摘要
在夜视网络环境或低照度环境下,数字图像的采集和处理容易产生图像噪点增多、细节丢失的问题。而且现有图像边缘增强算法存在识别率差、效率低等问题。基于深度学习网络模型的图像边缘增强算法,依据图像边缘像素点的灰度差异,对原始图像做滤波预处理确定出数字图像的边缘特征,构建深度置信网络模型并定义概率分布函数;基于网络转换和稀疏编码规则融合图像边缘像素的多特征,并在深度置信网络模型输出层加入sigma函数降低迭代训练过程的复杂度,最后提取图像边缘的形态学梯度特征,剔除边缘孤立的特征点,达到数字图像边缘增强的目的。实验结果表明,增强算法处理后样本图像轮廓清晰,细节特征明显,且在相同的样本条件下迭代训练步数更少,图像边缘处理效果优于传统方案。
In the night vision network environment or low illumination environment,the acquisition and processing of digital image will easily lead to the increase of image noise and the loss of details,and the existing image edge enhancement algorithms have the problems of poor recognition rate and low efficiency.Therefore,the image edge enhancement algorithm based on deep learning network model,according to the gray difference of image edge pixels,does filtering preprocessing to the original image,determines the edge features of digital image,constructs the depth confidence network model and defines the probability distribution function;based on network transformation and sparse coding rules,fuses the multi features of image edge pixels,and uses the depth confidence network model Finally,the morphological gradient features of image edge are extracted,and the isolated feature points are removed to achieve the purpose of digital image edge enhancement.The experimental results show that the sample image processed by the enhancement algorithm has clear contour,obvious detail features,less iterative training steps under the same sample conditions,and the image edge processing effect is better than the traditional scheme.
作者
白玲玲
韩天鹏
BAI Lingling;HAN Tianpeng(Educational Affairs Office,Party School of Fuyang Municipal Party Committee of Anhui Province,Fuyang 236034,China;School of Computer and Information Engineering,Fuyang Normal University,Fuyang 236037,China)
出处
《北部湾大学学报》
2020年第11期26-30,70,共6页
Journal of BeiBu Gulf University
基金
阜阳市2019年社科规划项目课题:阜阳市基层社会综合治理网格化问题研究(FSK2019011)。
