期刊论文

中文

中国农机化学报

2095-5553

2020

41

6

159-165

10.13733/j.jcam.issn.2095-5553.2020.06.026

湖南省自然科学基金面上项目(2019JJ40133) 国家自然科学基金(61972146)。

本校为第一机构

随着近几年深度学习的不断发展,计算机视觉需要处理大量的图像数据。为解决图像阴影给计算机视觉领域带来的干扰,基于YCbCr颜色空间,给出了图像的本影检测框架和图像的半影与边界去除框架,从而提出了一种基于YCbCr颜色空间的近景图像的阴影去除方法。基于绿萝数据集的实验结果,在阴影检测中平均IOU值为93.4%,平均精度为0.962,误检率为0.010,漏检率为0.017,整体的SSIM为0.979,PSNR为36.84,不仅去除了图像的阴影,也较好的保护了图像的重要信息,在近景图像阴影的去除效果中表现良好。

As the application of deep learning in computer vision,more and more image data needs to be processed,but image shadows affect the efficiency of processing.To solve the interference caused by image shadows,we proposed a shadow removal method for the close-up images based on the YCbCr color space.Moreover,the researchers gave two frames of the image.One is the umbral detection frame,the other is the penumbra and boundary removal frame.The experimental result showed that the average IOU value in the shadow detection is 93.4%,the average precision value is 0.962,false detection rate is 0.010,miss...