Studies On Spectral Encoded Acquisition And Color Reproduction Of Images

Traditionally,both imaging and display technologies are based on the color encoding of images,which compress the information of the spectral dimension to three channels,i.e.red,green and blue.This methodology is straightforward and practical.However,it throws out lots of information carried by the spectrum.For traditional color imaging devices,they struggle to realize accurate spectral and color detection because of a limited information acquisition amount.For traditional three-primary displays,the output information amount is also limited.Therefore,they cannot satisfy the demand for more abundant and accurate color display.In fact,spectrum not only reflects the characteristic of materials,but also the“fingerprint”of color.To realize the accurate acquisition and reproduction of spectrum and color,the spectral performance of imaging and display devices has to be improved,and thus increase the spectral information amount in the imaging acquisition-processing-reproduction chain.In recent years,with the continuous development of computer science and advanced manufacturing technology,spectral imaging and multi-primary display technology are pushed forward,which significantly improves the spectral performance of imaging and display devices.However,the existing de-vices and methods still have some disadvantages.Focused on them,corresponding studies and analyses were carried out.For the encoding acquisition of spectral images,the study started from the large volume,low image quality,and slow reconstruction speed problems of the current spectral imaging devices.Instructed by compressive sensing and deep learning,the compact spectral imaging systems based on broadband spectral encoding filters were designed and built respectively.Based on compressive sensing theory,the size of the spectral imaging system was shrunken to several centimeters while maintaining a pixel amount of above 105.Based on deep learning theory,the quantitative optimization design of the filters was resolved throughout the co-design of hardware and software.The computational simulation and experimental analysis show that the proposed methods realized a dramatic increase in reconstruction speed.The accuracy of filter design and spectral reconstruction was also improved.The mean square error(MSE)of spectral reconstruction reaches 10-3.For the color reproduction of spectral images,the current displays can only reproduce a small part of colors in spectral reflectance images.Focused on this problem,a wide display color gamut evaluation method was proposed,which was based on the optimal color coverage of the uniform color space.Based on this method,a practical multi-primary laser display design was offered,which gives a nearly ultimate color gamut as well as a high optical efficiency.The main creativity of this dissertation includes:Proposed and realized the compact spectral encoded imaging system based on compressive sensing.Simplified and improved the classical compressive sensing reconstruction algorithm,thus improved the spectral reconstruction accuracy by 7.5 times and increase the computation speed by 12 times.Proposed and realized the compact spectral encoded imaging system based on deep learn-ing.Regarding the spectral encoding process of filters as a fully connected layer of a neural network,solved the inverse design problem of random filters within a deep learning frame-work.Compared with the compressive sensing algorithm,the spectral reconstruction accuracy was improved by 6 times,and the computation speed was increased by 4000 to 11000 times.Proposed the display gamut optimization design method based on the spectral reflectance color coverage in uniform color space(UCS).Taking laser display as an example,analyzed the trade-off between display gamut and optical efficiency.The proposed design scheme realized an ultra-wide color gamut covering almost the entire spectral reflectance color gamut while maintaining a relatively high optical efficiency.