基于改进Graph Cut算法的生猪图像分割方法

发布时间：2018-05-30 06:15

本文选题：图像处理 + 图像分割　；参考：《农业工程学报》2017年16期

【摘要】：生猪图像分割为生猪行为特征提取、参数测量、图像分析、模式识别等提供易于理解和分析的图像表示,准确有效的生猪图像分割是生猪行为理解和分析的基础。针对传统Graph Cut算法分割精度差、分割效率低及不能准确分割特定目标的问题,该文结合交互分水岭算法,提出基于改进Graph Cut算法的生猪图像分割方法。采用交互分水岭算法对图像进行区域划分,划分的各个区域块看作超像素,用超像素替代传统加权图中的像素点,构造新的网络图替代传统加权图,重新构造能量函数以完成前景背景的有效分割。试验结果表明:该方法峰值信噪比平均范围为[30,40],结构相似度平均范围为[0.9,1],两种评价准则的结果与主观评价一致,图像分割质量、精度得到明显提升;平均耗时缩短到传统Graph Cut算法的33.7%,提高了分割效率;在复杂背景、噪声干扰、光照强度弱等条件下可以快速分割出特定目标生猪,具有较高鲁棒性。
[Abstract]:Pig image segmentation is the basis of pig behavior understanding and analysis, such as extraction of pig behavior characteristics, parameter measurement, image analysis, pattern recognition and so on, which provide easy to understand and analyze the image representation. Accurate and effective pig image segmentation is the basis of pig behavior understanding and analysis. Aiming at the problems of poor segmentation accuracy, low segmentation efficiency and inaccurate segmentation of specific targets in traditional Graph Cut algorithm, this paper proposes an improved Graph Cut algorithm for pig image segmentation based on the interactive watershed algorithm. The interactive watershed algorithm is used to divide the region of the image. Each area block is regarded as super pixel, and the pixel points in the traditional weighted map are replaced by the super pixel, and a new network graph is constructed to replace the traditional weighted map. The energy function is reconstructed to complete the efficient segmentation of foreground background. The experimental results show that the average range of peak signal-to-noise ratio (PSNR) of this method is [30 ~ 40] and the average range of structural similarity is [0.9 ~ 1]. The results of the two evaluation criteria are consistent with subjective evaluation, and the image segmentation quality and accuracy are improved obviously. The average time consuming is shortened to 33.7 of the traditional Graph Cut algorithm, and the segmentation efficiency is improved. Under the conditions of complex background, noise interference and weak illumination intensity, the hogs can be quickly segmented with high robustness.
【作者单位】：中国农业大学信息与电气工程学院;
【基金】：国家高技术研究发展计划(863计划)资助项目(2013AA102306)
【分类号】：S828;TP391.41

【相似文献】