融合PLSA和随机游走模型的自动图像标注  

Integrating PLSA and Random Walk Model for Automatic Image Annotation

作　　者：田东平[1,2] 

机构地区：[1]宝鸡文理学院计算机软件研究所,山西宝鸡721007 [2]宝鸡文理学院计算信息科学研究所,山西宝鸡721007

出　　处：《小型微型计算机系统》2017年第8期1899-1905,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61202212)资助;陕西省教育厅专项科研计划项目(15JK1038)资助;宝鸡文理学院校级重点科研计划项目(ZK16047)资助

摘　　要：为了有效克服语义鸿沟问题,提出一种融合概率潜语义分析(PLSA)和随机游走(random walk,RW)模型的图像语义标注方法.从已标注图像的文本信息出发构建一个非对等模态的PLSA模型,以此计算未知图像的初始语义标注;,基于初始标注的语义信息和与之关联的图像的视觉信息构造标签相似性图,以有效避免图像标注过程中因多义词而引入的噪声数据;在所构造的相似性图上执行随机游走过程,进一步挖掘和分析初始标注之间的潜在语义关联,从而获得未知图像的精确化语义标注.通过在Corel5k图像集上的实验表明,本文方法(PLSA-RW)的性能明显优于若干经典的自动图像标注方法,而且具有更好的检索性能.To efficiently narrow down the problem of semantic gap, the current paper presents a semantic image annotation method by integrating probabilistic latent semantic analysis （ PLSA ） and random walk model （ abbreviated as PLSA-RW ）. To start with, a PLSA model with asymmetric modalities is constructed from textual information of the annotated images so as to predict the initial semantic annotations of the unseen images. Subsequently, a label similarity graph is constructed by a weighted linear combination of label similarity and visual similarity of images associated with the corresponding labels, which can effectively avoid the noise data caused by the polysemous words. Finally, a random walk process is implemented over the built label similarity graph to further mine and analyze the latent semantic correlations of the candidate annotations so as to capture the refining semantic annotations for the unseen images. Con- ducted experiments on the general-purpose Corel5k dataset demonstrate that the proposed PLSA-RW not only significantly outperforms several state-of-the-art automatic image annotation methods but also has better retrieval performance.

关 键 词：图像语义标注 概率潜语义分析 随机游走 语义鸿沟 图像检索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]