摘要
为了检索图像中不同位置和不同大小的感兴趣目标,提出一种基于多尺度深度卷积特征的图像检索方法.首先利用卷积神经网络构造一个深度学习框架,利用随机梯度下降和后向传播算法训练深度学习模型;其次利用训练得到的模型提取图像在不同尺度下的卷积特征,对不同尺度下的卷积特征进行PCA降维,研究降维后的检索性能;最后为了提高深度特征对图像的刻画能力,对不同尺度下降维后的卷积特征进行特征融合.大量的实验表明本文所提算法对图像检索是有效的.
To detect the object of interests of different location and scales,it puts forward a image retrieval method based on muti-scale deep convolutional features. Firstly,it trains a deep learning model using convolutional neural network and extract convolutional features of image patches at different scales. And then,it reduces the dimensionality of the extracted features by PCA and study the effect of the reduced features of different scales on image retrieval. Finally,to improve the description of high features,it fuses the processed features by combining them. A large number of experiments show that our method is effective for image retrieval.
出处
《福建师范大学学报（自然科学版）》
CAS
CSCD
北大核心
2016年第5期17-23,共7页
Journal of Fujian Normal University：Natural Science Edition
关键词
卷积神经网络
基于内容的图像检索
特征提取
深度学习
多尺度
特征融合
convolutional neural network
content-based image retrieval
feature extraction
deep learning
muti-scale
feature fusion
