基于视觉差异特征的跨域图像匹配方法研究

项目名称： 基于视觉差异特征的跨域图像匹配方法研究

项目编号： No.61502364

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 计算机科学学科

项目作者： 李静

作者单位： 西安电子科技大学

项目金额： 21万元

中文摘要： 人类视觉系统能够轻易在不同场合中辨别具有明显特征的事物，这种“跨域图像匹配”能力既不会受光照、色彩、视角、尺度、遮挡等因素影响，也不会因为照片、素描、油画等载体的差别而减弱。如何模拟人类视觉感知能力进行“跨域图像匹配”是当前图像匹配领域的难点问题。本项目在借鉴认知生理学和神经科学的基础上，从视觉感知建模、跨域视觉特征集合构建、跨域视觉差异特征提取与分类三方面出发，研究一套新的视觉差异特征驱动的跨域图像匹配模型与方法，包括: (1)研究视觉感知机理启发的跨域图像显著区域提取方法；(2)研究跨域图像底层视觉特征单元构建方法；(3)研究视觉差异特征选择与分类器学习算法；(4)构建专用数据库与性能测试平台,对跨域图像匹配方法进行实验验证和评估。本项目的研究将为互联网跨域图像检索识别，异源传感器配准等领域的相关研究提供理论基础和实验依据，具有重要的理论和应用价值。

中文关键词： 图像匹配；视觉差异特征；特征提取与描述

英文摘要： Human visual system can easily recognize object with unique features in different scenes. This “cross domain image matching” ability is stable by illumination, color, viewpoint, scale, occlusion, and it is robust with different carrier such as photo, sketch or oil painting. Simulating the human visual perception ability for cross domain image matching is one of the essential problems in the field of image matching. To solve this problem, in this project we plan to research and develop a novel distinctive visual feature driven based cross domain image matching model. Inspired by the theory and recent improvements of cognitive physiology science and neuroscience, we start the research work from three aspects, which contain visual perception modeling, cross domain visual feature set creation, and distinctive visual feature extraction and classification. The research content of this project mainly include three parts: (1)human visual perception mechanism inspired cross domain image salient region extraction, (2)low level cross domain visual feature unit creation,(3)distinctive cross domain visual feature extraction and learning, and(4)designing and development of dataset and experimental platform for performance evaluation. The research project will provide theory foundation and experimental references for many application fields such as cross domain image retrieval and recognition in internet and multi-sensor image registration and fusion, and it has important theoretical and practical value.

英文关键词： image matching;distinctive visual feature;feature extraction and description