目标自适应图像分块与小目标提取  

Object Adaptive Image Division and Object Extraction

作　　者：朱立新[1] 夏德深[1] 

机构地区：[1]南京理工大学计算机系,南京210094

出　　处：《计算机工程与应用》2005年第25期88-90,共3页Computer Engineering and Applications

摘　　要：针对传统二维直方图方法的难点,提出了采用基于分水岭变换的图像自适应分块的解决方法,新方法能使得每个小目标都被分割在同一个图像区域内,克服了传统图像分块方法采用固定分块,易造成将同一目标分到多个区域的缺点。方法中首先采用了基于标记点的灰度图像重建方法对图像进行预处理,在自适应增强目标的同时也克服了分水岭变换易造成过度分割的影响,在此基础上进一步地对图像采取了基于分水岭变换的图像分块,接着在每一个分块区域中采用引入目标分布信息阈值选取方法,得到二值化的结果。实验表明该方法目标分割结果稳定,适合于小目标的分割提取。The conventional selt-adaptive image thresholding based on two-dimensional histogram is improved through proposed object adaptive image partition.This new method,using watershed transform to split image into regions,makes every wanted object contained in a one and only region.As is well-known,traditional image partition of two-dimensional histogram often causes one wanted object to be splitted into several regions and needs complicated post-processing.Our methods applies image gray reconstruction operation to the original image at first and this eliminated over-segmentation caused by watershed transform,then partition based on watershed transform is applied,at last an improved thresholding method which considers the distribution of objects in same region,is used to get final result.Many experimental results show that this proposed method is efficient in small object extraction.

关 键 词：分水岭变换 灰度重建 目标提取 图像分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]