Search

CN-122023200-A - Cultural relic pattern form digital restoration method and system based on generated model

CN122023200ACN 122023200 ACN122023200 ACN 122023200ACN-122023200-A

Abstract

The application discloses a method and a system for digitally repairing a pattern shape of a cultural relic based on a generated model, which relate to the technical field of digital cultural relic repair and currently provide a scheme which comprises the steps of acquiring a surface image set of a target cultural relic, and (3) obtaining a substrate grid through three-dimensional reconstruction and standardization, and fusing registration projections of the surface image set in the substrate grid to form a grid pattern evidence and multi-view consistency measure, wherein the grid pattern evidence and multi-view consistency measure is based. According to the method, the event closed-loop error is constructed and is thrown back to the three-dimensional substrate grid positioning error section, the central line and the transverse width distribution are iteratively optimized along the section to update the geometric parameters of the strip-shaped body, the defect that the structural mismatch is difficult to trace to the source and the global deviation is caused by local correction caused by correction divergence is overcome, and a self-correction closed-loop control mechanism from the structural error to the geometric parameters is established, so that the repair process has convergence optimization paths and stable global structural consistency.

Inventors

  • ZHANG YUYING
  • WU HAO
  • ZHENG DIE
  • WU MAOQI

Assignees

  • 四川师范大学

Dates

Publication Date
20260512
Application Date
20260415

Claims (10)

  1. 1. The cultural relic pattern shape digital restoration method based on the generated model is characterized by comprising the following steps of: obtaining a surface image set of a target cultural relic, obtaining a substrate grid through three-dimensional reconstruction and standardization processing, and fusing registration projections of the surface image set in the substrate grid to form a grid pattern evidence and a multi-view consistency measure; Based on the lattice pattern evidence and multi-view consistency measurement, determining a pattern strip body which takes the main trend of the pattern as a long axis and has limited transverse width on a substrate grid, and extracting the distribution of the central line and the transverse width of the strip body to form strip geometric parameters; parameterizing and expanding the strip-shaped body of the strip sample into a two-dimensional strip repair domain by utilizing strip geometric parameters, recording the mapping relation between the strip-shaped body of the strip sample and a substrate grid, and resampling the grid strip-shaped evidence into a strip sample according to the mapping relation; dividing a two-dimensional strip repair domain into a sound area and a defect mask based on a strip pattern, extracting repeated pattern units from the sound area to construct a pattern unit library, detecting structural event points representing symmetry references, bifurcation and closure or convergence of adjacent relations from the pattern unit library, and sequencing according to the main trend of the pattern to generate an event sequence and pitch; Under the constraint of an event sequence, a pitch and a pattern unit library, performing pattern unit assembly and complementation on the defect mask to obtain candidate repair patterns, and calculating an event closed loop error according to the closed consistency and pitch consistency deviation of the event sequence; When the event closed-loop error exceeds a preset error threshold, the event closed-loop error is thrown back to a substrate grid positioning error section according to the mapping relation, the central line or transverse width distribution is adjusted along the error section so as to update the strip-shaped geometric parameters and regenerate candidate repair patterns until the stopping condition is met; And inputting the candidate repair pattern into a generative model, encoding the event sequence and the pitch into conditional constraint input, and outputting the plane repair pattern with enhanced details.
  2. 2. The method according to claim 1, characterized in that the determination of a strip-like body with a major axis of the strip-like main trend and a limited lateral width on the substrate grid comprises: Generating a consistency weight graph on a substrate grid based on the grid pattern evidence and the multi-view consistency measurement, and screening a high-confidence region by combining a preset confidence criterion; extracting local gradient vectors of grid pattern evidence in a substrate grid, constructing a pattern direction field, counting pattern direction field distribution in a high confidence region to determine a pattern main trend, and intercepting local grid fragments along the pattern main trend; Performing local parameterization expansion based on a tangent plane of the local grid segment, resampling the grid pattern evidence into a local stripe pattern, detecting geometric event points in the local stripe pattern, and writing back to the substrate grid according to a local parameterization expansion relation to form a candidate point set; Constructing an event adjacency graph by using a candidate point set, calculating side weights, taking an event main chain with the largest side weight accumulation and the highest consistency between the main direction and the pattern main trend as a central line, and calculating the tangential direction and the normal direction of the central line; Generating a plurality of groups of boundary point pairs on the normal direction of each sampling point of the central line, forming candidate width sets, and resampling each candidate width set into a local stripe pattern according to a local parameterization unfolding relationship; Calculating local pitches on each local stripe sample and obtaining pitch stability, wherein the pitch stability is obtained by combining the dispersion of adjacent local pitches and the consistency of rhythm peak values; and selecting a boundary point pair with the maximum pitch stability as the boundary of each sampling point, splicing the boundary point pair into transverse width distribution, and constructing a stripe-shaped ribbon body on the substrate grid based on the central line and the transverse width distribution.
  3. 3. The method of claim 1, wherein the steps of extracting repeated pattern units in the intact region to construct a pattern unit library, and detecting structural event points representing symmetry references, adjacency furcation and closure or convergence therefrom, and generating event sequences and pitches according to pattern main trend ordering comprise: Mapping the main trend of the stripe sample to a two-dimensional stripe repair domain by using a mapping relation, carrying out sliding autocorrelation calculation on the stripe sample along the mapped main trend of the stripe sample in a perfect region, and obtaining a relevant peak value position to generate a rhythm candidate point column; Cutting repeated segments in a stripe sample according to the rhythmic candidate point column, extracting segment boundary curves, writing the boundary curves into a stripe sample cell library after normalizing the boundary curves, and recording cell contours; Mirror image matching and registration are carried out on the unit contours to obtain mirror image registration axes and registration errors thereof, and the mirror image registration axes with the registration errors lower than a preset registration threshold value are recorded as symmetrical references; constructing a unit adjacency graph according to the relative pose of the unit outline in the stripe sample and the adjacency relation, detecting symmetrical reference switching, adjacency relation bifurcation or mergence and boundary curve closing or convergence positions on the unit adjacency graph, and marking the positions as structure event points; And ordering the structural event points according to the main trend of the pattern to obtain an event sequence, and executing steady statistics on the distance between the event points of the adjacent similar structures to generate a pitch.
  4. 4. The method of claim 1, wherein the event closed loop error is projected back to the base grid positioning error segment according to the mapping relationship, and the centerline or lateral width distribution is adjusted along the error segment to update the ribbon geometry, comprising: The event closed-loop errors are projected back to the substrate grid according to the mapping relation to generate an error weight field, and connected domain analysis is performed in the central line neighborhood to extract an error section; Performing lateral displacement search along the normal direction of the central line in the error section, and taking the displacement amount of the grid pattern evidence and multi-view consistency measurement weighted accumulation value meeting a preset accumulation criterion to update the central line; And (3) performing bilateral stepping correction on boundary point pairs of the transverse width distribution in the error section, updating the transverse width distribution by using correction amounts with consistent gradient signs of grid pattern evidence at the boundary points and width change meeting preset smoothness constraint, and writing the updated central line and transverse width distribution into the strip geometric parameters.
  5. 5. The method according to claim 2, wherein before performing pattern unit assembly and repair on the defect mask to obtain candidate repair patterns, further comprising constructing a structure phase coordinate and a strip repair weight field, specifically comprising: in a two-dimensional strip repair domain, building a structure phase coordinate along the main trend of the pattern by utilizing pitch, and projecting a structure event point to the structure phase coordinate to form a phase event index table; In the intact area, carrying out phase alignment sampling on the stripe sample according to the structure phase coordinates, extracting a phase texture section, and carrying out robust statistics based on the phase texture section to obtain a phase reference template; and projecting the consistency weight map to a two-dimensional strip restoration domain according to the mapping relation to obtain a strip consistency weight map, and fusing the strip consistency weight map with the normalized gradient amplitude of the phase reference template to generate a strip restoration weight field.
  6. 6. The method according to claim 5, wherein performing pattern unit assembly and repair on the defect mask to obtain candidate repair patterns, specifically comprises: dividing phase strips according to structure phase coordinates in a defect mask, generating a defect block set, and marking structure neighborhood identifiers for each defect block according to a phase event index table; searching candidate pattern units which are consistent with the structure neighborhood identifier and are matched with the phase reference template in the pattern unit library, and performing phase alignment and transverse scale normalization on the candidate pattern units along the main trend of the pattern to generate an alignment unit; And calculating boundary gradient consistency scores and pitch consistency scores of the alignment units based on the strip repair weight fields, selecting the alignment unit with the largest weight of the boundary gradient consistency scores and the pitch consistency scores, writing the alignment unit into the corresponding defect block, and splicing to obtain the candidate repair pattern.
  7. 7. The method of claim 6, further comprising performing a boundary fusion update on the candidate repair pattern prior to inputting the candidate repair pattern into the generative model, comprising: equidistant expanding a preset bandwidth in a two-dimensional strip repair domain along the boundary of the defect mask to obtain a transition zone, calculating a distance field from the transition zone to the boundary of the defect mask, and normalizing to obtain a distance weight map; Backtracking to a defective mask boundary point along the gradient direction of a distance field in a transition zone, reversely translating a distance equal to the distance field value of the point to position an inward movement sampling point, extracting pixel values of candidate repair pattern at the inward movement sampling point, writing the pixel values into a corresponding position of the transition zone, and constructing a transition zone candidate pattern; Calculating a pixel difference map of a transition zone candidate pattern and a stripe pattern in a transition zone, and fusing a normalization result of the pixel difference map, a distance weight map and a stripe consistency weight map to generate a transition zone fusion weight field; The transition zone candidate pattern and the stripe pattern are subjected to pixel-by-pixel weighted fusion by a transition zone fusion weight field to generate a transition zone fusion pattern; and splicing and writing the fusion pattern of the transition zone into the corresponding position of the transition zone, and reserving the candidate repair pattern in the defect mask to obtain the spliced and updated candidate repair pattern.
  8. 8. The cultural relic pattern shape digital restoration system based on the generative model is characterized by being used for realizing the cultural relic pattern shape digital restoration method based on the generative model as set forth in any one of claims 1-7, and comprising the following steps: the three-dimensional reconstruction fusion module is used for acquiring a surface image set of the target cultural relics, obtaining a substrate grid through three-dimensional reconstruction and standardization processing, and fusing registration projections of the surface image set in the substrate grid to form a grid pattern evidence and a multi-view consistency measure; The strip parameter extraction module is used for determining a strip-shaped strip body which takes the main trend of the strip sample as a long axis and is limited in transverse width on the basis of the grid-shaped evidence and multi-view consistency measurement, and extracting the strip-shaped geometric parameters formed by the distribution of the central line and the transverse width of the strip-shaped strip body; the two-dimensional unfolding resampling module is used for parameterizing and unfolding the strip-shaped body of the strip sample into a two-dimensional strip repairing domain by utilizing the strip-shaped geometric parameters, recording the mapping relation between the strip-shaped body of the strip sample and the substrate grid, and resampling the grid strip-shaped evidence into a strip sample according to the mapping relation; The event sequence generation module is used for dividing the two-dimensional strip repair domain into a sound area and a defect mask based on a strip pattern, extracting repeated pattern units in the sound area to construct a pattern unit library, detecting structural event points representing symmetry reference, adjacent relation bifurcation and closure or convergence from the pattern unit library, and generating an event sequence and pitch according to pattern main trend sequencing; The repair error calculation module is used for performing pattern unit assembly and complementation on the defect mask under the constraint of the event sequence, the pitch and the pattern unit library to obtain candidate repair patterns, and calculating event closed loop errors according to the closed consistency and the pitch consistency deviation of the event sequence; The iterative optimization adjustment module is used for projecting the event closed-loop error back to the substrate grid positioning error section according to the mapping relation when the event closed-loop error exceeds a preset error threshold value, adjusting the center line or transverse width distribution along the error section to update the strip-shaped geometric parameters and regenerating candidate repair pattern until the stop condition is met; And the model enhancement output module is used for inputting the candidate repair pattern into the generated model, encoding the event sequence and the pitch as conditional constraint input, and outputting the plane repair pattern with enhanced details.
  9. 9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the method of any of claims 1-7 when executing the computer program.
  10. 10. A computer readable storage medium storing a computer program, which when executed by a processor performs the method according to any one of claims 1-7.

Description

Cultural relic pattern form digital restoration method and system based on generated model Technical Field The application relates to the technical field of digital cultural relic restoration, in particular to a cultural relic pattern shape digital restoration method and system based on a generated model. Background The surface pattern of the cultural relic bears aesthetic and technological information of a specific age, and is oriented to requirements of protection record, digital display, resource utilization of the pattern and the like, and the collected surface pattern of the cultural relic is usually required to be subjected to morphological digitization and repair and reconstruction of a defect area. The existing pattern digital restoration of cultural relics is generally based on image acquisition and three-dimensional digitization, pattern information is obtained by carrying out multi-view imaging or scanning on the surface of the cultural relics, pattern extraction, alignment and fusion are carried out on the surface of a two-dimensional image or a three-dimensional model, on the basis, the missing areas caused by damage, abrasion or shielding are complemented by adopting the modes of patch filling, repeating unit copying, image repair or learning generation and the like based on similar textures, and the restoration results are adjusted by combining with manual correction so as to obtain continuous and complete pattern expression. With the increasing demands of application on pattern precision, consistency and reusability, the problems exposed by the above processes in actual processing of complex curved surfaces, multi-view information and structured pattern rules are also increasingly highlighted. In the existing digital restoration of the cultural relic pattern, the common difficulty is that the pattern is attached to a complex curved surface and is influenced by shielding, reflection and visual angle change, multi-view pattern information is difficult to align stably and fuse credibly, key geometric relations such as pattern main trend and bandwidth are easy to drift, meanwhile, the pattern has structural rules such as repetition, symmetry and closing/converging, the conventional complementation is more dependent on local texture similarity or direct generation, and a verifiable structural constraint and a locatable error feedback mechanism are lacked, so that dislocation is easy to occur on pitch, phase and closing relation, a result that local looks reasonable but integral structural distortion appears is difficult to directly serve as a high-consistency pattern material for multiplexing scenes such as clothing design and decoration, and the like, so that the digital restoration method and the system for the cultural relic pattern form based on a generated model are provided to solve the problem. Disclosure of Invention In order to solve the technical problems, the method and the system for digitally repairing the pattern form of the cultural relics based on the generated model are provided. In order to achieve the above object, the technical scheme of the present invention is as follows: in a first aspect, the present application provides a method for digitally repairing a pattern of a cultural relic based on a generative model, the method comprising: obtaining a surface image set of a target cultural relic, obtaining a substrate grid through three-dimensional reconstruction and standardization processing, and fusing registration projections of the surface image set in the substrate grid to form a grid pattern evidence and a multi-view consistency measure; Based on the lattice pattern evidence and multi-view consistency measurement, determining a pattern strip body which takes the main trend of the pattern as a long axis and has limited transverse width on a substrate grid, and extracting the distribution of the central line and the transverse width of the strip body to form strip geometric parameters; parameterizing and expanding the strip-shaped body of the strip sample into a two-dimensional strip repair domain by utilizing strip geometric parameters, recording the mapping relation between the strip-shaped body of the strip sample and a substrate grid, and resampling the grid strip-shaped evidence into a strip sample according to the mapping relation; dividing a two-dimensional strip repair domain into a sound area and a defect mask based on a strip pattern, extracting repeated pattern units from the sound area to construct a pattern unit library, detecting structural event points representing symmetry references, bifurcation and closure or convergence of adjacent relations from the pattern unit library, and sequencing according to the main trend of the pattern to generate an event sequence and pitch; Under the constraint of an event sequence, a pitch and a pattern unit library, performing pattern unit assembly and complementation on the defect mask to obtain candidate repair patterns, a