Improved DenseNet-DCGAN for Enhanced Digital Restoration of Embroidery Cultural Heritage

Abstract

At present, embroidery image restoration technology still has deficiencies in terms of color uniformity and detail restoration. To address these issues, the study improves the densely connected convolutional network and the deep convolutional generative adversarial network through spatial pyramid pooling, and proposes a novel method for embroidery image classification and restoration. The experimental results showed that the research method largely restored the details and colors of the original image and effectively addressed the uneven color issue. The average prediction accuracy, recall rate, and specificity of the image classification model on Suzhou embroidery, Hunan embroidery, Guangdong embroidery, and Shu embroidery reached 96.3%, 98.5%, and 99.4%, respectively. The structural similarity index of the image restoration model has reached 0.99. The restored image was almost indistinguishable to the naked eye in terms of details, texture, and color. The research method has significant advantages in classifying embroidery images and high-quality restoration tasks, and can provide reliable technical support for the digital protection and intelligent restoration of traditional embroidery cultural relics.