摘要
通过研究图像分割算法的原理和实验对比,可以发现标记分水岭分割方法是以边缘特性为基础,通过设置光谱特征标记来分割图像,其分割结果边缘精度高,但是仍然存在较为严重的过分割和欠分割情况。而mean shift分割方法和eCognition分割方法是以光谱特征为分割依据的分割方法,虽然它们分割结果的过分割和欠分割情况较少,但是分割对象的边缘精度较差。分析以上存在的问题后,通过融合边缘特征和区域特征,并且依据一定的特征来选择种子点,尽量避免种子点选择在边缘区,从而实现提高图像分割的效果。通过实验取得了好的分割效果,说明改进的图像分割方法是可行的。
By analysing their principles and experiment contracts, we can find that the marker controlled watershed segmentation is an edge-based segmentation method, it segments the image by setting the spectral feature marker and has high edge precision in segmentation result, but still has a lot of over-segmentation and under-segmentation situations. As the segmentation methods based on spectral feature, the mean shift segmentation method and the eCognition segmentation method have less over-segmentation and under-segmentation situations in segmentation results, but the edge precision of segmenting objects is poorer. With the analysis on the problems above, in this paper we select the seed points according to certain features through fusing the edge featu~＇es and regional features, and do best to avoid the seed points to be chosen near the edges, so as to improve the image segmentation effects. Better result is also obtained through the experiment, this proves that the improved segmentation method is feasible.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第1期194-196,200,共4页
Computer Applications and Software
基金
国家自然科学基金项目(41071274)
国家科技支撑计划课题(2012BAC16B01)
