基于DnCNN函数的分水岭算法
|
摘要
在日常生活中,由于系统和设备问题,图片经常会引入噪声,图像也会因此变得很模糊.而分水岭算法经常因为噪声的缘故使图像出现过分割.针对传统分水岭算法进行图像分割易受噪声影响的问题,研究了一种基于Dn CNN函数的分水岭算法.基于当下神经网络对于图像识别、归类、特征提取这一领域所做出的卓越成绩,利用神经网络的良好图像统计特性进行图像去噪,并改进传统的激活函数,图像经过神经网络去噪后图像细节更加明显和清晰,再利用改进的分水岭算法,使用最大相似性合并规则对去噪后的图像进行分割.并将该算法与K-means算法和噪声图片+分水岭算法这两个算法进行比较,基于Dn CNN函数的分水岭算法对图像的分割精度更高.
In daily life,noise,which is often introduced into pictures because of system and equipment problems,blurs images dramatically. However,watershed algorithm often causes image segmentation because of noise.In order to solve the problem that the traditional watershed algorithm in image segmenting is easily affected by noise,the current study was conducted on a watershed algorithm based on Dn CNN function. With reference to the outstanding achievements made by the current neural network in the field of image recognition,classification and feature extraction,the image denoising was carried out by adopting the good image statistical characteristics of the neural network,and thus the traditional activation function was improved. After the image was denoised by the neural network,the details of the image are more obvious and clearer. Furthermore,by applying the improved watershed algorithm and the maximum similarity merging rule,the denoised image was segmented. The comparison between the three algorithms—the current algorithm,the K-means algorithm and the noise picture + watershed algorithm—showed that the segmentation accuracy of the image segmented by the watershed algorithm based on DnCNN function is the highest.
In daily life,noise,which is often introduced into pictures because of system and equipment problems,blurs images dramatically. However,watershed algorithm often causes image segmentation because of noise.In order to solve the problem that the traditional watershed algorithm in image segmenting is easily affected by noise,the current study was conducted on a watershed algorithm based on Dn CNN function. With reference to the outstanding achievements made by the current neural network in the field of image recognition,classification and feature extraction,the image denoising was carried out by adopting the good image statistical characteristics of the neural network,and thus the traditional activation function was improved. After the image was denoised by the neural network,the details of the image are more obvious and clearer. Furthermore,by applying the improved watershed algorithm and the maximum similarity merging rule,the denoised image was segmented. The comparison between the three algorithms—the current algorithm,the K-means algorithm and the noise picture + watershed algorithm—showed that the segmentation accuracy of the image segmented by the watershed algorithm based on DnCNN function is the highest.
引文
[1]陈强．基于聚类技术的多阈值图像分割技术[J]．太赫兹科学与电子信息学报，2018,16(4):715-718.
[2]魏巍，张嘉宇，陈千，等．基于熵的多尺度决策表最优尺度选择[J]．海南热带海洋学院学报，2018,25(5):61-65.
[3]龙金辉．融合同态滤波与马尔可夫随机场的棉花图像分割[J]．江苏农业科学，2018,46(20):232-235.
[4]王栋，于威威，庄斐弘．基于标记分水岭和FLICM模糊聚类的图像分割方法研究[J]．微型机与应用，2017,36(17):49-51+58.
[5]路正佳．基于改进分水岭算法的彩色图像分割方法[J]．科技通报，2018,34(9):172-175.
[6]CHEN JIAN,YAN BIN,JIANG HUA,et al. Interactive image segmentation by improved maximal sim-ilarity based region merging[C]//Proceedings of 2013 IEEE International Conference on Medical Imaging Physics and Engineering,IEEE-Beijing Section,2013:279-282.
[7]HASAN M,EL-SAKKA M R. Improved BM3D Image Denoising Using SSIM-Optimized Wiener Filter[J]. Eurasip Journal on Image&Video Processing,2018,25:1-12.
[8]AHRON M,ELAD M,BRUCKSTEIN A. K-SVD:An Algorithm for Designing Overcomplete Dictionaries for Sparse Representation[J]. IEEE Transaction on signal Processing,2006,54(11):4311-4322.
[9]VINCENT L,SOILLE P. Watersheds in digital space:An efficient algorithms based on immersion simulation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1991,13(6):583-598.
[10]刘玉兰，刘小平，邹艳妮．改进的自适应权值核范数最小化去噪算法[J]．计算机工程与设计，2018,39(1):212-217+229.
[11]林越，刘廷章．基于区间分析的新能源并网削峰填谷调度策略研究:以光伏微网为例[J]．海南热带海洋学院学报，2018,25(5):66-73.
[12]李芝峰，景源．基于直方图峰值优化的阶梯k-means聚类算法[J]．电子技术与软件工程，2019,153(6):66-67.
[2]魏巍，张嘉宇，陈千，等．基于熵的多尺度决策表最优尺度选择[J]．海南热带海洋学院学报，2018,25(5):61-65.
[3]龙金辉．融合同态滤波与马尔可夫随机场的棉花图像分割[J]．江苏农业科学，2018,46(20):232-235.
[4]王栋，于威威，庄斐弘．基于标记分水岭和FLICM模糊聚类的图像分割方法研究[J]．微型机与应用，2017,36(17):49-51+58.
[5]路正佳．基于改进分水岭算法的彩色图像分割方法[J]．科技通报，2018,34(9):172-175.
[6]CHEN JIAN,YAN BIN,JIANG HUA,et al. Interactive image segmentation by improved maximal sim-ilarity based region merging[C]//Proceedings of 2013 IEEE International Conference on Medical Imaging Physics and Engineering,IEEE-Beijing Section,2013:279-282.
[7]HASAN M,EL-SAKKA M R. Improved BM3D Image Denoising Using SSIM-Optimized Wiener Filter[J]. Eurasip Journal on Image&Video Processing,2018,25:1-12.
[8]AHRON M,ELAD M,BRUCKSTEIN A. K-SVD:An Algorithm for Designing Overcomplete Dictionaries for Sparse Representation[J]. IEEE Transaction on signal Processing,2006,54(11):4311-4322.
[9]VINCENT L,SOILLE P. Watersheds in digital space:An efficient algorithms based on immersion simulation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1991,13(6):583-598.
[10]刘玉兰，刘小平，邹艳妮．改进的自适应权值核范数最小化去噪算法[J]．计算机工程与设计，2018,39(1):212-217+229.
[11]林越，刘廷章．基于区间分析的新能源并网削峰填谷调度策略研究:以光伏微网为例[J]．海南热带海洋学院学报，2018,25(5):66-73.
[12]李芝峰，景源．基于直方图峰值优化的阶梯k-means聚类算法[J]．电子技术与软件工程，2019,153(6):66-67.