摘要
为提升织物图像检索的准确性,采用改进的广义产品量化(generalized product quantization, GPQ)半监督神经网络实现弱纹理织物图像的检索。通过CLAHE方法增强织物图像纹理,加强底层纹理特征,以降低深度学习特征过拟合的概率。利用GPQ框架中产品量化、基于余弦相似性分类器和子空间极小最大熵损失计算,对提取的特征向量进行归一化,寻找最相似织物图像。实验中采用的织物数据集包含了12类不同纹理形式的织物试样,共计1 800幅图像。分别对比了基于颜色直方图的词袋模型、尺寸不变特征变换模型、最近邻和优化产品量化算法。结果表明:改进的GPQ半监督神经网络方法的平均精度均值达到89.47%,检索性能最优。该方法能批量、低成本检索出相似织物图像,提高织物图像检索的准确性。
To enhance the accuracy of fabric image retrieval,an improved semi-supervised neural network based on generalized product quantization(GPQ)is employed for efficient retrieval of weakly textured fabric images.The texture of the fabric image dataset is enhanced using the clip limited adaptive histogram equalization(CLAHE)method,thereby reinforcing the underlying texture features to reduce the probability of overfitting in deep learning features.Within the GPQ framework,which includes product quantization,a cosine similarity-based classifier,and computation of the subspace minimum-maximum entropy loss,the extracted feature vectors were normalized to identify the most similar fabric images.The experimental fabric dataset comprises 1800 images,representing 12 different texture styles of fabric samples.Comparative analyses were conducted with respect to the color histogram-based bag-of-words model,size-invariant feature transform model,nearest neighbors,and optimized product quantization algorithm.The results indicate that the improved GPQ semi-supervised neural network achieves a m AP value of 89.47%,demonstrating optimal retrieval performance.This method facilitates batch retrieval of similar fabric images at a low cost,and improves the accuracy of fabric image retrieval.
作者
熊枫情
罗芊芊
蒋汶秦
吕萧羽
徐平华
XIONG Fengqing;LUO Qianqian;JIANG Wenqin;LYU Xiaoyu;XU Pinghua(School of Fashion Design&Engineering,Zhejiang Sci-Tech University,Hangzhou 310018,China;Digital Intelligence Style and Creative Design Research Center,Key Research Center of Philosophy and Social Sciences,Zhejiang Province,Zhejiang Sci-Tech University,Hangzhou 310018,China;Key Laboratory of Silk Culture Heritage and Products Design Digital Technology,Ministry of Culture and Tourism,P.R.China,Zhejiang Sci-Tech University,Hangzhou 310018,China)
出处
《纺织高校基础科学学报》
CAS
2024年第1期42-48,共7页
Basic Sciences Journal of Textile Universities
基金
浙江省哲学社会科学规划交叉学科课题(24LMJX09YB)
浙江省高校重大人文社科攻关计划项目(2023QN092)
中国纺织工业联合会科技指导性项目(2023029)
国家级大学生创新创业训练计划项目(202310338047,202210338019)。
关键词
织物图像
神经网络
图像增强
深度学习
图像检索
fabric image
neural networks
image enhancement
deep learning
image retrieval
