Establishment Of A Light Source And Plant Spectral Reflectance Database And Color Recovery Of Low-light Images

With the advancement of image sensor CCD/CMOS technology,the acquisition of true color images using ordinary three-channel cameras under low illumination conditions has become a research topic in this field.The color is determined by the spectral reflectance of the surface of objects and the illumination source.A great deal of research has been done on estimating light sources under normal lighting conditions and on estimating the spectral reflectance of an object’s surface.This thesis explores color recovery methods for ordinary three-channel cameras in low illumination conditions.A database of light source spectra,including natural low light levels from the full moon and artificial light sources,and a database of natural plant spectral reflectance have been established.On this basis,the use of the principal components obtained from the database to estimate light sources and spectral reflectance for image color recovery is investigated.The specific work is described as follows.（1）A database of the spectral power distribution of daylight,full moonlight,starlight and some artificial light sources has been established.The wavelength range of the natural light database is from 400-1000 nm at 4 nm intervals,while the wavelength range of the artificial light sources is from 400-700 nm at 4 nm intervals.Principal component analysis is conducted on daylight,full moonlight and artificial light respectively,and the experimental results showed that the first principal component contribution of daylight is as high as 99.00%,and the first principal component contribution of full moonlight is as high as 98.90%;while the first principal component of artificial light is 62.51%,and the cumulative contribution of the first three principal components is 99.78%.Thus,both daylight and full moonlight can express most of the characteristics of a light source using the first principal component;artificial light sources require the use of the first three principal components to express most of the characteristics of a light source.Analysis of the color temperature,chromaticity coordinates,color rendering index and color difference to the 24 color card under a standard D65 light source for artificial light sources provides the basis for light source selection.（2）A database of spectral reflectance of selected plants is established for the wavelength range 400-1000 nm at 4 nm intervals.The characteristics of the spectral reflectance are analysed in terms of the spectral shape,in conjunction with the first-order derivative and red-edge parameters,and the possible factors responsible for this phenomenon.Principal component analysis is performed on leaves and flowers in the 400-700 nm and 400-1000 nm bands,respectively.The leaf analysis experiments show that the cumulative contribution of the first three principal components in the visible and visible and NIR bands is 98.62%and 94.97%respectively.For the flower analysis experiments showed that the cumulative contribution of the first three principal components in the visible and visible and NIR bands is 97.21%and 89.67%respectively.Thus the spectral reflectance of the plant leaves can be well reconstructed by the first three principal components.For flowers the first three principal components only give a good reconstruction of the spectral reflectance of flowers in the 400-700 nm band.（3）A color recovery method for estimating plant spectral reflectance using the principal components of light source and plant spectral reflectance is proposed in the thesis.The method of the estimation of light source and plant spectral reflectance is investigated natural daylight and moonlight conditions using a 24 color standard color card with known spectral reflectance,the first principal component of the light source and the first three principal components of the plant spectral reflectance.The experiments are conducted under two lighting conditions:daylight and full moonlight.The results show that the correlation coefficient between the estimated daylight and the data is 0.9592;while the correlation coefficients between the estimated light sources and the database for the four scenes under low illumination conditions at night are 0.9876,0.9913,0.9923 and 0.9929,in that order.The color recovery experiments with estimated plant spectral reflectance show that the estimated spectral reflectance in the 400-700 nm band reflects the spectral properties of plants under daylight;the RGB color difference between the reconstructed image and the original image is less than 10^-13.However,the spectral reflectance reconstructed under low illumination conditions does not accurately represent the spectral characteristics of the plant.Nevertheless,replacing the estimated light source with daylight can improve the visual impact of images taken at night to some extent.