基于CNN与HOG特征的决策级融合图像识别
首发时间:2016-06-24
摘要:图像识别在计算机视觉领域广泛应用。深度学习热潮兴起之后,卷积神经网络(Convolutional Neural Network, CNN)作为其中的一种典型模型,在图像识别上取得了良好的应用效果。针对卷积神经网络对不同样本需要调整模型以及模型调整比较繁琐的问题,本文提出一种基于CNN与HOG特征的决策级融合算法。该模型从原始图像中分别提取CNN特征和HOG特征,使用SVM分类器产生预测,通过设定权值进行决策级融合,利用多源特征的优势。本文在不同数据库上进行实验,结果证明该方法使用基本的CNN模型即能有效提高图像识别率,避免了繁琐地调整CNN模型。
关键词: CNN特征 HOG特征 SVM 决策级融合 图像识别
For information in English, please click here
Image Recognition via Decision Fusion of CNN and HOG Features
Abstract:Image recognition is widely used in computer vision. After deep learning has rised, the Convolution Neural Network (CNN) has obtained good effects in image recognition as one of typical models. Due to a problem that the Convolutional Neural Network needs to adjust the model with different samples and the adjustment of the model is complicated, this paper proposes a decision fusion algorithm based on CNN and HOG features. This model extracts CNN features and HOG features respectively from original images, and uses SVM classifier to generate predictions, then fuses the decision with weighted values, making use of multi-source features. Experiments have been conducted on the different databases in this paper, the results have proved that this fusion of using the basic CNN model has effectively improved the images recognition rates and avoided to adjust the CNN model repeatedly.
Keywords: CNN features HOG features SVM Decision fusion Image recognition
论文图表:
引用
No.4698328862381146****
同行评议
共计0人参与
勘误表
基于CNN与HOG特征的决策级融合图像识别
评论
全部评论0/1000