计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (16): 242-248.DOI: 10.3778/j.issn.1002-8331.2101-0126

• 图形图像处理 • 上一篇    下一篇

基于pix2pix网络结构的单曝光HDR图像生成方法

张媛媛,张红英   

  1. 1.西南科技大学 信息工程学院,四川 绵阳 621010
    2.西南科技大学 特殊环境机器人技术四川省重点实验室,四川 绵阳 621010
  • 出版日期:2022-08-15 发布日期:2022-08-15

Single Exposure HDR Image Generation Method Based on Pix2pix Network Structure

ZHANG Yuanyuan, ZHANG Hongying   

  1. 1.School of Information Engineering, Southwest University of Science and Technology, Mianyang, Sichuan 621010, China
    2.Robot Technology Used for Special Environment Key Laboratory of Sichuan Province, Southwest University of Science and Technology, Mianyang, Sichuan 621010, China
  • Online:2022-08-15 Published:2022-08-15

摘要: 针对单曝光高动态范围(high dynamic range,HDR)图像生成过程中图像欠曝光区域的细节恢复和图像色彩偏移的问题,提出一种基于pix2pix网络结构的单曝光HDR图像生成算法,采用pix2pix自编码网络结构,利用卷积步长代替池化层简化网络结构,在卷积过程中采用Leaky ReLU激活函数,有效地保留图像的细节信息,并使用绝对误差和余弦相似度函数构建损失函数,使网络能够对图像阴暗区域进行较好的扩展,保证图像色彩不发生偏移。实验结果证明,该算法能在有效地扩展图像阴暗区域的同时保证图像色彩不发生偏移。

关键词: 高动态成像, pix2pix网络, 余弦相似度, 图像色彩

Abstract: Aiming at the problem of detail recovery and color shift of underexposed area in single exposure HDR(high dynamic range) image generation, a single exposure HDR image generation algorithm based on pix2pix network structure is proposed. Pix2pix self coding network structure is adopted, convolution step is used to replace pooling layer to simplify network structure, and leaky network is used in convolution process, ReLU activation function can effectively retain the details of the image, and the absolute error and cosine similarity function are used to construct the loss function, so that the network can better expand the dark area of the image, and ensure that the image color does not shift. The experimental results show that the algorithm can effectively expand the dark area of the image while ensuring that the image color does not shift.

Key words: high dynamic imaging, pix2pix network, cosine similarity, image color