基于卷积神经网络和语义分割的苗族刺绣风格数字模拟  

Digital Simulation of Miao Embroidery Style Based on Convolution Neural Network and Semantic Segmentation

作　　者：谭永前[1] 曾凡菊[1] TAN Yong-qian;ZENG Fan-ju(Kaili University,Kaili,Guizhou,556011,China)

机构地区：[1]凯里学院,贵州凯里556011

出　　处：《凯里学院学报》2023年第3期64-73,共10页Journal of Kaili University

基　　金：贵州省2022年度基础研究计划(自然科学)项目(黔科合基础-ZK[2022]一般526)。

摘　　要：传统神经网络算法在对苗族刺绣图像风格模拟过程中,存在模拟图像边缘轮廓模糊,生成的风格化模拟图像纹理细节缺乏立体感、线条扭曲以及模拟效果图的前景和背景层次不清等问题.本文提出了一种基于卷积神经网络和语义分割的苗族刺绣风格数字模拟算法:首先对内容图像和风格图像进行预处理,检测内容图像的边缘轮廓并增强,利用FCN-CRF图像语义分割算法对内容图像和苗族刺绣风格图像进行语义分割并进行标注,同时获取相应的二值化掩膜.然后把内容图像和苗族刺绣风格图像及相应掩膜输入到CNN中进行网络计算,完成苗族刺绣风格的转换,获得苗族刺绣风格初始模拟图像.最后把边缘轮廓增强图像和绣布与初始模拟图像进行融合,达到更好的效果.实验表明,改进后的方法能提高风格模拟效果图中前景图和背景图的层次感,风格模拟效果图的边缘信息清晰度得到改善,取得了比传统算法更好的风格模拟视觉效果.In view of the problems of the traditional neural network algorithm in the style simulation of Miao embroidery image,such as the fuzzy edge contour of the simulated image,the lack of stereoscop⁃ic texture details of the generated stylized simulated images,the distortion of lines,and the unclear lev⁃el of the foreground and background of the simulated effect image,a digital simulation algorithm of Miao embroidery style based on convolutional neural network and semantic segmentation is proposed.First,the content image and the style image are preprocessed;the edge contour of the content image is detected and enhanced;the content image and the Miao embroidery style image are semantically seg⁃mented and labeled using the FCN-CRF image semantic segmentation algorithm;and the correspond⁃ing binary mask is obtained at the same time;then input the content image,Miao embroidery style im⁃age and corresponding mask into CNN for network calculation,complete the transformation of Miao em⁃broidery style,and obtain the initial simulation image of Miao embroidery style;Finally,the edge con⁃tour enhancement image and embroidered fabric are fused with the initial simulation image to improve the effect of blurred edge information and unclear foreground and background levels in the style simula⁃tion image.The experiment shows that the improved method can achieve a better visual effect of style simulation by improving the hierarchical sense of foreground and background images in style simulation renderings,improving the edge information clarity of style simulation renderings,and improving the edge information clarity of style simulation renderings.

关 键 词：卷积神经网络 边缘信息 语义分割 风格化处理 苗族刺绣 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]