基于图像处理的男西装前身褶皱弊病识别  

Recognition of Front Bodice Wrinkle Defects in Men’s Suits Based on Image Processing

作　　者：张启泽[1] 杜聪 庹武[1] 郝潇潇 ZHANG Qize;DU Cong;TUO Wu;HAO Xiaoxiao(College of Fashion,Zhongyuan University of Technology,Zhengzhou 451191,China;School of Intelligent Enginee-ring,Huanghe Jiaotong University,Jiaozuo 454950,China;College of Fashion Design,Shanghai Institute of Visual Arts,Shanghai 201620,China)

机构地区：[1]中原工学院服装学院,河南郑州451191 [2]黄河交通学院智能工程学院,河南焦作454950 [3]上海视觉艺术学院时尚设计学院,上海201620

出　　处：《服装学报》2024年第6期484-491,共8页Journal of Clothing Research

基　　金：河南省高等学校重点科研项目(25A540002,23A540007)。

摘　　要：前身不良褶皱是西装常见的弊病之一,在西装定制过程中对专业人员依赖较强、耗时且易受主观影响。以男西装为例,构建一种前身褶皱弊病自动识别方法。从企业收集男西装弊病图,借助分割标注工具EISeg提取目标图像,采用双三次插值统一图像分辨率,规范特征参数;对图像进行灰度化、伽马变换及阈值分割,简化运算数据,增强图像信息,突出褶皱走势;根据褶皱局部灰度曲线图提取褶皱的宽度、深度和密度3种参数指标,按照阈值分割图提取褶皱方向和褶皱部位2种参数指标;在BP神经网络中加入粒子群算法改进网络模型,输出弊病类别。研究表明:与传统BP神经网络模型相比,优化网络模型的测试集准确率上升了8.3%,该方法可准确实现男西装前身褶皱弊病的自动识别,并为行业带来新的技术手段和方法。The poor front bodice wrinkles is one of the common drawbacks of suits, which is highly dependent on professionals, time-consuming and susceptible to subjective influence in the suit customization process. Taking men’s suits as an example, this paper constructs an automatic recognition method for the defects of predecessor pleats. The male suit malpractice map was collected from the enterprise, and the target image was extracted by using the segmentation and annotation tool EISeg. The bicubic interpolation was used to unify the image resolution and standardize the feature parameters. The image was grayed, gamma transform and threshold segmentation to simplify the operation data, enhance the image information, and highlight the fold trend. According to the local gray curve of the fold, the width, depth and density of the fold are extracted. According to the threshold segmentation map, the fold direction and the fold position are extracted. The particle swarm optimization algorithm is added to the BP neural network to improve the network model and output the malpractice category. The results show that the test set accuracy of the optimized network model is 8.3% higher than that of the traditional BP neural network model. This method can accurately realize the automatic identification of the defects of the front of the men’s suit, and bring new technical means and methods to the industry.

关 键 词：西装前身弊病 图像处理 参数提取 粒子群算法 BP神经网络 

分 类 号：TS941.26[轻工技术与工程—服装设计与工程]