基于多尺度注意力残差网络的图像阴影去除算法  

Image Shadow Removal Algorithm Based on Multi-Scale Attention Residual Network

作　　者：陈腾飞 CHEN Tengfei(Xi'an Shiyou University,Xi'an Shaanxi 710065,China)

机构地区：[1]西安石油大学,陕西西安710065

出　　处：《信息与电脑》2022年第24期113-116,共4页Information & Computer

摘　　要：针对在复杂环境下的阴影去除存在阴影区域颜色、纹理失真以及半阴影区域去除不够彻底的问题,提出了一种基于生成式对抗网络的多尺度注意力残差网络的图像阴影去除算法。首先,使用4个不同大小的卷积核提取输入图像特征,以获得不同尺度上的特征信息;其次,通过压缩和激励网络(Squeeze and Excitation Networks,SE-Net)网络校准每个通道的特征信息,以增强重要特征信息并弱化非重要特征信息;最后,拼接所有通道,得到最终具有多尺度特征信息的特征图。实验表明,所提出算法在复杂阴影场景中能准确恢复阴影区域的颜色、纹理及阴影边缘等细节,其性能优于现有的主流算法。In view of the color and texture distortion of shadow region and the incomplete removal of penumbra region in complex environment, an image shadow removal algorithm based on multi-scale attention residual network based on generative adversarial network is proposed. Firstly, four convolution kernels of different sizes are used for feature extraction of the input images to obtain feature information of at different scales, and then the feature information of each channel is calibrated through Squeeze and Excitation Networks(SE-Net) to enhance important feature information and weaken non-important feature information, Finally, all channels are spliced to obtain the final feature map with multi-scale feature information. Numerical experiments show that the proposed algorithm can accurately recover the color,texture, shadow edge and other details of the shadow region in complex shadow scenes.

关 键 词：图像处理 阴影去除 生成式对抗网络 残差网络 多尺度注意力 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]