交互式图像分割与实现.pdf

交互式图像分割与实现 张磊 （清华大学土木工程系， 土博14， 2014310053 ） 摘 要：图像分割是从输入图像中提取目标或感兴趣区域的过程，是目标检测和识别过程中的重要步骤。由于目 前的计算机人工智能还远未达到理解高层语义的水平，全自动分割算法很难获得令人满意的结果。交互式图像分 割相对于传统图像分割的最大的优点在于，加入了用户主观的指导，使得过去许多因为数据歧义而很难解决的问 题变得定义清晰。因此，人机交互式图像处理于近年兴起，并得到越来越多的重视。本文研究实现的是一种针对 彩色图像的交互式通用目标分割系统。首先，自行设计图割算法中的区域惩罚项和边界惩罚项函数，并在此基础 上增加K 均值聚类预处理以提升分割速度。然后，采用基于特殊聚类预处理的区域增长算法对分割后的图像进行 二次交互分割，提高分割结果的准确率。其次，用形态学开闭运算对图像分割结果进行二值滤波，进一步提高分 割准确率。最后在图像分割结束后，系统还可以实现以下功能：为目标更换背景、平移目标位置、对图像进行平 滑处理。 关键词：图像分割；交互式；图割；区域生长 Application and Implementation of the Interactive Image Segmentation System Zhang Lei (Department of Civil Engineering, Tsinghua University, The first grade of Dr.in 2014, 2014310053) Abstract: Segmenting foreground objects from natural images are a fundamental task in image understanding. But it is a challenging problem. A fully automatic algorithm cannot yet achieve satisfactory results. The difficulties lie in the high complexity in visual pattern modeling and the intrinsic ambiguity in visual pattern grouping. To make the problems solvable, one method is to construct interactive frameworks, and allow the user to specify the foreground objects and the background. The biggest advantage of interactive image segmentation, compared to tradition segmentation methods, is that with the guidance of the user, problems which previously were ambiguous in a semantic sense now become clearly defined and solvable. Therefore, interactive image processing was developed in recent years, and has been receiving more and more attention. In this paper, a kind of interactive general target segmentation system for color images is cre