APP下载

基于改进神经网络的图像边缘分割技术

2018-08-21卫洪春

现代电子技术 2018年16期

卫洪春

摘 要: 采用梯度下降法进行图像边缘分割时受到噪声的干扰,训练过程中存在局部最佳解,从而导致图像边缘分割效果和泛化性能差。为此,提出基于改进神经网络的图像边缘分割方法,采集样本图像的中值特征量、基于梯度的特征量、Krisch算子方向特征量,融合三个特征向量塑造具备较强抗噪性能的样本图像特征向量,通过基于特征向量和BP神经网络的边缘检测算法,将样本图像特征向量输入四层BP神经网络,采用改进BP算法训练四层BP神经网络,采用训练后的改进神经网络完成图像边缘分割。实验结果表明,所提图像边缘分割方法细节保有性能强,分割精度和泛化能力强。

关键词: 改进神经网络; 图像边缘; 图像分割; 梯度特征; 中值特征; 改进BP算法

中图分类号: TN911.73?34; TP391.41 文献标识码: A 文章编号: 1004?373X(2018)16?0112?04

Abstract: There exist noise interference when the gradient descent method is used for image edge segmentation and local optimal solution in its training process, resulting in poor image edge segmentation effect and generalization performance. Therefore, an image edge segmentation method based on improved neural network is proposed. The median feature quantity, gradient?based feature quantity, and Krisch operator direction feature quantity of sample images are collected. The three feature vectors are fused to shape the feature vector with strong anti?noise performance for the sample image. The feature vectors of sample images are input into the four?layer BP neural network by means of the edge detection algorithm based on the feature vector and BP neural network. The improved BP algorithm is used to train the four?layer BP neural network. The improved neural network is used to complete image edge segmentation. The experimental results show that the proposed image edge segmentation method has strong detail preservation performance, segmentation precision, and generalization capability.

Keywords: improved neural network; image edge; image segmentation; gradient feature; median feature; improved BP algorithm

圖像的关键特征是边缘,其广泛应用在计算机视觉、模式识别等领域。图像边缘检测是数字图像处理领域研究的重点[1]。寻求有效的图像边缘分割方法,对于医疗、军事、探测以及工业等领域具有重要的应用价值。以往BP神经网络方法大都采用梯度下降法实现图像边缘分割,其训练时会出现局部最佳值,导致图像分割精度大大降低,泛化性能差。图像边缘检测是一种分类过程,可通过BP神经网络进行有效分类,本文通过样本图像训练BP神经网络,通过训练好的神经网络检测图像边缘。神经网络训练过程中,采集的特征需要对噪声点同真实边缘间的差异进行充分分析,该方法抗噪性能强[2]。……

登录APP查看全文