基于卷积神经网络的刺绣风格数字合成

针对刺绣风格数字化模拟方法立体感不强、缺少线条方向等问题，提出了一种基于深度学习和卷积神经网络的算法，将刺绣艺术风格传输到目标图像。利用图像语义分割网络及风格迁移网络，分别对目标内容图像与刺绣艺术风格图像进行目标提取和风格迁移。首先，输入目标内容图像与刺绣艺术风格图像，采用基于条件随机场的图像语义分割,将目标内容图与刺绣艺术风格图的前景与背景分离，并进行二值化处理，形成掩模图像；其次，将目标内容图与刺绣艺术风格图的RGB颜色空间转换为YIQ；最后，参照掩模图像使用VGG19网络模型提取目标内容图的内容特征及刺绣艺术风格图的风格纹理特征进行目标区域内的风格迁移，从而对刺绣艺术进行数字化模拟。该算法能模拟出具有刺绣艺术效果的结果图像，能更好地模拟真实刺绣艺术的线条方向，突出了线条的立体感。通过使用语义分割与风格迁移相结合的方法，有效模拟了色彩艳丽、立体感强的刺绣艺术风格图像，是对非真实感绘制的有效补充，为刺绣数字化保护与非物质文化传承奠定了基础。

Synthesis of embroidery based on convolutional neural network

To remedy the deficiency of the current digital embroidery algorithm in arousing the original linear and stereo perception, this paper proposes an artificial system based on deep-learning and convolutional neural network to synthetize the style of embroidery. First, we input the content image and the embroidery style image, perform image semantic segmentation which is based on conditional random field to separate the foreground and the background of both images and construct masks by image binarization. Then, we convert the color spaces of both input images from RGB into YIQ, extract the features of embroidery by VGG19 and transfer the content image into embroidery style in the foreground by using mask, meanwhile emphasizing the gorgeous colors and the stereoscopic textures of the embroidery. Experimental results show that the proposed method can enhance images with embroidery style effectively. It lays a foundation for digital inheritance of the traditional embroidery.