【作者】 李新元；

【导师】 林克正；

【作者基本信息】 哈尔滨理工大学 ， 计算机应用技术， 2009， 硕士

【摘要】 随着计算机技术的发展,数字图像处理技术己经成为视觉图像研究的主要手段。但是,迄今为止人类还未能构建出一种达到高等生物视觉性能标准的自主视觉系统。其中,有一个重要因素就是生物视觉系统能够很容易完成复杂的图像分割,而机器视觉的分割却很难达到相应的分割速度及精度。因此要使机器视觉技术进一步发展,必须研究快速有效的图像分割技术。本论文借鉴人类视觉机理,将先验知识和客观信息相结合,主要进行以下三个方面的工作。1.针对特定类自然图像分割提出了一种新的基于外观模板的图像对象分割方法,主要有三个改进之处,首先引入一种新的分片方法和分片记分方法生成形状片断,使高质量的形状码片产生效率更高,且更具代表性;其次鉴于以前的模型方法中特征提取或者抽象度不够,或者抛弃了太多的图像信息,提出一种新的包含了形状和纹理的对象特征模型,该特征模型具有更加良好的对象类别区分能力,而且能够很大程度容忍不同对象实例的外观差异和成像条件变化;最后将基于密度和层次的快速聚类方法引入到算法体系当中,初步解决多对象图像分割的问题,并且对遮挡对象图像的分割效率也有一定的程度的提高。2.基于Snake模型的图像分割研究。探讨气球Snake模型和GVF Snake模型以及GVF-Balloon Snake模型的原理,比较分析他们的优点和缺陷。根据GVF-Balloon Snake模型的特点,选取基于外观模型的图像对象分割算法得到的轮廓线作为初始轮廓线,完成对象的精确分割。3.基于Snake模型的分割方法与基于纹理和形状的图像对象分割方法(T&C-SEG)相结合的组合式分割方法研究。分析自底向上与自顶向下分割各自的优势与不足,根据特定类自然图像特点,设计了两种方式兼用的组合式对象分割方法B&T-IOSEG(Bottom-up and top-down combined image object segmentation)算法。实验表明该算法是有效的,且提高了分割效果。更多还原

【Abstract】 With the development of computer technology, digital image technology has become the principal mean of the image vision research. But now human beings are still unable to build an autonomous vision system which is able to meet the standards of biological vision. There is one important factor that the biological visual system can easily complete complex image segmentation, on the other hand, the segmentation of machine vision is very difficult to meet the appropriate speed and accuracy. Therefore, for the further development of machine vision technology, it is necessary to study the image segmentation technique to find a rapid and effective method.In this thesis, combined with the priori knowledge and objective information,and guided by mechanim of human vision, more works has been done as follows: First, the objective of this work is the detection of object classes. The thesis first develops a novel technique to extract class-discriminative boundary fragments and the texture features near the boundary then boosting is used to select discriminative boundary fragments (weak detectors) to form a strong“Boundary-Fragment-Model”detector. A new appearance model is built with those entire detectors and the texture features. And then, the boundary fragment and the texture features and used to complete detection. To the end, a new fast cluster algorithm is used to deal with the centroid image. The generative aspect of the model is used to determine an approximate segmentation. In addition, this thesis presents an extensive evaluation of the new method on a series of test images and compares its performance with the existed methods which are from the literature. As it is shown in the experiment, the new method outperforms previously published methods with the overlap part of the object in multiple-object scene.And then, this thesis researches the differerce among the three kinds of snake model. According to the characteristics of the snake model, this thesis takes the results of the T&C-SEG method as the initial contour of the snake model. Then, the GVF-Balloon Snake model is choosen to complete the segmentation. Experiments show that this method can enhance the effect of segmentation.Finally, this thesis analyzes the advantages and disadvantages between bottom-up segmentation method and top-down segmentation method, and researches the modular segmentation method between snake model and the template-based image segmentation method. According to the characteristics of natural image, this thesis develops a combined image segmentation method. Experiments show that the method can significantly enhance the efficiency of segmentation.更多还原