一种新的基于语义聚类和图算法的自动图像标注方法

A New Approach for Automatic Image Annotation Based on Semantic Clustering and Graph Algorithm

针对图像检索中的语义鸿沟问题,提出了一种新颖的自动图像标注方法。该方法首先采用了一种基于软约束的半监督图像聚类算法(SHMRF-Kmeans)对已标注图像的区域进行语义聚类,这种聚类方法可以同时考虑图像的视觉信息和语义信息。并利用图算法——Manifold排序学习算法充分发掘语义概念与区域聚类中心的关系,得到两者的联合概率关系表。然后利用此概率关系表标注未知标注的图像。该方法与以前的方法相比可以更加充分地结合图像的视觉特征和高层语义。通过在通用图像集上的实验结果表明,本文提出的自动图像标注方法是有效的。

A novel automatic image annotation approach is proposed to bridge the semantic gap of content-based image retrieval. Our approach first performs segmentation of images into regions, followed by the clustering of regions to blobs using a semi-supervised image clustering algorithm with soft constraints which utilizing the visual and semantic information of images. And a graph-based algorithm is used to compute the probabilistie relation between concepts and region blobs which can he used to annotate new images. Experiments conducted on standard dataset demonstrate the effectiveness and efficiency of the proposed approach for image annotation.