摘要
图像变化检测是区分同一场景不同时间所获取图像的变化区域与未变化区域的重要方法。然而现有的基于深度学习的方法往往需要大量的数据样本进行训练和调参,使得模型的训练过程非常耗时且对内存的要求很高。另一方面,现有的大多数变化检测模型中,输入层的图像必须归一化到同样的大小,但是调整图像尺寸又会造成图像小目标信息缺失等问题,导致检测结果的精确度降低。针对上述问题,提出了一种集成多特征信息的街景图像变化检测方法。首先,将参考图像和检测图像分块,提取图像块的颜色特征、纹理特征和形状特征;然后,计算对应图像块之间上述三种特征的欧氏距离,同时计算图像块之间的感知哈希序列和灰度方差;最后,使用集成模型将几种分类器组合为变化检测的分类模型,将上述几种特征作为模型的输入特征,得到图像的变化检测结果。该方法对图像的输入尺寸没有固定的要求,并且有效降低了模型的复杂性。在CDnet2014数据集上的实验结果表明,只需要少量的训练样本,就可以得到较为鲁棒的变化检测结果。
Image change detection is an important method to distinguish the changed and unchanged areas of images acquired at different times in the same scene.However,existing deep learning-based methods often require a large number of data samples for training and parameter tuning,which makes the model training process quite time-consuming and requires high memory.On the other hand,in most existing change detection models,the images of input layer must be normalized to the same size,but adjusting the image size will cause problems such as loss of small target information,which will reduce the precision of the detection results.To solve the above problems,a change detection method for street view images with integrated multi-feature information is proposed.Firstly,chunking the reference image and the detection image to extract the color features,texture features and shape features of the image patches.Then,the Euclidean distances of the above three features between corresponding image patches is calculated,along with the perceptual hash sequences and gray variances between image patches.Finally,several classifiers are combined into a classification model for change detection using the ensemble model,and the above-mentioned features are used as the input features of the model to obtain the change detection results of images.The proposed method has no fixed requirements on the input size of images,and effectively reduces the complexity of the model.The experimental results on CDnet2014 dataset show that only a small number of training samples,whose relatively robust change detection results are obtained.
作者
薛颂东
张轩冉
王斌
李靖
曹旺旺
乔钢柱
XUE Song-dong;ZHANG Xuan-ran;WANG Bin;LI Jing;CAO Wang-wang;QIAO Gang-zhu(School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China;Department of Data Science and Technology,North University of China,Taiyuan 030051,China)
出处
《计算机技术与发展》
2023年第6期69-74,共6页
Computer Technology and Development
基金
教育部产学合作协同育人项目(202102076011)
山西省高等学校科技创新项目(2021L322)
山西省基础研究计划(自由探索类)项目(20210302124165)。
关键词
图像变化检测
特征提取
集成模型
分类器
麻雀搜索算法
image change detection
feature extraction
ensemble model
classifier
sparrow search algorithm
