Scholarlink

<p>Reflected light can be seen as a mixture of body reflection (diffuse component) and interface reflection (specular component). Applications using the diffuse reflection model assume Lambertian surfaces, and consider the specular reflection pixels as outliers. To recover or remove the highlights, a computational algorithm is used to separate the components in color space. In this paper, we generalize a polarization-based separation method to the case of multispectral images with K channels. A spectropolarimetric synthetic dataset of 10 scenes, including diffuse and specular reference images, is made available for both RGB and multispectral modalities. We found that our spectral adaptation qualitatively and quantitatively improves the separation results as compared to the RGB version. We also propose an algorithm optimization to reduce the computation time without sacrificing image quality.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Diffuse and specular component separation using multispectral and polarization images

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月20日

查看全文
  • Pierre-Jean Lapray,
  • Benjamin Mourllion,
  • Laurent Bigué

摘要

Reflected light can be seen as a mixture of body reflection (diffuse component) and interface reflection (specular component). Applications using the diffuse reflection model assume Lambertian surfaces, and consider the specular reflection pixels as outliers. To recover or remove the highlights, a computational algorithm is used to separate the components in color space. In this paper, we generalize a polarization-based separation method to the case of multispectral images with K channels. A spectropolarimetric synthetic dataset of 10 scenes, including diffuse and specular reference images, is made available for both RGB and multispectral modalities. We found that our spectral adaptation qualitatively and quantitatively improves the separation results as compared to the RGB version. We also propose an algorithm optimization to reduce the computation time without sacrificing image quality.