基于SVM和D-S证据理论场景图像分类

Scene Image Classification Based on SVM and D-S Evidence Theory

Gist特征和PHOG特征分别作为描述场景图像全局性质和局部性质的特征,两者各自有不足之处。若能吸取两者优势互补,则场景图像分类准确率将得到提升。论文提出了一种基于D-S证据理论的融合Gist特征和PHOG特征的场景图像描述方法。该方法首先提取场景图像的Gist特征和PHOG特征,然后基于D-S证据理论得到融合的特征向量。使用支持向量机作为分类器,在OT场景图像库下,分别建立单一的Gist特征、单一的PHOG特征、传统串联融合特征以及证据理论融合特征的分类模型,采用正确率和混淆矩阵作为评价指标,分别进行四组实验。实验结果表明,论文提出的方法有效提高了场景图像分类的准确度。

As respectively describing global and local property of images,Gist and PHOG have their own disadvantage.If the advantage of the two can be combined together,then the classification accuracy will increase.A new method based on D-S evidence theory which combine Gist and PHOG is proposed in this paper.First,Gist and PHOG characteristic vector of images is gotten,then combined characteristic vector can be got based on theory D-S evidence theory.Using the SVM as classifier,four group of experiments,including Gist、PHOG、traditional serial combined and combined based on D-S evidence theory,would be conducted with Oliva & Torralba scene image library.Accuracy and confusion matrix is used as evaluation criterion.The result for the experiment show that this method has improved classification accuracy obviously.