计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (6): 204-208.DOI: 10.3778/j.issn.1002-8331.1711-0367

• 图形图像处理 • 上一篇    下一篇

基于改进的Chan-Vese模型与边缘转换的图像分割算法

李  康1,杨玉东2   

  1. 1.南京工业大学 计算机科学与技术学院,南京 211800
    2.淮阴工学院 电子信息工程学院,江苏 淮安 223003
  • 出版日期:2019-03-15 发布日期:2019-03-14

Image Segmentation Algorithm Based on Improved Chan-Vese Model and Edge Transformation

LI Kang1, YANG Yudong2   

  1. 1.School of Computer Science and Technology, Nanjing University of Technology, Nanjing 211800, China
    2.School of Electronic Information Engineering, Huaiyin Institute of Technology, Huai’an, Jiangsu 223003, China
  • Online:2019-03-15 Published:2019-03-14

摘要: 为了实现物联网环境下果园飞鸟的自动驱离,使其复杂条件下能够准确驱赶空中的飞鸟,提出了一种基于改进的Chan-Vese模型与边缘转换的空中飞鸟分割算法。通过准确识别飞鸟,为系统自动发出超声波驱离飞鸟提供准确的信息。利用Canny算子获取飞鸟图像的边缘信息;使用欧氏距离计算得到二进制边缘的距离映射;引入S形函数,构建边缘转换图;引入自动局部比,对Chan-Vese模型进行改进,以准确分割边缘映射图。实验结果表明:与SBGFRLS算法、G-CV算法和FAST EDGE算法相比,该算法具有更高的分割精度,在面对单目标图像分割时,其区域匹配率最高,约为70%,而均方根误差比率只有13%;对于含双目标的图像分割时,其区域匹配率最高,约为85%,而均方根误差比率只有5%。

关键词: 几何主动轮廓, Chan-Vese算法, 自动局部比, 边缘信息, 欧氏距离, 边缘变换图

Abstract: In order to realize the automatic displacing of the bird in the orchard under the environment of the Internet of things so that it can accurately drive the birds in the complex conditions, an image segmentation algorithm based on improved Chan-Vese model and edge conversion is proposed in this paper. Through the accurate identification of flying birds, it can provide accurate information for system automatically emitting ultrasonic waves to drive away flying birds. Firstly, the Canny operator is used to obtain the edge information of the bird image. And the Euclidean distance is used to get the distance map of binary edge. Then the edge conversion map is constructed by introducing S shape function. The automatic local ratio is introduced to improve the Chan-Vese model for accurately segmenting edge map. Experimental results show that this algorithm has higher segmentation accuracy compared with the SBGFRLS algorithm, G-CV algorithm and FAST EDGE algorithm, which the region matching rate is the highest, about 70%, while the root mean square error ratio is only 13% in the face of single target image segmentation, and the region matching rate is the highest, about 85%, while the root mean square error ratio is only 5% for the image segmentation with double targets.

Key words: geometric active contours, Chan-Vese algorithm, automatic local ratio, edge information, Euclidean distance, edge transform graph