基于感受野空间属性的视觉计算模型及图像检索研究

项目名称： 基于感受野空间属性的视觉计算模型及图像检索研究

项目编号： No.61463008

项目类型： 地区科学基金项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 刘广海

作者单位： 广西师范大学

项目金额： 45万元

中文摘要： 基于视觉信息处理机制的计算模型已成为研究热点。本项目以视觉计算模型和图像检索为研究对象，主要研究内容包括视觉显著图的计算，整合特征显著图以及结构共生词汇模型等, 依据模式识别和神经生理学等领域的最新研究成果，提出一种基于感受野空间属性的视觉计算模型，在此基础上进一步提出结构共生词汇模型，并应用于图像检索。拟解决两个关键科学问题：（1）如何建立视觉特征整合模型，使其更加符合视觉信息处理机制；（2）如何在视觉词汇中融入先验知识，以便于缩小视觉词汇的歧义性。本项目提出的视觉计算模型是Itti视觉注意模型的重大改进，结构共生词汇模型则是对视觉词汇包模型的升华。它们能够为建立更符合视觉信息处理机制的视觉计算模型以及缩小视觉词汇的歧义性提供新思路。

中文关键词： 视觉注意；图像检索；感受野；视觉词汇

英文摘要： Visual attention model based on the mechanisms of visual information processing has become a hot topic. The research objects are visual computational model and image retrieval.The main works include the computation of saliency maps, the integration of feature saliency maps and the visual words model of structure co-occurrence. We adopt the recent achievements of pattern recognition and neurophysiology, and present a visual computational model based on the spatial properties of receptive fields, and further put forward the visual words model of structure co-occurrence on this basis, and use it to image retrieval.There are two key scientific problems need to be solved, one is how to build the feature integration model, and makes it more akin to the human's mechanisms of visual information processing, the other is how to reduce the ambiguity of visual words by embedding priori knowledge into visual words. The proposed visual computational model is the significant improvement of Itti visual attention model, in addition，the proposed visual words model of structure co-occurrence is the significant improvement of bag-of-visual words. They can provide new ideas on how to build a visual computational model that more akin to the mechanisms of visual information processing and on how to reduce the ambiguity of visual words.

英文关键词： visual attention;image retrieval;receptive field;visual words