基于视觉感知的图像分割评价方法研究

项目名称： 基于视觉感知的图像分割评价方法研究

项目编号： No.61202190

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 彭博

作者单位： 西南交通大学

项目金额： 24万元

中文摘要： 图像分割是计算机视觉和图像处理领域中一个重要且具挑战性的研究问题。图像分割质量的评价对分割算法的理论发展和实际应用都有重要的影响，然而目前还缺乏成熟完善的图像分割质量评价体系。本项目从视觉感知的角度出发对图像分割质量进行评价，目的是构建图像分割质量评价的新理论和方法，弥补现有统计学等计算方法在该领域能力的不足。研究内容以自然图像的分割结果为对象从两个方面展开：1. 基于参考分割（有监督）的质量评价，建立包含多个参考分割的图像数据库，设计基于边界/像素点和区域的评价方法，讨论分析这些方法在评价中的性能表现。2. 无参考分割的质量评价，研究统计学意义下具有较稳定属性特征的低层边界结构对图像目标的表达能力，通过模式识别和机器学习方法设计在线评价算法。本项目研究将为图像分割质量客观评价提供理论、方法和算法，并完善现有的图像分割质量客观评价体系，进一步促进图像分割技术的应用和发展。

中文关键词： 图像分割评价；图像分割；参考分割；；

英文摘要： Image segmentation is an important yet still challenging problem in computer vision and image processing. Evaluation of image segmentation quality has an essential impact on the theoretical and practical development of segmentation algorithms. However, now it still lacks of mature and complete evaluation framework of the segmentation quality evaluation. This project studies the evaluation problem from the view of visual perception. The goal is to develop new theory and methodologies for the task, and overcome the limitations and defects in the existing techniques. The study focuses on the natural images and includes two parts. The first one is the ground-truth based (supervised) evaluation, for which an image segmentation database will be constructed. The boundary/pixel based and region based evaluation methods will be studied, as well as the performance of these methods. The second one is the non-ground-truth based (unsupervised) evaluation, where we will put the effort on the representation of objects by relatively stable features from low-level boundary structures. Then on-line evaluation algorithms can be designed based on the pattern recognition and machine learning techniques.This project will contribute to the segmetnation quality evaluation with theory, methodology and algorithms. Moreover, it can improv

英文关键词： image segmentation evaluation；image segmentation；ground truth；；