织锦文物纹样超分修复中双三次插值的非局部均值拓展OA

Silk and brocade textiles are regarded as among the most challenging categories of cultural relics to preserve.Owing to corrosion,mildew,insect damage,fading,and the limitations of historical imaging conditions,a large number of brocade cultural relics have been preserved only in the form of low-resolution images,by which the extraction,restoration,and subsequent vectorized modeling of pattern information are seriously hindered.In recent years,digital protection methods for textile relic patterns have been mainly developed from two aspects,namely image restoration and pattern modeling.Specifically,super-resolution reconstruction is considered an important preprocessing step for vectorized modeling,because the quality of the original pattern image directly affects the fidelity of subsequent digital restoration.Existing super-resolution methods can generally be classified into interpolation-based,reconstruction-based,and learning-based methods.Although remarkable performance has been achieved by deep learning-based approaches in natural image super-resolution,several challenges are still encountered when they are applied to cultural heritage images,including the scarcity of domain-specific training samples,the diversity of degradation conditions,and the high demand for authenticity and traceability.In particular,synthesized details inconsistent with the original brocade patterns may be generated by deep models,by which later restoration and modeling tasks may be adversely affected.Therefore,a super-resolution method that does not rely on training data and can better preserve the structural authenticity of brocade cultural relic patterns is still urgently needed. To address the above problems,a non-local means(NLM)-augmented bicubic interpolation method,termed NLM-Bicubic,was proposed in this study for the super-resolution preprocessing of brocade cultural relic patterns.In the proposed method,NLM-based weight optimization was introduced into the conventional bicubic interpolation framework,so that reconstructed pixel values were determined not only by the local 4×4 neighborhood required in bicubic interpolation,but also by structurally similar image patches distributed over a larger search region.In this way,the repetitive motifs and periodic structures widely present in brocade pattern images could be effectively utilized.In the experiments,twelve representative brocade cultural relic images were selected as test samples,with different levels of high-frequency information,structural complexity,self-similarity,and noise.Parameter sensitivity analysis,ablation experiments,comparative experiments,and subjective-objective evaluations were then carried out.It was shown by the results that,at a magnification factor of×2,an average PSNR of 29.12 dB and an average SSIM of 0.866 9 were achieved by the proposed method.In the ablation experiment,PSNR was increased from 21.29 dB to 24.73 dB and SSIM was improved from 0.597 2 to 0.690 7,as compared with the Bicubic baseline,by which the important role of the NLM module in improving reconstruction quality was confirmed.When compared with several representative super-resolution methods,competitive performance was exhibited by NLM-Bicubic for pattern samples with pronounced repetitive structures and high self-similarity,and better structural consistency was observed in some cases.At a magnification factor of×4,relatively high SSIM values were still maintained across most pattern types,showing that strong reconstruction capability was retained by the proposed method for both local repetitive units and overall pattern structures.In addition,the highest comprehensive score,0.967 3,was obtained in the entropy-weighted TOPSIS evaluation,indicating that a favorable balance among pattern fidelity,edge clarity,and detail continuity was achieved by the proposed method. A reliable and interpretable solution was thus provided by the proposed NLM-Bicubic method for the digital restoration of brocade cultural relic patterns.Without relying on large-scale training data,pattern authenticity and structural consistency can be effectively preserved,while the reconstruction of repetitive textures can also be enhanced.It is expected that a practical preprocessing approach will be provided by this study for the subsequent vectorized modeling of brocade patterns.Moreover,useful technical reference may also be offered for the digital preservation and restoration of other traditional woven textile patterns.