摘要
针对GrabCut算法在分割图像时效率低,且容易出现欠分割与过分割的问题,提出了一种基于概率神经网络(PNN)改进的GrabCut(PNN_GrabCut)算法。该算法用PNN模型替换GrabCut算法中的高斯混合模型(GMM)进行t-links权值计算,以提升算法的计算效率;通过构建前景和背景直方图,选取像素值出现频率较高的像素作为PNN模型的训练样本,以提高算法的分割精度。在公开的ADE20K数据集中选取图像进行分割实验,结果表明,PNN_GrabCut算法的分割精度优于其他对比算法,且效率较高。对前景与背景相似度高的图像进行分割实验,结果表明,PNN_GrabCut算法的分割精度明显高于GrabCut算法。
Aiming at the low efficiency of GrabCut algorithm in image segmentation,and the problems of under-segmentation and over-segmentation,an improved GrabCut algorithm based on probabilistic neural network(PNN)(PNN_GrabCut)is proposed in this paper.The algorithm replaces the Gaussian mixture model(GMM)in the GrabCut algorithm with PNN model to calculate the weight of t-links to improve the calculation efficiency of the algorithm.By constructing the foreground and background histograms,the pixels with higher pixel values are selected as training samples of the PNN model to improve the segmentation accuracy of the algorithm.In the public ADE20K data set,images are selected for segmentation experiments.The results show that the segmentation accuracy of PNN_GrabCut algorithm is better than other comparison algorithms,and the efficiency is higher.For image segmentation experiments with high similarity between foreground and background,the results show that the segmentation accuracy of PNN_GrabCut algorithm is significantly higher than that of GrabCut algorithm.
作者
张翠军
赵娜
Zhang Cuijun;Zhao Na(School of Information Engineering,Hebei GEO University,Shijiazhuang,Hebei 050031,China;Hebei Center for Ecological and Environmental Geology Research,Hebei GEO University,Shijiazhuang,Hebei 050031,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2021年第2期236-243,共8页
Laser & Optoelectronics Progress
基金
河北省高等学校科学技术研究重点项目(ZD2019134)
河北省研究生创新资助项目(CXZZSS2019113)。
关键词
图像处理
概率神经网络
高斯混合模型
图像分割
image processing
probabilistic neural network
Gaussian mixture model
image segmentation
