding output module, which is used to encapsulate the repaired image sequence into a standard video file;

[0060] The verification module is used to detect the image quality indicators of the repaired video.

[0061] This system constructs a complete processing chain consisting of a preprocessing module, an AI video restoration module, an auxiliary restoration module, an encoding output module, and a verification module. The multiple modules work together to achieve intelligent restoration of video resolution, including the removal of facial and text distortions and compression loss. The preprocessing module reduces hardware load and improves processing efficiency through block-based inference. Combined with the image quality verification module, the restoration effect is stable and consistent, and it is applicable to a wide range of scenarios.

[0062] Specifically, Real-ESRGAN convolutional neural network is used to restore single-frame image quality degradation; global video inter-frame information is fused by relying on BasicVSR bidirectional temporal propagation mechanism; inter-frame pixel alignment is completed by outputting optical flow field through RAFT optical flow network; and the three types of networks are initialized by loading corresponding pre-trained weights to reduce the difficulty of model training convergence.

[0063] The AI ​​face restoration module is used to identify the face region within a video frame, specifically restore the facial features and skin details, suppress face distortion, edge artifacts and color blemishes caused by over-resolution magnification, and ensure the consistency of face feature sequence across frames and the stability of face recognition.

[0064] The subtitle enhancement module is used to locate text areas such as subtitles and road signs within video frames, enhance the outline of text strokes and eliminate text blurring and edge artifacts caused by over-resolution, and ensure the text clarity and temporal display consistency of the entire video sequence.

[0065] The distortion removal module mainly removes blocky blurring and color banding defects caused by h264 / h265 transcoding.

[0066] In a preferred embodiment, the post-processing optimization module is also included. The post-processing optimization module performs color balancing, detail softening, defect fine-tuning, and temporal smoothing on the repaired video frames to unify the visual effect of the final product. In addition to calling various auxiliary repair modules, the post-processing optimization module also uses traditional image processing algorithms such as color balancing, temporal smoothing, and edge blending for collaborative processing.

[0067] The construction of the AI ​​video restoration module specifically includes the following steps:

[0068] S1. Training Dataset Construction: Collect lossless high-definition materials as the standard answer, construct a hybrid real degradation library (simulate all image quality damage in real videos, used to generate low-definition LR maps), and generate paired samples (divide the dataset: 80% training set, 15% validation set, and 5% test set, with no overlap).

[0069] S2. Network structure initialization and pre-training configuration: BasicVSR: Initializes weights based on the pre-trained ResNet residual network, loads publicly available video super-resolution pre-trained weights to avoid slow convergence from training from zero; Real-ESRGAN: Loads the image super-resolution pre-trained model as the initial weights for the degradation repair branch; RAFT optical flow network: Uses official mature optical flow pre-trained weights to fix the optical flow layer parameters in the early training stage.

[0070] S3. Phased and Layered Training: Phase 1: Freeze the backbone and train the RAFT optical flow alignment network separately; Phase 2: Unfreeze the temporal backbone and train the BasicVSR super-resolution network; Phase 3: Jointly train BasicVSR and Real-ESRGAN multi-branch fusion training for complete joint training.

[0071] S4. Composite loss functions used in training: including L1 pixel loss, perceptual loss VGG Loss, temporal consistency loss, and adversarial loss;

[0072] S5. Real-time verification and optimization during training: Automatic testing on the validation set every fixed iteration round; Optimization strategies: Motion blurring: Increase the learning weight of the optical flow network; Fake texture or oil painting effect in the picture: Increase the perceptual loss and enhance the discriminator weight; Amplified noise in old videos: Expand the degradation dataset and increase the weight of the Real-ESRGAN branch; Severe inter-frame flicker: Increase the weight of the temporal consistency loss.

[0073] S6. Model Fine-tuning and Deployment: After the general basic model is trained, it is fine-tuned a second time using a special dataset; the trained model is deployed on the workstation GPU to process business materials in real time.

[0074] Please see Figure 2 The auxiliary repair module also includes an old video repair module and a stabilization module. The old video repair module is used to eliminate the snow noise, physical scratches and color fading problems inherent in tape and film materials. The stabilization module is used to calculate the global motion offset between video frames and correct the image shake and frame misalignment caused by shooting shake.

[0075] The old video restoration module is specifically designed to identify grainy, snowy, and compressed color block noise produced by tape recordings, old surveillance footage, and low-light materials. It distinguishes between real textures and random noise, suppresses noise amplification while preserving image details, and avoids the restored image being filled with messy white dots and color spots.

[0076] For horizontal and vertical scratches, dotted damage, and tearing marks caused by old film and worn videotapes, the system learns the texture patterns of intact images and automatically fills in the damaged areas, eliminating line defects that run through the entire image.

[0077] By studying standard film and television color benchmarks, we can correct fading problems in old images caused by storage aging, such as yellowing, reddishness, whitening, and loss of saturation, and unify the color temperature and color gamut of the image to restore a natural and balanced color performance.

[0078] Image stabilization module: Motion monitoring footage combined with AI electronic image stabilization corrects image shake and frame misalignment caused by shooting shakiness.

[0079] Please see Figure 3The preprocessing module includes a decapsulation unit, a timing alignment unit, a color space conversion unit, and a block processing unit. The decapsulation unit is used to decapsulate the original video file and decode the bitstream to output the original image frames. The timing alignment unit extracts video frames and binds timing markers according to a set standard frame rate, constraining the correspondence between frames. The color space conversion unit performs mutual conversion between RGB and YUV color spaces and unifies them into a model-adaptive data format, simultaneously cropping invalid black border areas and blocking watermark-occluded image blocks. The block processing unit performs image block splitting on the video to distribute the single-frame video memory load.

[0080] The de-encapsulation unit is used to be compatible with multiple encapsulation and encoding formats such as MP4, MOV, MKV, old AVI, monitoring H.264, and H.265, and completes the de-encapsulation and decoding of the original video file to output the original image frames.

[0081] The timing alignment unit extracts video frames evenly and binds timing markers according to a set standard frame rate, constrains the correspondence between frames, and avoids frame content misalignment and ghosting problems in motion scenes;

[0082] The color space conversion unit completes the mutual conversion between RGB and YUV color spaces and unifies them into a data format that is compatible with the model, while simultaneously cropping invalid black border areas and blocking watermark-occupied areas of the image.

[0083] The segmentation and partitioning processing unit preferentially performs image segmentation and inference on ultra-wide resolution videos, distributing the video memory load of a single frame, reducing the upper limit of hardware video memory usage, and ensuring stable model operation.

[0084] In a preferred embodiment, the system also includes a basic noise reduction preprocessing unit that uses Gaussian filtering and median filtering algorithms to filter out salt-and-pepper noise and snow-like noise from old magnetic tapes, thereby reducing the interference of native random noise on subsequent super-resolution inference.

[0085] As a preferred embodiment, the AI-based intelligent video resolution restoration system further includes an audio restoration module, which is used to separate the video audio track, filter out audio noise, equalize the volume, and correct audio-visual timing deviations.

[0086] The audio repair module separates the video audio track and performs intelligent repair, filtering out tape hiss, electrical noise, and pulse pops, suppressing recording echoes, enhancing voice clarity, unifying audio loudness and timing offset, and ensuring audio-visual synchronization after repair. After decapsulation, the video image stream and audio stream are separated. The image stream is repaired as described above. The audio stream is sent separately to the audio repair unit for synchronous processing. Finally, the encoding and scheduling layer re-encapsulates the repaired image frames and repaired audio to synthesize a complete video.

[0087] The present invention also provides a device for a video resolution intelligent restoration system based on AI algorithms.

[0088] The device for the AI-based intelligent video resolution restoration system is used in the correction module of the AI-based intelligent video resolution restoration system, and includes a color calibration display and a colorimeter 6. The color calibration display includes a display screen 1, a light shield 2, an L-shaped plate 3, and a positioning component 4.

[0089] The display screen 1 includes an IPS LCD panel and a Mini-LED backlight structure;

[0090] The light shield 2 includes a top shield 21 and two side shields 22. The top shield 21 is installed on the top of the display screen 1, and the two side shields 22 are installed on the top shield 21 and located on both sides of the display screen 1.

[0091] The top cover 21 has a detection hole 201. One end of the L-shaped plate 3 is rotatably installed in the detection hole 201 through a rotating shaft 31. The other end of the L-shaped plate 3 is detachably connected to the top cover 21. The L-shaped plate 3 is used to block the detection hole 201.

[0092] The positioning component 4 includes a positioning block 41 and a positioning pin 42. The positioning block 41 is mounted on the L-shaped plate 3. The positioning pin 42 passes through the positioning block 41 and is threadedly connected to the positioning block 41. The axis of the positioning pin 42 is perpendicular to the display screen 1.

[0093] The colorimeter 6 is used to collect color data from the display screen 1 and generate a calibration file to correct color deviation.

[0094] The color accuracy display is used to play the repaired video to check for image defects, verify the true performance of color and sharpness, and determine whether it is qualified. It can also assist in manually adjusting the color of the video file to optimize the final visual effect. In addition to using the color accuracy display to check the repaired video, this invention also includes a quantization detection algorithm module. The quantization detection algorithm module automatically outputs objective image quality indicators such as video sharpness and temporal stability, such as sharpness indicators: PSNR, SSIM, LPIPS; temporal indicators: inter-frame fluctuation, flicker coefficient, motion ghosting error; and defect indicators: noise intensity, compression artifact residue, and text blur.

[0095] The display screen 1 consists of an IPS LCD panel and a Mini-LED backlight structure. The Mini-LED (sub-millimeter light-emitting diode) has a large number of independent backlight zones, which can individually adjust the brightness of local areas. The bright areas are clear and the dark areas are pure and free of haze. It can clearly distinguish and repair minor flaws such as dark field noise, compression banding, and inter-frame flicker in videos, greatly improving the ability to identify image quality defects. At the same time, it has a high brightness limit and is suitable for monitoring high dynamic range repair materials.

[0096] IPS is responsible for stabilizing image color, while Mini-LED backlight is responsible for fine brightness and darkness layering. It takes into account the three core requirements of wide viewing angle, low color shift, and high contrast. Compared with ordinary monitors, it can completely restore the true color and time-series brightness and darkness differences of the video after AI restoration, providing a reliable display foundation for image quality verification that combines subjective and objective factors. The use of IPS LCD panel with Mini-LED backlight has the advantages of wide viewing angle, low color shift, multi-zone precise light control, and high contrast. It can truly restore the color and brightness details of the restored video, making it easy to accurately identify subtle image quality defects such as noise, color banding, and time-series flickering, and is suitable for professional color verification work in video restoration.

[0097] Among them, by setting up a light shield, it is used to isolate ambient light and reflected light from directly hitting the screen, eliminate screen reflection and glare interference, and prevent external light from covering up minor flaws such as noise in dark areas of the video, color deviation, and timing flicker, so as to ensure accurate and reliable color calibration and image quality detection results.

[0098] The colorimeter 6 periodically collects color data from the color calibration monitor and works with software to check whether the color of the color calibration monitor is up to standard and qualified, thus avoiding problems with the video files being tested due to the color calibration monitor itself.

[0099] During testing, the testing end of the colorimeter 6 needs to be placed against the display screen 1 for 2-12 minutes, preferably at the center position;

[0100] Specifically, during testing, first rotate the L-shaped plate 3 downwards until it is perpendicular to the top cover 21. At this point, the detection hole 201 opens. Then, place the colorimeter 6 on the L-shaped plate 3, pass the data cable of the colorimeter 6 through the detection hole 201, and connect it to the interface of the testing computer. Make sure the detection end of the colorimeter 6 is in contact with the display screen 1. Then, tighten the positioning pin 42. Through the threaded action, the positioning pin 42 abuts against the colorimeter 6. Figure 8 Therefore, the L-shaped plate 3 and the positioning part 4 can be used to assist in supporting and limiting the colorimeter 6, so that it can be stably attached to the display screen 1 for testing. This eliminates the need for staff to hold the colorimeter 6 for testing, or avoids the problem of poor stability of the display screen 1 when the colorimeter 6 is supported by the data cable alone.

[0101] After the test is completed, remove the colorimeter 6, and then use the L-shaped plate 3 to seal the test hole 201 again to prevent light from being reflected onto the display screen 1 through the test hole 201.

[0102] Among them, a stop block is provided at the bottom end of the L-shaped plate 3 near the rotating shaft 31. When the L-shaped plate 3 rotates to ninety degrees, the stop block abuts against the inner wall of the detection hole 201, so that the L-shaped plate 3 remains perpendicular to the top cover 21. It is also preferable to provide an anti-slip sleeve on the rotating shaft 31 to increase the friction between it and the L-shaped plate 3, so that the L-shaped plate 3 can be stably maintained at the rotation angle. The anti-slip sleeve is made of rubber, leather or other materials.

[0103] The L-shaped plate 3 comprises a straight plate and a vertical plate. The two ends of the straight plate are rounded and contact the two ends of the detection hole 201, thus sealing the entire detection hole 201. One end of the straight plate is rotatably mounted inside the detection hole 201 via a rotating shaft 31. The vertical plate is vertically mounted at the other end of the straight plate, leaving a gap between it and the inner wall of the detection hole 201, forming the L-shaped plate 3. The gap between the vertical plate and the inner wall of the detection hole 201 provides sufficient space for the L-shaped plate 3 to rotate, preventing the inner wall of the detection hole 201 from obstructing it during rotation. The detection hole 201 is correspondingly designed as a rectangular hole.

[0104] Mounting holes can be pre-set on the top of the back of the display screen 1 housing, with nuts pre-set in the mounting holes, and the top cover 21 can be fixedly installed using screws and nuts; alternatively, fasteners can be installed on the back of the display screen 1 housing using adhesive or other methods, and the top cover 21 can be installed on the fasteners using screws or other methods, with corresponding threaded holes or pre-embedded nuts on the fasteners.

[0105] A base 11 is provided at the bottom of the display screen 1 to support the display screen 1.

[0106] Please see Figures 4 to 6 As a preferred embodiment of the present invention, the device for the intelligent video resolution restoration system based on AI algorithm further includes a wiping structure 5. A frame 211 is provided on the top of the top cover 21 and outside the detection hole 201. The wiping structure 5 is disposed inside the frame 211 and located on the L-shaped plate 3. The wiping structure 5 includes a water storage box 51, a wiping part 52, and a valve 55. The wiping part 52 includes a base plate 521 and a wiping cotton 522. The base plate 521 is detachably installed on the water storage box 51, and the wiping cotton 522 is detachably installed on the base plate 521. A wetting hole 520 is provided on the base plate 521. A connecting hole 510 is provided on the top of the water storage box 51. The wetting hole 520 communicates with the connecting hole 510. The valve 55 is used to control the communication state of the connecting hole 510.

[0107] Since the colorimeter's 6 probe is directly attached to the screen surface to collect light, contaminants on the screen surface will change the light transmission and reflection path, leading to systematic measurement errors. Therefore, the display screen 1 must be wiped clean before testing; at the same time, it will affect the accuracy of the color accuracy monitor when detecting AI-repaired video defects.

[0108] By setting a frame 211 on the top cover 21 and forming a storage groove with the L-shaped plate 3, the wiping structure 5 can be placed, thus facilitating the wiping and cleaning of the display screen 1. When wiping, the valve 55 is used to open the connecting hole 510, so that the connecting hole 510 is connected to the wetting hole 520. At this time, the cleaning liquid inside the water storage box 51 can enter the wiping cotton 522, and the moistened wiping cotton 522 can easily clean the screen of the display screen 1.

[0109] During the cleaning process, the valve 55 can be used to adjust the connection status of the connecting hole 510 and adjust the amount of cleaning fluid used. After cleaning, the valve 55 can be used to seal the connecting hole 510, and then it can be placed inside the frame 211 and positioned on the L-shaped plate 3.

[0110] The wiping part 52 can be removed for cleaning, and a spare wiping part 52 can be installed, or the wiping part 52 can be cleaned before reassembly.

[0111] The wiping cotton 522 preferably includes a cotton block and a fiber cloth, with the fiber cloth wrapping the cotton block to form the wiping cotton 522; and the lens of the colorimeter 6 probe needs to be cleaned simultaneously.

[0112] The cleaning solution stored inside the water storage box 51 can be deionized water or alcohol-free and ammonia-free neutral screen cleaning solution, etc.

[0113] like Figure 10 In this embodiment, the connecting hole 510 is configured as multiple rectangular holes to facilitate the slow outflow of cleaning fluid; the wetting hole 520 can be multiple corresponding rectangular holes or a long strip hole whose length covers multiple rectangular holes.

[0114] Preferably, the side of the water reservoir 51 away from the wiping part 52 is made deformable, so that the cleaning liquid can flow out toward the wiping cotton 522 by pressing the water reservoir 51.

[0115] Please see Figure 9 As a preferred embodiment, the wiping structure 5 further includes a drying cotton 53, which is installed on the side of the water storage box 51.

[0116] By providing the drying sponge 53, after cleaning the screen of the display 1 using the wiping structure 5, the drying sponge 53 on the side of the water reservoir 51 can be used again to wipe away any remaining water stains on the screen. This makes it convenient to remove any remaining water stains from the screen.

[0117] Furthermore, the drying sponge 53 is preferably detachably installed on the side of the water storage box 51, thereby facilitating the replacement of the drying sponge 53. Installation methods include: opening a fixing groove on the side of the water storage box 51; opening a limiting groove on the inner circumferential side of the fixing groove; providing a limiting flange on the side of the drying sponge 53; installing the drying sponge 53 in the fixing groove; and having the limiting flange extend into the limiting groove to achieve installation; or opening mounting holes on the drying sponge 53 and threaded holes on the water storage box 51, using bolts to install the drying sponge 53.

[0118] Please see Figure 6 and Figure 10 The water storage box 51 has mounting grooves 511 at both ends. The wiping part 52 also includes two inserts 523. The two inserts 523 are symmetrically installed on the bottom of the base plate 521. The inserts 523 are inserted into the mounting grooves 511 respectively. The inserts 523 are detachably connected to the water storage box 51 through connectors 54.

[0119] During installation, insert the two plugs 523 at both ends of the wiping part 52 into the mounting slots 511, and then use the connector 54 to lock the plugs 523 and the water storage box 51 to install the wiping part 52. During disassembly, unlock the connector 54 to unlock the plugs 523 and the water storage box 51, so that the wiping part 52 can be removed for cleaning.

[0120] Please see Figure 6 , Figure 7 and Figure 10 As an optional solution in this embodiment, the connector 54 includes a connecting pin 541, a knob 542, and a nut 543. The knob 542 is sleeved and fixed on the connecting pin 541. The nut 543 is installed at the end of the water storage box 51. One end of the connecting pin 541 passes through the nut 543, the end of the water storage box 51, and a plug 523 in sequence, and is threadedly connected to the nut 543.

[0121] Limiting holes 32 and positioning holes 202 are respectively provided on the L-shaped plate 3 and the frame 211, and the center line connecting the limiting holes 32 and the positioning holes 202 is collinear with the axis of the connecting pin 541.

[0122] When the other end of the connecting pin 541 passes through the limiting hole 32 and the positioning hole 202 in sequence, one end of the connecting pin 541 separates from the insertion hole 501 on the insertion block 523.

[0123] After the wiping structure 5 has been used to wipe and clean the display screen 1, and after the colorimeter 6 has been used to calibrate the display screen 1, rotate the L-shaped plate 3 ninety degrees to block the detection hole 201. Then, place the wiping structure 5 inside the frame 211 and on the L-shaped plate 3, and align the connecting pin 541 with the limiting hole 32 and the positioning hole 202. Then, by turning the knob 542, the connecting pin 541 is moved out of the insertion hole 501 of the insert block 523, so that the connecting pin 541 passes through the limiting hole 32 and the positioning hole 202 respectively. This allows the wiping part 52 and the water tank 51 to be unlocked at the same time, and the L-shaped plate 3 to be connected and assembled with the frame 211. This allows the L-shaped plate 3 to stably block the detection hole 201 and stably support the wiping part 52.

[0124] Similarly, when using the colorimeter 6 later, the connecting pin 541 is rotated in the opposite direction by the knob 542, so that the connecting pin 541 re-enters the insertion hole 501 of the insertion block 523, thereby connecting the insertion block 523 with the water storage box 51. At the same time, the connecting pin 541 moves out of the positioning hole 202 and the limiting hole 32, so that the L-shaped plate 3 can be rotated downwards, which is convenient for supporting the colorimeter 6 later and is easy to use.

[0125] One end of the water storage box 51 has a through hole corresponding to the nut 543, which is connected to the mounting groove 511.

[0126] Preferably, the connecting pin 541 is threaded near the knob 542. When the thread on the connecting pin 541 is unlocked from the nut 543, the connecting pin 541 can be moved directly to pass through the limiting hole 32 and the positioning hole 202, thereby simplifying the operation.

[0127] A rubber ring is preferably provided inside the limiting hole 32 to increase friction with the connecting pin 541 and improve stability during assembly.

[0128] Alternatively, one connector 54 can be provided to connect one of the plugs 523 to the water storage box 51, or two connectors 54 can be provided to connect the two plugs 523 to the two ends of the water storage box 51 respectively. In the connector 54 corresponding to the plug 523 on the side away from the limiting hole 32, the knob 542 is provided at the end of the connecting pin 541 to shorten the length of the connecting pin 541.

[0129] As another optional solution in this embodiment, the connector 54 can also be set as a bolt, and a positioning pin 42 can be used to pass through the positioning hole 202 and the limiting hole 32 to limit the L-shaped plate 3;

[0130] Alternatively, a magnetically attachable metal sheet can be installed on the L-shaped plate 3, and a magnet can be installed in the detection hole 201 by means of a fixing block, so that the L-shaped plate 3 can be limited by magnetic attraction.

[0131] Please see Figure 6 and Figure 11 In this embodiment, the valve 55 includes an inverted U-shaped plate 551 and a connecting rod 552. The inverted U-shaped plate 551 is slidably disposed in the water storage box 51. A valve hole 550 is provided on the inverted U-shaped plate 551. One end of the connecting rod 552 is connected to the inverted U-shaped plate 551, and the other end is rotatably connected to the connecting pin 541 through a through hole 512. The bottom of the insertion hole 501 is connected to a strip hole 502.

[0132] When the connecting pin 541 passes through the insertion hole 501, the valve hole 550 communicates with the connecting hole 510.

[0133] The bottom end of the strip hole 502 is opened at the bottom of the through-hole block 523.

[0134] When using the wiping structure 5, the clean wiping part 52 is assembled with the water storage box 51, and the inserts 523 on both sides of the bottom of the wiping part 52 are inserted into the mounting grooves 511. The insert 523 with the insertion hole 501 is inserted into the mounting groove 511 on the side near the limiting hole 32. At this time, the insert 523 is sleeved on the connecting rod 552 through the strip hole 502.

[0135] Then, turn knob 542 to rotate connecting pin 541, causing one end of connecting pin 541 to separate from limiting hole 32 and positioning hole 202 in sequence, while the other end of connecting pin 541 gradually inserts into insertion hole 501. Simultaneously, connecting pin 541 pushes inverted U-shaped plate 551 to move via connecting rod 552, gradually aligning valve hole 550 with connecting hole 510. When inverted U-shaped plate 551 moves to abut against the inner wall of water storage box 51, valve hole 550 and connecting hole 510 are completely aligned. Figure 7 (a)

[0136] Thus, when manipulating the connector 54 to connect the plug 523 and the water tank 51, the connection between the L-shaped plate 3 and the frame 211 can be unlocked at the same time, and the inverted U-shaped plate 551 can be adjusted to connect the valve hole 550 and the connecting hole 510, so that the cleaning liquid can enter the wiping part 52 for wiping and cleaning, which is convenient to use.

[0137] After the colorimeter 6 completes the test on the display screen 1, similarly, when one end of the connecting pin 541 passes through the limiting hole 32 and the positioning hole 202 again, the other end of the connecting pin 541 moves out of the insertion hole 501. At this time, the wiping part 52 is disassembled for subsequent cleaning, and the inverted U-shaped plate 551 is driven by the connecting rod 552 to block the connecting hole 510.

[0138] At this time, one side of the inverted U-shaped plate 551 blocks the through hole 512, thereby preventing the cleaning fluid from entering the mounting groove 511 through the through hole 512 after the wiping part 52 is removed and the insert 523 is separated from the mounting groove 511. The through hole 512 is formed on the inner wall of the water storage box 51 and communicates with the mounting groove 511. The diameter of the through hole 512 is the same as the diameter of the connecting pin 541.

[0139] The number of valve holes 550 and connecting holes 510 are set accordingly.

[0140] The bottom of the inverted U-shaped plate 551 is embedded with ball bearings, forming a sliding connection with the bottom of the inner wall of the water storage box 51.

[0141] In other embodiments, the connecting rod 552 and the connecting pin 541 can be staggered. The connecting rod 552 passes through the water storage box 51, and a sealing sleeve is provided on the surface of the connecting rod 552 to ensure the sealing of the penetration. The position of the inverted U-shaped plate 551 can be controlled by directly pulling the connecting rod 552, so that the valve hole 550 is aligned with or staggered from the connecting hole 510.

[0142] Please see Figure 10 A water replenishment hole 513 is provided at the bottom of the water storage box 51 for replenishing cleaning fluid. Preferably, a sealing plate is provided on one side of the inverted U-shaped plate 551 to seal the water replenishment hole 513. When it is necessary to replenish the liquid, the position of the inverted U-shaped plate 551 is adjusted by the connector 54 so that the sealing plate is staggered from the water replenishment hole 513, which facilitates the replenishment of cleaning fluid.

[0143] As a preferred embodiment, a light-blocking curtain can be installed on the top of the frame 211. One end of the light-blocking curtain is fixedly connected to one end of the frame 211. The light-blocking curtain is provided with concave and convex horizontal folds, which can be unfolded or retracted along the concave and convex horizontal folds. The light-blocking curtain further covers the top of the frame 211, improves the blocking effect, and at the same time prevents dust from falling on the wiping structure 5 and the L-shaped plate 3.

[0144] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A video resolution intelligent restoration system based on AI algorithms, characterized in that, include: The preprocessing module is used to decode, correct, segment, and normalize the original video footage. The AI ​​video restoration module is based on a deep learning algorithm that uses convolutional neural networks, bidirectional temporal propagation, and optical flow alignment to build an AI video super-resolution model. The AI ​​video super-resolution model is then used to process and restore pre-processed video materials. The auxiliary repair module includes an AI face repair module, a subtitle enhancement module, and a distortion removal module. The AI ​​face repair module is used to locate the face area in the image and reconstruct the face details, eliminating face distortion. The subtitle enhancement module is used to locate the text area within the video frame, enhance the outline of the text strokes, and eliminate text blurring, adhesion, and edge artifacts. The distortion removal module is used to eliminate compression artifacts caused by multi-layer transcoding. An encoding output module, which is used to encapsulate the repaired image sequence into a standard video file; The verification module is used to detect the image quality indicators of the repaired video.

2. The intelligent video resolution restoration system based on AI algorithm according to claim 1, characterized in that, The auxiliary repair module also includes an old video repair module and a stabilization module. The old video repair module is used to eliminate the snow noise, physical scratches and color fading problems inherent in tape and film materials. The stabilization module is used to calculate the global motion offset between video frames and correct the image shake and frame misalignment caused by shooting shake.

3. The intelligent video resolution restoration system based on AI algorithm according to claim 1, characterized in that, The preprocessing module includes a decapsulation unit, a timing alignment unit, a color space conversion unit, and a block processing unit. The decapsulation unit is used to decapsulate the original video file and decode the bitstream to output the original image frames. The timing alignment unit extracts video frames and binds timing markers according to a set standard frame rate, constraining the correspondence between frames. The color space conversion unit performs mutual conversion between RGB and YUV color spaces and unifies them into a model-adaptive data format, simultaneously cropping invalid black border areas and blocking watermark-occluded image blocks. The block processing unit performs image block splitting on the video to distribute the single-frame video memory load.

4. The intelligent video resolution restoration system based on AI algorithm according to claim 1, characterized in that, It also includes an audio repair module, which is used to separate the video audio track, filter out audio noise and equalize the volume, and correct audio-visual timing deviations.

5. A device for an AI-based intelligent video resolution restoration system, characterized in that, The correction module of the AI-based intelligent video resolution restoration system as described in any one of claims 1-5 includes: a color calibration display and a colorimeter, wherein the color calibration display includes a display screen, a light shield, an L-shaped plate, and a positioning component; The display screen includes an IPS LCD panel and a Mini-LED backlight structure; The light shield includes a top cover and two side covers. The top cover is mounted on the top of the display screen, and the two side covers are mounted on the top cover and located on both sides of the display screen. The top cover has a detection hole. One end of the L-shaped plate is rotatably installed in the detection hole via a rotating shaft. The other end of the L-shaped plate is detachably connected to the top cover. The L-shaped plate is used to seal the detection hole. The positioning component includes a positioning block and a positioning pin. The positioning block is mounted on the L-shaped plate, and the positioning pin passes through the positioning block and is threadedly connected to the positioning block. The axis of the positioning pin is perpendicular to the display screen. The colorimeter is used to collect color data from the display screen and generate a calibration file to correct color deviation.

6. The apparatus for the intelligent video resolution restoration system based on AI algorithm according to claim 5, characterized in that, It also includes a wiping structure. A frame is provided on the top of the top cover and outside the detection hole. The wiping structure is located inside the frame and on the L-shaped plate. The wiping structure includes a water storage box, a wiping part, and a valve. The wiping part includes a base plate and a wiping cotton. The base plate is detachably installed on the water storage box, and the wiping cotton is detachably installed on the base plate. A wetting hole is provided on the base plate. A connecting hole is provided on the top of the water storage box. The wetting hole communicates with the connecting hole. The valve is used to control the communication state of the connecting hole.

7. The apparatus for intelligent video resolution restoration system based on AI algorithm according to claim 6, characterized in that, The wiping structure also includes a drying sponge, which is installed on the side of the water storage box.

8. The apparatus for intelligent video resolution restoration system based on AI algorithm according to claim 7, characterized in that, The water storage box has mounting slots at both ends. The wiping part also includes two inserts. The two inserts are symmetrically installed on the bottom of the base plate. The inserts are inserted into the mounting slots and are detachably connected to the water storage box via connectors.

9. The apparatus for the intelligent video resolution restoration system based on AI algorithm according to claim 8, characterized in that, The connector includes a connecting pin, a knob, and a nut. The knob is sleeved and fixed on the connecting pin. The nut is installed at the end of the water storage box. One end of the connecting pin passes through the nut, the end of the water storage box, and a plug in sequence, and is threadedly connected to the nut. Limiting holes and positioning holes are respectively provided on the L-shaped plate and the frame, and the center line connecting the limiting holes and the positioning holes is collinear with the axis of the connecting pin; When the other end of the connecting pin passes through the limiting hole and the positioning hole in sequence, one end of the connecting pin separates from the insertion hole on the insertion block.

10. The apparatus for intelligent video resolution restoration system based on AI algorithm according to claim 9, characterized in that, The valve includes an inverted U-shaped plate and a connecting rod. The inverted U-shaped plate is slidably disposed in the water storage box. A valve hole is provided on the inverted U-shaped plate. One end of the connecting rod is connected to the inverted U-shaped plate, and the other end is rotatably connected to the connecting pin through a through hole. A strip hole is connected to the bottom of the insertion hole. When the connecting pin passes through the insertion hole, the valve hole communicates with the connecting hole.

Images