N-to-sRGB Mapping for Single-Sensor Multispectral Imaging

Proposed Algorithm Overview

Abstract

     Visualization of a multispectral image in a standard color space, typically the sRGB space, is an important task for human color perception. When we reproduce the sRGB image from the multispectral image with N spectral bands, an N-to-sRGB mapping is required. The challenge of the N-to-sRGB mapping in single-sensor multispectral imaging with a multispectral filter array (MSFA) is to reduce demosaicking error amplification and propagation across different spectral bands, which are not trivial because of very sparse sampling of multiple spectral bands in a single image sensor. In this paper, we propose a novel N-to-sRGB mapping pipeline for effectively suppressing the demosaicking error amplification and propagation. Our idea is to apply guided filtering in the mapped sRGB space using one of input N band images before the amplification and propagation as a guide image. Experimental results demonstrate that our proposed pipeline improves the mapping accuracy for various MSFA types.

Reference

• N-to-sRGB Mapping for Single-Sensor Multispectral Imaging
  Yusuke Monno, Masayuki Tanaka and Masatoshi Okutomi
  Proc. of IEEE ICCV Workshop: 5th Color and Photometry in Computer Vision Workshop (CPCV2015), pp.66-73, December, 2015. [PDF, Slides]