摘要
当图像对比度低或者光照条件复杂时,图像目标形状、颜色等特征都具有很高的变异性,难以准确地识别目标边界,导致图像分割结果中目标区域与真实标注的重合程度较低。为此,文章研究基于生成对抗网络的小波域自适应图像分割技术。该技术利用小波域分析基函数,将图像从空间域转换为模糊集域;考虑图像对比度,计算图像变换的扩散强度与各向同性扩散系数;采用反模糊变换函数恢复图像中存在的模糊效应,获取目标边缘增强的图像,计算图像像素之间的相似度;通过引入距离信息,计算分割图像的相对熵和权值,对生成对抗网络进行优化处理,实现图像的自适应分割。实验结果表明,所提技术可以对图像进行精准分割,并保留丰富的细节信息,平均重叠率达到了96.3%,分割准确性较高,这表明该技术在图像分割任务中具有较高的可靠性,为后续的图像处理任务提供高质量的输入数据。
When the image contrast is low or the lighting conditions are complex,the image target shape,color and other characteristics have high variability,and it is difficult to accurately identify the target boundary,resulting in a low degree of overlap between the target area and the real annotation in the image segmentation results.To this end,the wavelet domain adaptive image segmentation technology based on generative adversarial networks is studied.Using the wavelet domain analysis basis function,the image is converted from the spatial domain to the fuzzy set domain.Considering the image contrast,the diffusion intensity of the image transformation is calculated.By using anti blur transformation function to restore the blurring effect in the image,the image with enhanced target edges is obtained,and the similarity between image pixels is calculated.The relative entropy and weight of the segmented image are calculated by introducing the distance information,furthermore to optimize the processing tasks of the generative adversarial network and realize the adaptive segmentation of the image.The experimental results show that the proposed technology can accurately segment the image and retain the rich detailed information.The average overlap rate reaches 96.3%,and the segmentation accuracy is high,indicating that the technology has high reliability in the image segmentation task and provides high quality input datas for the subsequent image processing tasks.
作者
梁丽香
李怀颖
何成亿
LIANG Lixiang;LI Huaiying;HE Chengyi(Kaili university,Kaili 556011,China)
出处
《无线互联科技》
2024年第14期117-119,共3页
Wireless Internet Technology
基金
凯里学院计算机图形学课程教学范式改革项目,项目编号:FS201817
黔东南州科技计划项目,项目名称:数字经济与黔东南旅游产业高质量发展——作用机制及区域异质研究,项目编号:黔东南科合J字[2022]89号。
关键词
生成对抗网络
小波域分析
图像分割
边缘增强
generative adversarial network
wavelet domain analysis
image segmentation
edge enhancement
