Old movie quality enhancement method and system based on artificial intelligence

Abstract

The application discloses an old movie picture quality enhancement method and system based on artificial intelligence, and the method comprises the following steps: collecting original film materials, normalizing pixel colors, correcting the whole picture, filtering regional noise, repairing definition, optimizing picture contour constraints and enhancing the picture quality of old movies. The application belongs to the field of image processing, and specifically relates to an old movie picture quality enhancement method and system based on artificial intelligence. In the scheme, linear gray scale remodeling is completed by using the pixel extreme interval of the reference color channel itself, pixel-level cross-channel color correlation mapping is constructed, color feature natural transmission is realized, pixel-level light and shadow transparency evaluation parameters are constructed, different regions are enhanced according to the differences in transparency attenuation degree, quantitative deformation degree is constructed to realize fine boundary correction, local sub-domain error aggregation is used to improve the single-frame correction accuracy, time sequence contour constraints are added to eliminate frame-to-frame contour flicker and jitter, pixel repair and time sequence contour joint optimization are used to further improve the final old movie picture quality enhancement effect.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a method and system for enhancing the image quality of old movies based on artificial intelligence. Background Technology

[0002] Conventional old film image enhancement uses general image processing techniques such as color correction, dehazing and noise reduction, sharpening and enhancement to uniformly repair, brighten, and improve the image quality of faded, grayed, blurry, and noisy images in old films, focusing on improving the visual effect of a single frame. However, conventional old film image enhancement methods suffer from problems such as disordered light and shadow levels in red, green, and blue channels, severe level compression, residual dust and fog effects, and the inability to distinguish between effective scene textures and invalid film scratches and grain noise, resulting in low image enhancement accuracy. Conventional old film image enhancement methods also suffer from problems such as blind boundary correction, lack of local contour correction, insufficient overall correction accuracy, and a significant decrease in dynamic viewing smoothness, resulting in a very poor visual experience of image restoration. Summary of the Invention

[0003] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides an artificial intelligence-based method and system for enhancing the image quality of old films. Addressing the problems of conventional old film image enhancement methods, such as disordered light and shadow levels in the red, green, and blue channels, severe level compression, residual dust and haze, and the inability to distinguish between effective scene textures and ineffective film scratches and grain noise, leading to low image enhancement accuracy, this solution uses pixel value domain statistics to distinguish between benchmark high-quality channels and degraded channels. It utilizes the pixel extreme value range of the benchmark color channel itself to complete linear grayscale reshaping, restoring the original natural light and shadow transitions of the film. It constructs pixel-level cross-channel color correlation mapping to achieve natural color feature transfer; it performs image feature space reconstruction, gathering all degraded pixels into the positive effective range and standardizing the chaotic pixel distribution of film aging; and it constructs pixel-level light and shadow transparency assessment based on homogeneous texture region division. The method estimates parameters and performs differentiated noise filtering and adaptive brightness restoration intensity adjustment in different regions, enhancing the clarity of different areas accordingly. This improves the accuracy of old film image enhancement. Addressing the problems of conventional old film image enhancement methods, such as blind boundary correction, lack of local contour correction, insufficient overall correction accuracy, and significant decrease in dynamic viewing smoothness, resulting in poor image restoration, this solution constructs a quantitative deformation metric to accurately identify the degree of contour distortion and achieve refined boundary correction. Through local sub-domain error aggregation, it achieves differentiated contour optimization in different regions, improving single-frame correction accuracy. It also adds temporal contour constraints to eliminate inter-frame contour flickering and ensures dynamic playback stability. Finally, it jointly optimizes pixel restoration and temporal contours to avoid conflicts between image quality and contour shape resulting from independent optimization, thereby improving the final old film image enhancement effect.

[0004] The technical solution adopted by this invention is as follows: The old movie image quality enhancement method based on artificial intelligence provided by this invention includes the following steps:

[0005] Step S1: Acquisition of original video footage;

[0006] Step S2: Pixel color regularization;

[0007] Step S3: Global image calibration;

[0008] Step S4: Area noise filtering;

[0009] Step S5: Clarity Restoration;

[0010] Step S6: Optimize image outline constraints;

[0011] Step S7: Enhance the picture quality of old movies.

[0012] Furthermore, in step S1, the acquisition of the original video material involves splitting the video stream frame by frame to obtain independent image frames, completing the time-series sorting and storage of the frame sequence, synchronously retaining the positional correspondence between frames, and unifying the pixel value range of all image frames to complete the preliminary coarse screening of the original images.

[0013] Further, in step S2, the pixel color normalization specifically includes:

[0014] Step S21: Pixel value range statistics. Statistically analyze the overall pixel average of the red, green and blue channels of a single frame of the image. Select the monochrome channel with the largest global pixel average as the reference color channel, and define the other channels as degraded color channels.

[0015] Step S22: Natural reconstruction of grayscale levels. The grayscale linear reshaping is completed by using the pixel extreme value range of the reference color channel itself, stretching the compressed light and shadow levels, and restoring the original natural light and shadow transition state of the film.

[0016] Step S23: Solve the associated mapping parameters. Extract the mapping parameters that carry the original color ratio and pixel distribution of the film from the base color channel that has completed the hierarchical reconstruction. Use these parameters as the core reference for color restoration of the degraded color channel to achieve natural cross-channel transmission of original image features.

[0017] Step S24: Degraded color channel feature matching and restoration. Using the reconstructed reference color channel as a reference standard, the original image distribution features obtained from solving the reference color channel are transferred to the degraded color channel according to the spatial arrangement rules of the image texture, and the extreme value ratio of the dual channels is matched at the same time.

[0018] Step S25: Color optimization. Based on the initial color restoration, the three types of information are mixed: degraded color, baseline original lighting and shadow, and local smoothing mean. The color restoration adjustment range is limited by the color balance ratio. The entire integrated image is subjected to global uniform color deviation correction to obtain the pixel value of the single-frame color-optimized image.

[0019] Further, in step S3, the global image correction specifically includes:

[0020] Step S31: Basis deviation correction, the entire frame of image pixels is offset with the natural degradation basis as the reference to remove the interference of reduced image transparency caused by the uniform basis;

[0021] Step S32: Feature space reconstruction transformation. Based on the distribution characteristics of film image blurring and degradation, image space reconstruction is performed to make all pixels converge to the effective value range.

[0022] Step S33: Image transparency assessment. In the reconstructed image feature space, the image is divided into similar texture feature regions based on pixel gradient magnitude, color similarity, and contour connectivity. The entire frame image is divided into several homogeneous texture sub-domains. In the reconstructed feature space, the offset difference of the pixel relative to the global hazy substrate is calculated again to obtain the spatial Euclidean degradation distance. The maximum value in the region to which the pixel belongs is used as the limit to complete the local normalization process, construct the single-pixel image light and shadow transparency parameters, and quantify the degree of blur attenuation of the film image.

[0023] Further, in step S4, the regional noise filtering involves aggregating and calculating the average degradation feature value of all pixels within a single texture region, comparing the deviation between the pixels within the region and the average degradation feature value, and identifying pixels with a deviation exceeding a threshold as abnormal noise pixels. The grayscale value of the abnormal noise pixels is then replaced with the average degradation feature value of the texture feature.

[0024] Furthermore, in step S5, the clarity restoration relies on the image's light and shadow transparency parameters to restore the lost facial details, background layers, and line clarity of the film.

[0025] Furthermore, in step S6, the image contour constraint optimization specifically includes:

[0026] Step S61: Deformation difference measurement construction. Establish a spatial mapping relationship between the optimized and repaired image contour features output by S5 and the high-definition standard reference contour features. Introduce an adaptive contour mapping transformation matrix to represent the spatial deformation relationship between the two types of contours. Quantify the degree of local contour distortion deviation through matrix difference.

[0027] Step S62: Contour error aggregation. The contour region of the single-frame repaired image is divided into sub-domains. The contour deformation difference value is calculated for each sub-domain and then accumulated and integrated to obtain the overall contour error of the whole frame image, which serves as the basis for fine correction of the single-frame static contour.

[0028] Step S63: Video frame contour consistency constraint. Using continuous video segments as the calculation unit, the contour error of each frame in the sequence is statistically analyzed and the mean is calculated to establish a unified temporal contour constraint rule to ensure the stable and continuous contour of dynamic images.

[0029] Step S64: Global image quality precision optimization, integrate the image pixel-level restoration precision error and temporal contour constraint error to construct a joint optimization loss function; use the joint optimization loss as the objective function for backpropagation iterative optimization of image quality restoration, and output the final restored image with the best performance.

[0030] Furthermore, in step S7, the old movie image quality enhancement is achieved by performing real-time frame-by-frame optimization in parallel processing, relying on the light and shadow temporal change patterns of continuous video frames to complete inter-frame fusion, re-encoding and encapsulating the optimized frame sequence into a standard video format, and outputting a complete restored film as the old movie image quality enhancement result.

[0031] The old movie image quality enhancement system based on artificial intelligence provided by the present invention includes an original film material acquisition module, a pixel color normalization module, a global image correction module, a regional noise filtering module, a sharpness restoration module, an image contour constraint optimization module, and an old movie image quality enhancement module.

[0032] The original video material acquisition module splits the video stream to obtain independent image frames;

[0033] The pixel color normalization module divides the independent image frames into benchmarks and degraded color channels, and completes pixel color normalization through pixel value range statistics, grayscale reconstruction, channel restoration and color ratio adjustment.

[0034] The global image correction module reconstructs the image feature space using an orthogonal matrix and calculates light and shadow transparency evaluation parameters by combining texture region division, thereby completing the global image correction.

[0035] The regional noise filtering module calculates the average degradation features of the region based on the divided texture feature regions and performs regional noise reduction on abnormal noise pixels that exceed the threshold.

[0036] The sharpness restoration module performs numerical correction on the image brightness channel based on the light and shadow transparency evaluation parameters to achieve sharpness restoration.

[0037] The image contour constraint optimization module constructs a contour deformation difference metric, aggregates single-frame contour error and temporal contour constraint error, and achieves image quality enhancement and optimization.

[0038] The old movie image quality enhancement module integrates the frame-by-frame restoration results to complete the restoration of the entire film and obtain the old movie image quality enhancement result.

[0039] The beneficial effects achieved by the present invention using the above solution are as follows:

[0040] (1) To address the problems of conventional old film image enhancement methods, such as disordered light and shadow layers in red, green and blue channels, severe layer compression, residual dust and fog, and inability to distinguish between effective scene textures and invalid film scratches and grain noise, resulting in low image enhancement accuracy, this solution uses pixel value domain statistics to screen and distinguish between benchmark high-quality channels and degraded attenuation channels. It uses the pixel extreme value range of the benchmark color channel itself to complete linear grayscale reshaping and restore the original natural light and shadow transition of the film. It constructs pixel-level cross-channel color association mapping to achieve natural transmission of color features. It performs image feature space reconstruction to gather all degraded pixels into the positive effective range and regulate the chaotic pixel distribution of film aging. It constructs pixel-level light and shadow transparency evaluation parameters based on homogeneous texture region division, and performs regional differentiated noise filtering and adaptive adjustment of brightness restoration intensity, and enhances the transparency attenuation of different regions differently. This improves the image enhancement accuracy of old films.

[0041] (2) In view of the problems of blind boundary correction, lack of local contour partition correction, insufficient overall correction accuracy, and a significant decrease in dynamic viewing smoothness and poor picture restoration of conventional old movie image enhancement methods, this solution constructs a quantitative deformation metric to accurately identify the degree of contour distortion and achieve fine boundary correction; through local subdomain error aggregation, it achieves regional differentiated contour optimization, improves single frame correction accuracy, adds temporal contour constraints, eliminates inter-frame contour flickering and jitter, and ensures dynamic playback stability; it combines pixel restoration with temporal contour optimization to avoid the conflict between picture quality and contour shape generated by independent optimization, thereby improving the final old movie image enhancement effect. Attached Figure Description

[0042] Figure 1 A flowchart illustrating the AI-based method for enhancing the image quality of old movies provided by this invention;

[0043] Figure 2 A schematic diagram of the old movie image quality enhancement system based on artificial intelligence provided by the present invention.

[0044] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0045] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0046] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0047] Example 1, see Figure 1 The present invention provides an artificial intelligence-based method for enhancing the image quality of old movies, which includes the following steps:

[0048] Step S1: Acquire raw video footage and split the video stream to obtain independent image frames;

[0049] Step S2: Pixel color regularization. The baseline and degraded color channels are divided for independent image frames. Pixel color regularization is completed through pixel value range statistics, grayscale reconstruction, channel restoration and color ratio adjustment.

[0050] Step S3: Global image correction. The image feature space is reconstructed by an orthogonal matrix, and the light and shadow transparency evaluation parameters are calculated by combining the texture region division, thereby completing the global image correction.

[0051] Step S4: Regional noise filtering. Based on the divided texture feature regions, the average degradation features of the regions are obtained, and regional noise reduction is performed on abnormal noise pixels that exceed the threshold.

[0052] Step S5: Sharpness restoration. Based on the light and shadow transparency evaluation parameters, the image brightness channel is numerically corrected to achieve sharpness restoration.

[0053] Step S6: Image contour constraint optimization, construct contour deformation difference measurement, aggregate single frame contour error and temporal contour constraint error to achieve image quality enhancement optimization;

[0054] Step S7: Enhance the image quality of the old movie, integrate the frame-by-frame restoration results, complete the restoration of the entire film, and obtain the enhanced image quality result of the old movie.

[0055] Example 2, see Figure 1This embodiment is based on the above embodiment. In step S1, the acquisition of original film material involves splitting the video stream frame by frame to obtain independent old movie frames, completing the time-series sorting and storage of the frame sequence, synchronously retaining the positional correspondence between frames, unifying the pixel value range of all image frames, removing redundant areas of video black borders, filtering severely damaged invalid frames, and completing the preliminary coarse screening of the original images.

[0056] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, pixel color normalization relies on the differential natural attenuation characteristics of the three primary color channels of old film images to complete the internal color level division, grayscale level reshaping, and cross-channel color feature adaptive completion, thereby restoring the original retro tone of the film. Specifically, it includes:

[0057] Step S21: Pixel value range statistics. Statistically analyze the overall pixel average of the red, green, and blue channels in a single frame of the image. Select the monochrome channel with the largest global pixel average as the reference color channel, and define the remaining channels as degraded color channels. This establishes a natural reference standard for subsequent color restoration, represented as: ;in, It is the baseline color channel pixel value with the optimal color retention state within a single frame; , and These are the original pixel values ​​for the red, green, and blue channels, respectively. , and These are the global pixel averages of the red, green, and blue channels of a single frame image, used to determine the overall color attenuation of the channels; x is the pixel coordinate.

[0058] Step S22: Natural reconstruction of grayscale levels. Old films also exhibit compression of light and dark levels and discontinuous grayscale transitions in their base color channels. Therefore, linear grayscale reshaping is performed using the pixel extreme value range of the base color channel itself, stretching the compressed light and shadow levels to restore the original natural light and shadow transitions of the film. This is represented as: ;in, It is the baseline color channel pixel value after grayscale reconstruction; and These are the minimum and maximum pixel grayscale values ​​for a single frame within the base color channel; It is a smooth term, a very small positive number;

[0059] Step S23: Solve the associated mapping parameters. Extract the mapping parameters that carry the original color ratio and pixel distribution of the film from the base color channel that has completed hierarchical reconstruction. The parameters are naturally generated from the real pixel distribution characteristics of the image and serve as the core reference for color restoration of degraded color channels, realizing the natural transfer of original image features across channels. This is represented as: ;in, These are pixel-level color distribution association mapping parameters;

[0060] Step S24: Degraded color channel feature matching and restoration. Using the reconstructed baseline color channel as a reference standard, the original image distribution features obtained from solving the baseline color channel are transferred to the degraded color channel according to the spatial arrangement rules of the image texture. At the same time, the extreme value ratio of the two channels is matched to fill in the missing colors while avoiding over-rendering of colors that deviates from the texture of the era. This is represented as: ;in, It is the global average value of the mapping parameters within a single frame; These are the degraded color channel pixel values ​​that have been initially restored to their original color state; These are the pixel space coordinates corresponding to the degraded color channel; It is the maximum pixel value of the original degraded color channel;

[0061] Step S25: Color Optimization. Based on the initial color restoration, three types of information are mixed: degraded colors, baseline original lighting and shadows, and local smoothing mean. The color restoration adjustment range is limited by the color balance ratio. A uniform color deviation correction is performed on the entire integrated image frame to obtain the pixel value of the single-frame color-optimized image, represented as: ;in, It is the overall color balance ratio within the frame, which is calculated by the total cumulative pixel count of the baseline color channel and the degradation channel. Based on the overall color storage distribution pattern of the image, the range of color restoration is limited. It is the final pixel value of a single frame image after color optimization and integration; , These are the pixel values ​​of the degradation channel and the reference color channel; It is the average pixel value of a local area in the degraded color channel; This is the light and shadow balance adjustment coefficient, with a value ranging from 0.7 to 0.9; It is the smoothing constraint coefficient, with a value ranging from 0.1 to 0.3; It is the sum of the pixel values ​​of all pixels in a single frame of the base color channel; It is the sum of the pixel values ​​of all pixels in a single frame of the degraded color channel.

[0062] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, global image correction addresses the overall degradation problems of global dusting, image haze, and overall grayscale shift caused by long-term storage of old films. It eliminates globally uniform degradation interference terms through substrate stripping and image feature space reconstruction, calculates the degree of image clarity attenuation, and provides a precise quantitative basis for subsequent detail clarity restoration. Specifically, it includes:

[0063] Step S31: Base deviation correction. Long-term storage of old films results in widespread dusting, image haziness, and overall grayscale shift. A uniform, hazy base value exists due to film aging. The entire frame's pixels are offset against this naturally degraded base, removing the interference of decreased image clarity caused by the uniform base and highlighting the true texture and color information of the scene. This is represented as: ;in, B is the corrected pixel value after removing global hazy background interference; B is the global film aging hazy background value of a single frame of film, which is obtained by extracting pixels in the dark, textureless areas of a single frame image and calculating the stable grayscale mean. These are the image pixel values ​​after color optimization processing;

[0064] Step S32: Feature Space Reconstruction Transformation. Within the conventional color space, the pixel distribution of hazy and degraded old films is chaotic and disordered, easily generating invalid negative pixel values, causing deviations in the image clarity assessment. Therefore, based on the distribution characteristics of hazy and degraded film images, image space reconstruction is performed to converge all pixels to the effective value range, reducing the difficulty of fitting image degradation features. This is represented as: ; ;in, is the image pixel value after the degradation feature space reconstruction is completed; M is the orthogonal transformation matrix; absolute value operation is used to uniformly converge pixels to the positive effective value range and avoid the generation of invalid negative pixel data;

[0065] Step S33: Image clarity assessment. In the reconstructed image feature space, the image is divided into homogeneous texture feature regions based on pixel gradient magnitude, color similarity, and contour connectivity. The entire frame image is divided into several homogeneous texture sub-domains. A non-overlapping global segmentation method is used to ensure that each pixel uniquely belongs to only one homogeneous texture sub-domain. The offset difference of the pixel relative to the global hazy substrate is calculated again in the reconstructed feature space. The spatial Euclidean degradation distance is calculated, and the maximum value within the texture region of the pixel is used to perform local normalization. This constructs the light and shadow transparency parameters of a single pixel image, quantifies the degree of blurring and attenuation of the film image, and is expressed as: ;in, It is an evaluation parameter for the light and shadow transparency of the image corresponding to a single pixel position. The closer the value is to 1, the higher the transparency of the image. It represents the spatial coordinates of any pixel within the image; It is the i-th texture feature region.

[0066] Example 5, see Figure 1This embodiment is based on the above embodiment. In step S4, the regional noise filtering is to use the regional mean feature to smooth and suppress noise in the divided texture feature regions, retain the effective texture features of the main scene, and remove invalid interference information. Specifically, the operation is as follows: the average degradation feature value of all pixels in a single texture region is aggregated and calculated. The deviation between the pixels in the region and the average degradation feature value is compared. Pixels with a deviation exceeding the threshold (1.5 to 2.0 times the standard deviation of the texture region pixel features) are judged as abnormal noise pixels such as film grains and small scratches that deviate from the overall degradation pattern. The gray value of the abnormal noise pixels is replaced with the average degradation feature value of the texture features to smooth out the numerical fluctuations caused by film grains and small scratches, and completely retain the original light and shadow layers and image texture of the region. The average degradation feature value is expressed as: ;in, It is the average degradation feature value corresponding to the same type of texture region; It is the i-th texture region of the same type; T is the total number of pixels contained in the current single texture region; the summation operation is used to integrate the degradation features of all pixels in the region to achieve noise smoothing filtering.

[0067] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the sharpness restoration only targets the brightness that causes the image to be blurry and lose details, while preserving all original pixel information of the color throughout the process. Relying on the image's light and shadow transparency parameters, it restores the facial details of people, the depth of distant scenes, and the sharpness of lines lost in the film. This can be expressed as: ;in, It is the pixel value of the image brightness channel after the detail sharpness has been restored; These are the original luminance channel values ​​obtained by splitting the CIELAB color space after regional noise filtering. It is the luminance reference component corresponding to the overall film hazy substrate.

[0068] By performing the above operations, this solution addresses the problems of conventional old film image enhancement methods, such as disordered light and shadow levels in the red, green, and blue channels, severe level compression, residual dust and haze, and the inability to distinguish effective scene textures from invalid film scratches and grain noise, resulting in low image enhancement accuracy. This solution uses pixel value domain statistics to screen and distinguish between benchmark high-quality channels and degraded channels. It utilizes the extreme value range of the benchmark color channel itself to complete linear grayscale reshaping, restoring the original natural light and shadow transitions of the film. It constructs pixel-level cross-channel color correlation mapping to achieve natural color feature transfer. It reconstructs the image feature space, gathering all degraded pixels into the positive effective range and standardizing the chaotic pixel distribution caused by film aging. Based on homogeneous texture region division, it constructs pixel-level light and shadow transparency evaluation parameters and performs differentiated noise filtering and adaptive brightness restoration intensity adjustment in different regions, enhancing the image quality of old films differently according to the degree of transparency attenuation in different areas. This ultimately improves the image quality enhancement accuracy of old films.

[0069] Example 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, the image contour constraint optimization addresses the problems of blurred character contours, offset scene boundaries, and flickering distortion of dynamic image contours that easily occur after old film restoration. First, it completes the fine-grained deviation correction of single-frame contours, then extends it to the contour consistency constraint of continuous temporal frames, and finally constructs a global joint optimization loss function to complete parameter iterative tuning. While preserving the main image quality restoration effect, it unifies and standardizes the image contours, improving the temporal stability of video playback. Specifically, it includes:

[0070] Step S61: Deformation difference metric construction. A spatial mapping relationship is established between the optimized and restored image contour features output from S5 and the high-definition standard baseline contour features. An adaptive contour mapping transformation matrix is ​​introduced to represent the spatial deformation relationship between the two types of contours. The degree of local contour distortion deviation is quantified by matrix difference, expressed as: ;in, is a measure of spatial deformation difference in the local contour of the image; p is the spatial location of the feature point of the image contour. It is the contour space adaptive mapping transformation matrix, which is obtained by feature registration; It is a third-order identity matrix, representing the standard contour mapping state without any morphological distortion; It is the L2 norm, used to quantify the overall numerical deviation of the matrix;

[0071] Step S62: Contour error aggregation. The contour region of the single-frame restored image is divided into sub-domains. The contour deformation difference value is calculated for each sub-domain, and the values ​​are accumulated and integrated to obtain the overall contour error of the entire frame image. This is used as the basis for fine correction of the single-frame static contour, as shown below: ;in, It refers to the overall outline error of a single frame of video footage; It is a contour feature map of the image extracted after the clarity restoration; It is the baseline contour feature map for high-definition restoration of old movies; N is the total number of contour subdomains in a single frame, ranging from 20 to 80, and s is the subdomain index; , It is the output contour feature and the standard contour feature corresponding to the s-th local subdomain;

[0072] Step S63: Video frame contour consistency constraint. Using continuous video segments as the calculation unit, the contour error of each frame in the sequence is statistically analyzed and the mean is calculated to establish a temporal contour uniform constraint rule, ensuring the stability and continuity of the dynamic picture contour, as shown below: ;in, Z is the overall contour constraint error of consecutive temporal movie frames; Z is the total number of consecutive movie frames participating in the temporal constraint calculation, with a value of 8 to 32; h is the index of consecutive movie frames. It is the contour error corresponding to the h-th frame in the time sequence;

[0073] Step S64: Global image quality accuracy optimization, fusing the pixel-level restoration accuracy error and temporal contour constraint error to construct a joint optimization loss function, expressed as: ;in, It is the joint optimization loss function for restoring the image quality of old movies; It is the error in the accuracy of global pixel-level restoration of the image; It is an image with restored clarity; It is a high-definition, genuine standard reference image; It is the contour constraint weight hyperparameter, with a value ranging from -0.2 to 0;

[0074] The joint optimization loss is used as the objective function for backpropagation iterative optimization of image quality restoration. During the training process, the optimization direction is to minimize the joint optimization loss. Based on the loss value, the hyperparameters of each step in the restoration process are adjusted in reverse. This continues until the loss value tends to stabilize and converge, thus completing the optimal parameter tuning for the entire set of old movie image quality restoration and outputting the final restored image with the best performance.

[0075] By performing the above operations, this solution addresses the problems of conventional methods for enhancing the image quality of old movies, such as blind boundary correction, lack of local contour partitioning correction, insufficient overall correction accuracy, and a significant decrease in dynamic viewing smoothness, resulting in extremely poor image restoration. It constructs a quantitative deformation metric to accurately identify the degree of contour distortion and achieve refined boundary correction. Through local sub-domain error aggregation, it achieves differentiated contour optimization in different regions, improving single-frame correction accuracy. It also adds temporal contour constraints to eliminate inter-frame contour flickering and jitter, ensuring dynamic playback stability. Furthermore, it jointly optimizes pixel restoration and temporal contours to avoid conflicts between image quality and contour shape resulting from independent optimization, thereby improving the final image quality enhancement effect for old movies.

[0076] Example 8, see Figure 1 This embodiment is based on the above embodiment. In step S7, the old movie image quality enhancement is carried out after the parallel processing of real-time frame-by-frame static color correction, noise reduction and smoothing, brightness and clarity enhancement and contour constraint optimization. It relies on the light and shadow temporal change law of continuous video frames to complete the inter-frame fusion, eliminate the sense of screen fragmentation caused by independent repair of single frames, re-encode and encapsulate the optimized frame sequence into a standard video format, and output a complete restored film with clear and natural image quality and retaining the original texture of the era, as the result of old movie image quality enhancement.

[0077] Example 9, see Figure 2 Based on the above embodiments, the artificial intelligence-based old movie image quality enhancement system provided by the present invention includes an original film material acquisition module, a pixel color normalization module, a global image correction module, a regional noise filtering module, a sharpness restoration module, an image contour constraint optimization module, and an old movie image quality enhancement module.

[0078] The original video material acquisition module splits the video stream to obtain independent image frames;

[0079] The pixel color normalization module divides the independent image frames into benchmarks and degraded color channels, and completes pixel color normalization through pixel value range statistics, grayscale reconstruction, channel restoration and color ratio adjustment.

[0080] The global image correction module reconstructs the image feature space using an orthogonal matrix and calculates light and shadow transparency evaluation parameters by combining texture region division, thereby completing the global image correction.

[0081] The regional noise filtering module calculates the average degradation features of the region based on the divided texture feature regions and performs regional noise reduction on abnormal noise pixels that exceed the threshold.

[0082] The sharpness restoration module performs numerical correction on the image brightness channel based on the light and shadow transparency evaluation parameters to achieve sharpness restoration.

[0083] The image contour constraint optimization module constructs a contour deformation difference metric, aggregates single-frame contour error and temporal contour constraint error, and achieves image quality enhancement and optimization.

[0084] The old movie image quality enhancement module integrates the frame-by-frame restoration results to complete the restoration of the entire film and obtain the old movie image quality enhancement result.

[0085] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0086] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0087] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for enhancing the image quality of old movies based on artificial intelligence, characterized by: The method includes the following steps: Step S1: Acquire raw video footage and split the video stream to obtain independent image frames; Step S2: Pixel color regularization. The baseline and degraded color channels are divided for independent image frames. Pixel color regularization is completed through pixel value range statistics, grayscale reconstruction, channel restoration and color ratio adjustment. Step S3: Global image correction. The image feature space is reconstructed by an orthogonal matrix, and the light and shadow transparency evaluation parameters are calculated by combining the texture region division, thereby completing the global image correction. Step S4: Regional noise filtering. Based on the divided texture feature regions, the average degradation features of the regions are obtained, and regional noise reduction is performed on abnormal noise pixels that exceed the threshold. Step S5: Sharpness restoration. Based on the light and shadow transparency evaluation parameters, the image brightness channel is numerically corrected to achieve sharpness restoration. Step S6: Image contour constraint optimization, construct contour deformation difference measurement, aggregate single frame contour error and temporal contour constraint error to achieve image quality enhancement optimization; Step S7: Enhance the image quality of the old movie, integrate the frame-by-frame restoration results, complete the restoration of the entire film, and obtain the enhanced image quality result of the old movie.

2. The method for enhancing the image quality of old movies based on artificial intelligence according to claim 1, characterized in that: In step S2, the pixel color normalization specifically includes: Step S21: Pixel value range statistics. Statistically analyze the overall pixel average of the red, green and blue channels of a single frame of the image. Select the monochrome channel with the largest global pixel average as the reference color channel, and define the other channels as degraded color channels. Step S22: Natural reconstruction of grayscale levels. The grayscale linear reshaping is completed by using the pixel extreme value range of the reference color channel itself, stretching the compressed light and shadow levels, and restoring the original natural light and shadow transition state of the film. Step S23: Solve the associated mapping parameters. Extract the mapping parameters that carry the original color ratio and pixel distribution of the film from the base color channel that has completed the hierarchical reconstruction. Use these parameters as the core reference for color restoration of the degraded color channel to achieve natural cross-channel transmission of original image features. Step S24: Degraded color channel feature matching and restoration. Using the reconstructed reference color channel as a reference standard, the original image distribution features obtained from solving the reference color channel are transferred to the degraded color channel according to the spatial arrangement rules of the image texture, and the extreme value ratio of the dual channels is matched at the same time. Step S25: Color optimization. Based on the initial color restoration, the three types of information are mixed: degraded color, baseline original lighting and shadow, and local smoothing mean. The color restoration adjustment range is limited by the color balance ratio. The entire integrated image is subjected to global uniform color deviation correction to obtain the pixel value of the single-frame color-optimized image.

3. The method for enhancing the image quality of old movies based on artificial intelligence according to claim 1, characterized in that: In step S3, the global image correction specifically includes: Step S31: Basis deviation correction, the entire frame of image pixels is offset with the natural degradation basis as the reference to remove the interference of reduced image transparency caused by the uniform basis; Step S32: Feature space reconstruction transformation. Based on the distribution characteristics of film image blurring and degradation, image space reconstruction is performed to make all pixels converge to the effective value range. Step S33: Image clarity assessment.

4. The method for enhancing the image quality of old movies based on artificial intelligence according to claim 1, characterized in that: In step S3, the image transparency assessment involves dividing the image into homogeneous texture feature regions based on pixel gradient magnitude, color similarity, and contour connectivity within the reconstructed image feature space. This divides the entire frame into several homogeneous texture sub-domains. Within the reconstructed feature space, the offset difference of the pixel relative to the global hazy substrate is calculated again to obtain the spatial Euclidean degradation distance. Using the texture region to which the pixel belongs as a limit, the maximum value within the region is used to complete local normalization processing, constructing single-pixel image light and shadow transparency parameters, and quantifying the degree of film image blur attenuation.

5. The method for enhancing the image quality of old movies based on artificial intelligence according to claim 1, characterized in that: In step S4, the regional noise filtering involves aggregating and calculating the average degradation feature value of all pixels within a single texture region, comparing the deviation between the pixels within the region and the average degradation feature value, and identifying pixels with a deviation exceeding a threshold as abnormal noise pixels. The grayscale value of the abnormal noise pixels is then replaced with the average degradation feature value of the texture feature.

6. The method for enhancing the image quality of old movies based on artificial intelligence according to claim 1, characterized in that: In step S5, the clarity restoration relies on the image's light and shadow transparency parameters to restore the lost facial details, background layers, and line clarity of the film.

7. The method for enhancing the image quality of old movies based on artificial intelligence according to claim 1, characterized in that: In step S6, the image contour constraint optimization specifically includes: Step S61: Deformation difference measurement construction. Establish a spatial mapping relationship between the optimized and repaired image contour features output by S5 and the high-definition standard reference contour features. Introduce an adaptive contour mapping transformation matrix to represent the spatial deformation relationship between the two types of contours. Quantify the degree of local contour distortion deviation through matrix difference. Step S62: Contour error aggregation. The contour region of the single-frame repaired image is divided into sub-domains. The contour deformation difference value is calculated for each sub-domain and then accumulated and integrated to obtain the overall contour error of the whole frame image, which serves as the basis for fine correction of the single-frame static contour. Step S63: Video frame contour consistency constraint; Step S64: Global image quality precision optimization, integrate the image pixel-level restoration precision error and temporal contour constraint error to construct a joint optimization loss function; use the joint optimization loss as the objective function for backpropagation iterative optimization of image quality restoration, and output the final restored image with the best performance.

8. The method for enhancing the image quality of old movies based on artificial intelligence according to claim 1, characterized in that: In step S6, the video frame contour consistency constraint uses continuous video segments as the calculation unit, calculates the contour error of each frame in the sequence and takes the mean value, establishes a temporal contour uniform constraint rule, and ensures the stable and continuous contour of dynamic images.

9. The method for enhancing the image quality of old movies based on artificial intelligence according to claim 1, characterized in that: In step S7, the old movie image quality enhancement is achieved by performing real-time frame-by-frame optimization in parallel processing, then relying on the light and shadow temporal change patterns of continuous video frames to complete inter-frame fusion, re-encoding and encapsulating the optimized frame sequence into a standard video format, and outputting a complete restored film as the old movie image quality enhancement result.

10. An AI-based old movie image quality enhancement system, used to implement the AI-based old movie image quality enhancement method as described in any one of claims 1-9, characterized in that: It includes modules for acquiring original video footage, pixel color normalization, global image correction, regional noise filtering, sharpness restoration, image outline constraint optimization, and old movie image quality enhancement. The original video material acquisition module splits the video stream to obtain independent image frames; The pixel color normalization module divides the independent image frames into benchmarks and degraded color channels, and completes pixel color normalization through pixel value range statistics, grayscale reconstruction, channel restoration and color ratio adjustment. The global image correction module reconstructs the image feature space using an orthogonal matrix and calculates light and shadow transparency evaluation parameters by combining texture region division, thereby completing the global image correction. The regional noise filtering module calculates the average degradation features of the region based on the divided texture feature regions and performs regional noise reduction on abnormal noise pixels that exceed the threshold. The sharpness restoration module performs numerical correction on the image brightness channel based on the light and shadow transparency evaluation parameters to achieve sharpness restoration. The image contour constraint optimization module constructs a contour deformation difference metric, aggregates single-frame contour error and temporal contour constraint error, and achieves image quality enhancement and optimization. The old movie image quality enhancement module integrates the frame-by-frame restoration results to complete the restoration of the entire film and obtain the old movie image quality enhancement result.

Images