基于多线索特征融合的图像分类方法

针对图像本身存在噪声和冗余信息而导致分类准确率不高的问题进行了研究,提出一种基于多线索特征融合图像分类算法。通过改进全局显著性和稀有性度量方法得到显著图像;分别在原图像、压缩图像和显著图像上提取方向梯度直方图(Histogram of Oriented Gradient,HOG)特征;将提取到的特征向量融合;采用基于欧氏距离的二叉树支持向量机(Distance Binary Tree SVM,DBT-SVM)进行图像分类。利用Caltech101 和花卉图像数据集进行实验测试,结果表明提出的算法能够有效地提高图像分类的准确率。

Multi-Cue Feature Fusion Based Image Classification

Due to the noise and redundant information in the image, the result of classification is not accurate. This paper proposes a feature fusion classification algorithm that based on multiple clues. Firstly, it gets a significant image by the improved global saliency and rarity metrics. Next, it extracts the Histogram of Oriented Gradient(HOG)features on the original image, including the compressed image and the salient image. And then it merges the extracted feature vectors. At last, it uses Distance Binary Tree Support Vector Machines(DBT-SVM)based on Euclidean distance for image classification. Experiments with Caltech101 and flower image datasets show that the proposed algorithm can effectively improve the accuracy of image classification.