A Novel Multi-Manifold Classification Model via Path-Based Clustering for Image Retrieval
Nowadays, with digital cameras and mass storage devices becoming increasingly affordable, each day thousands of pictures are taken and images on the Internet are emerged at an astonishing rate. Image retrieval is a process of searching valuable information that user demanded from huge images. However, it is hard to find satisfied results due to the well known “semantic gap. Image classification plays an essential role in retrieval process. But traditional methods will encounter problems when dealing with high-dimensional and large-scale image sets in applications. Here, we propose a novel multi-manifold classification model for image retrieval. Firstly, we simplify the classification of images from highdimensional space into the one on low-dimensional manifolds, largely reducing the complexity of classification process. Secondly, considering that traditional distance measures often fail to find correct visual semantics of manifolds, especially when dealing with the images having complex data distribution, we also define two new distance measures based on path-based clustering, and further applied to the construction of a multi-class image manifold. One experiment was conducted on 2890 Web images. The comparison results between three methods show that the proposed method achieves the highest classification accuracy.
Image retrieval image classification semantic gap visual semantics manifold learning distance measure path-based clustering
Rong Zhu Zhijun Yuan Junying Xuan
School of Science & Information Engineering, Jiaxing University, Jiaxing, China, 314001 Modern Education Technology Center, Xinjiang Vocational University, Urumqi, China, 830011
国际会议
桂林
英文
1-8
2011-11-01（万方平台首次上网日期，不代表论文的发表时间）