A Hyperspectral Reflectance Reconstruction Method Based on RGB Image

Abstract

The invention discloses a hyperspectral reflectance reconstruction method based on RGB images, relates to a hyperspectral reflectance reconstruction method, and belongs to the field of computational photography. The invention is divided into a training phase and a use phase. In the training phase, in the training set, the hyperspectral reflectance is mapped to the RGB color space, and the chromaticity of each pixel is solved according to the RGB value; the pixels are clustered according to the chromaticity of each pixel; the pixel reflectance of each cluster is Use dictionary learning to obtain a sparse dictionary of reflectance; map the sparse dictionary to RGB space to obtain an RGB dictionary. Use the stage to white balance the collected RGB image; solve the chromaticity of each pixel of the image, and find the cluster to which each pixel belongs according to the chromaticity; for the pixels in each cluster, use the cluster RGB dictionary to carry out constraints Sparse coding: Reconstruct the hyperspectral reflectance of a pixel from the clustered reflectance dictionary and sparse coding. The invention can improve the reconstruction accuracy without special equipment and has a fast reconstruction speed.

Description

technical field [0001] The invention relates to a hyperspectral reflectance reconstruction method, in particular to a hyperspectral reflectance reconstruction algorithm based on RGB images, and belongs to the field of computational photography. Background technique [0002] Hyperspectral imaging technology is different from traditional color image imaging technology. The images it acquires usually include dozens or hundreds of narrow-band channels, far more than the 3 or 4 channels of traditional color images. The image obtained by this technique is usually called a data cube. Compared with traditional images, which only have spatial dimensions, it has three dimensions: space and spectrum. [0003] Hyperspectral imaging technology has a wide range of application scenarios. This technology can be used for target segmentation, tracking and recognition in the field of computer vision. It was mainly used in remote sensing in the early days, but in recent years it has also been ...

AI Technical Summary

Problems solved by technology

[0006] There are two types of hyperspectral reflectance reconstruction methods in the prior art: the first method collects the hyperspectral image of the scene, and calculates the hyperspectral reflectance according to the known scene illumination. This method requires that the scene illumination is known and needs to be In the darkroom, a special light source is used to illuminate the scene and collect it, which requires high equipment and environment, and the collection of hyperspectral images usually takes a long time

The second method uses sparse representation technology to obtain a single sparse dictionary using the training set of hyperspectral reflectance, and then collects the RGB image of the scene to estimate the hyperspectral reflectance of each pixel. The disadvantage of this method is that the accuracy is relatively Low, but requires no special equipment and rebuilds are usually faster

Images