【作者】 田春娜；

【导师】 高新波；

【作者基本信息】 西安电子科技大学 ， 信号与信息处理， 2005， 硕士

【摘要】 图像分割是连接低级视觉和高级视觉的桥梁。然而分割因其不适定性,成为阻碍计算机视觉发展的瓶颈问题之一。因此,几十年来图像分割始终是计算机视觉中的一个研究热点。人们希望通过引入人的作用来解决分割的不适定性问题,这样交互式图像分割方法就应运而生了。在分析现有的基于相对模糊连通度的交互式图像分割算法的基础上,本文提出了一种序列图像分割的快速算法。对于单帧图像,新算法在保持分割精度的同时,运行速度提高了3 倍。作为该算法的扩展,一方面将原算法中单目标分割推广到多目标分割,另一方面将单帧图像的分割推广到序列图像的分割,实现了复杂背景下多目标的图像分割和图像序列的批处理分割。为了提高算法的分割速度和准确性,本文还提出了一种预分割方案,即先对图像进行预分割,然后再用快速模糊连通度的交互式图像分割算法进行精细分割。并进行了分割结果的后处理:提取出目标图像的单像素宽度的光滑边缘;为便于人眼分辨图像更精细的结构和区分多目标分割的结果,引入了伪彩色增强技术;并对序列图像的分割结果进行三维重建。此外,为了提高分割算法的抗噪声性能,还提出一种基于冗余小波变换的相对模糊连通图像交互式分割算法。对于单帧图像,与原基于尺度的相对模糊连通交互式图像分割算法相比,新算法在几乎不改变分割效果的前提下,速度提高了十几倍。并且该新算法对不同的噪声模型都具有较好的鲁棒性。文中还对小波变换的不同尺度的影响进行了较为系统地研究。实验结果表明,本文所提出的新算法可实现对非噪声和噪声图像中目标物体的快速而准确的分割。用人造图像和实际的医学图像和图像序列所做的测试实验取得了令人满意的分割效果。这些新算法的总体性能皆优于传统的基于模糊连通度的交互式图像分割算法,具有较高的实用价值。更多还原

【Abstract】 Image segmentation is the important foundation of higher-level computer vision.And it remains an open problem. However, image segmentation is an improperlyposed problem, which makes it one of the bottlenecks of computer vision. To solve theimpropriety, assistance of human is necessary. As a consequence, the interactivesegmentation has been arisen.A fast interactive segmentation algorithm of image-sequences based on relativefuzzy connectedness is presented, through modifying the existing one. In comparis-on with the existing algorithm, the proposed one, with the same accuracy, acceleratesthe segmentation speed by three times for single image. Meanwhile, this fastsegmentation algorithm is extended from single object to multiple objects and fromsingle-image to image-sequences. Thus the segmentation of multiple objects fromcomplex background and batch segmentation of image-sequences can be achieved. In order to improve the accuracy and speed of the segmentation algorithm, apre-segmentation scheme is presented to extract the ROI. In addition, a post-processing scheme is incorporated, in which we can extracts smooth edge with one-pixel-width for each segmented object. Pseudo-color processing is introduced todistinguish the fine structure of objects and the segmented multi-objects. And 3Dreconstruction is imported for the sake of illustrating image-sequences segmentationresults.Furthermore, a relative fuzzy connected interactive segmentation algorithmusing redundant wavelet transform is proposed to reduce the sensitivity to noise ofsegmentation. In contrast to the original scale-based algorithm, the proposed one,with almost the same accuracy, accelerates the segmentation speed by tens times forsingle image. And it has robustness to different noise models. In addition, a study ofthe influence of different redundant wavelet decomposition scales is also included. The experimental results illustrate that those proposed algorithms in this papercan obtain the ROI accurately from medical images as well as man-made images,which are nonnoised or noised, quickly and reliably with only a little interaction.And they have better performances over the traditional interactive segmentationalgorithms based on fuzzy connectedness.更多还原