Computer Vision-Driven AI Extraction and Redesign of Carved Lacquer Relief Patterns

Abstract

This study proposed a closed-loop knowledge management framework that integrates computer vision with craft rules to extract and redesign traditional carved lacquer relief patterns. A dataset of 120 complete images and 480 local segments was constructed across four motif types. Seven threshold metrics were established to evaluate pattern quality and design applicability. Modular experiments showed that adding boundary enhancement, style encoding, manual verification, and process rewriting progressively improved contour closure (from 0.78 to 0.91), symmetry stability (0.73 to 0.89), and redesign acceptance (74 to 87), while reducing misclassification (0.24 to 0.11). Among three redesign strategies, the balanced translation achieved the highest overall score (84) by preserving cultural authenticity while enabling moderate innovation. The findings demonstrated that a process-rewriting mechanism can effectively externalize tacit craft knowledge and prevent error propagation, offering a replicable knowledge management solution for the digital preservation of intangible cultural heritage.