摘要
为提高图像增强算法的图像识别有效性,提出了基于图像增强的低光照图像识别算法。首先,采用直方图均衡化的图像增强算法对开源低光照图像数据集(ExDark)进行增强处理;然后,设计卷积神经网络进行图像识别训练,通过多重卷积—池化操作,实现图像特征提取;最后,将识别结果与其他增强方法结果进行对比实验。结果表明,与其他传统方法相比,在低光照图像的处理中采用直方图均衡化的图像增强法可获得更高的图像信息熵与图像对比度,图像识别准确率提升了14.4%,对低光照条件下的图像识别具有参考价值。
A low-illuminated image recognition algorithm based on image enhancement is proposed to improve image recognition.Firstly,image enhancement algorithm of histogram equalization is used to enhance the open-source ExDark low-light image data set.Then,a convolutional neural network is designed for im age recognition training,from which image feature is extracted through multiple convolution and pooling operations.Finally,the results are compared with those obtained from the other enhancement methods.It is found that the image enhancement method using histogram equalization in the processing of low-light images can obtain higher image information entropy and image contrast,and image recognition accuracy increases by 14.4 %,which provides a reference for further study on image recognition under low illumination.
作者
王海鹏
WANG Hai-peng(School of Intelligent Manufacturing,Jiangsu College of Engineering and Technology,Nantong 226007,China)
出处
《南通职业大学学报》
2023年第1期66-69,共4页
Journal of Nantong Vocational University
关键词
直方图均衡化
低光照图像
神经网络
图像识别
histogram equalization
low-illuminated image
neural network
image recognition
