基于二维图像的三维模型分割算法研究

发布时间：2018-05-06 00:22

本文选题：映射关系 + 轮廓线提取　；参考：《西北大学》2017年硕士论文

【摘要】：本文设计实现了一种基于二维图像的三维模型分割方法。传统的三维模型分割过程常常伴随着较大的计算量与复杂度。与之相对的,二维图像分割技术已经非常成熟,而且图像分割过程相比三维模型分割过程而言更加的直观,计算效率高。针对上述问题,本文利用图像分割的结果指导完成三维模型分割。该方法利用三维模型预处理与图像分割相结合的方式,建立投影图像与三维模型的映射关系,提取投影图像中几何对象的轮廓线,并对其进行分割,最后将分割结果映射至三维模型。具体研究内容如下:1.提出了一种基于CPCA的二维图像至三维模型的映射方法。该方法利用CPCA方法对模型进行预处理,建立三维模型包围盒,并利用该包围盒对模型进行投影操作,获得二维图像至三维模型的映射关系。2.提出了一种基于点集的轮廓线提取方法。该方法通过二值图提取前景像素对应的点集,对该点集进行几何图像的构建,继而在此基础上进行轮廓线提取。通过该方法提取的轮廓线具有较高的精度,可以很好的保存物体的边缘信息,不易受到噪声点的影响。3.提出了一种改进了的基于曲率多尺度的自适应角点检测方法(AMCP方法)。该方法使用由点集提取的轮廓线来取代传统的Canny算子轮廓线,并且在AMCP方法的基础上对检测到的角点进行进一步的筛选,提高了检测准确率。4.提出了一种基于弱相关性的图像分割方法。该方法通过对邻域内角点间相关性的计算,排除具有强相关性的点,对具有弱相关性的点进行连接。形成了满足人类视觉特性的图像分割结果,并用该结果完成三维模型分割。
[Abstract]:In this paper, a three-dimensional model segmentation method based on two-dimensional image is designed and implemented. The traditional three-dimensional model segmentation process is often accompanied by a large amount of computation and complexity. In contrast, two-dimensional image segmentation technology has been very mature, and the process of image segmentation is more intuitive than three-dimensional model segmentation process, high computational efficiency. In order to solve the above problems, we use the result of image segmentation to complete 3D model segmentation. In this method, the mapping relationship between projection image and 3D model is established by combining 3D model preprocessing with image segmentation, and the contour line of geometric object in projection image is extracted and segmented. Finally, the segmentation results are mapped to the 3D model. The specific contents of the study are as follows: 1. A mapping method from two-dimensional image to three-dimensional model based on CPCA is proposed. This method uses CPCA method to preprocess the model, establishes the three-dimensional model bounding box, and uses the bounding box to project the model, and obtains the mapping relation between 2D image and 3D model. A method of contour extraction based on point set is proposed. In this method, the point set corresponding to foreground pixels is extracted by binary graph, and the geometric image of the point set is constructed, and then the contour line is extracted. The contour line extracted by this method has a high accuracy and can preserve the edge information of the object very well, and is not easily affected by the noise point. 3. An improved adaptive corner detection method based on curvature multi-scale is proposed. In this method, the contours extracted from the point set are used to replace the traditional Canny operator contours, and the detected corner points are further filtered on the basis of the AMCP method, which improves the detection accuracy. 4. A method of image segmentation based on weak correlation is proposed. By calculating the correlation between the corner points in the neighborhood, the method eliminates the points with strong correlation and connects the points with weak correlation. The image segmentation results satisfying the human visual characteristics are obtained, and the 3D model segmentation is completed with the results.
【学位授予单位】：西北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】