基于上下文信息的自适应场景分类算法

An adaptive scene classification algorithm based on contextual information

本文针对场景分类中存在的目标物以及之间的相互关系错综复杂,图像的拍摄角度、光照强度不同,造成的场景内容难以辨识等问题,提出了一种利用上下文关联信息进行自适应判别的分类算法。通过检测图像中目标物及其相互之间的上下文关系,利用该信息对图像的细节纹理特征进行增强,最后利用支持向量机(SVM)进行训练和测试,从而实现场景分类。在三个公共标准图像集上的测试实验结果表明,都具有较好的分类效果。

In this paper,aiming at the problems by the complexities of the objects and their relationship in the scene classification,the shooting angles of the images,the different light intensities and the difficulty of recognizing the scene contents.We proposed a classification algorithm that uses contextual information for adaptive discrimination.By detecting the objects in the image and their contextual relationships,the information is used to enhance the detailed texture features of the image.Finally,training and testing are performed using a support vector machine(SVM)to achieve scene classification.The test results from the three public standard image datasets show that the method obtains good classification results.