A Study Of Key Technologies In Multispectral Image Color Reprocuction

Traditional color reproduction technologies cannot achieve the faithfulreproduction of the object color. To solve this problem, the spectral-based colorreproductiontechnologiesareresearchedinthisdissertation.Newschemesareproposedinthespectralcolorreproductionsystemframework,spectralreflectancereconstruction,dimension reduction of spectral data, and spectral color calibration. Then based on theabove theoretic research, a multispectral image color management system isconstructed.First, the architecture of the multispectral image color reproduction system isdiscussed. By analyzing the logical structure under ideal condition, and taking thedevelopment status of mustispectral devices and image processing technologies intoaccount, the achievable logical structure of the reproduction system is presented.Accordingtothis,thehardwareandsoftwareframeworkofthesystemis designed.Twoimplementationprocesses are proposed. Oneis basedonlow-dimensional interim spaceand another is based on spectral device model. Then the core technologies anddifficultiesofthetwoprocessesareanalyzed.Next, the spectral reconstruction technologies are researched to reach the goal ofremoving the information of observation condition and the camera characteristic fromtheinputspectralimagesandreconstructingthespectralreflectanceofthesourcescenes.The hardware of the narrow-band and wide-band spectral imaging systems issummarized. Three types of reconstruction methods are reviewed. They are directreconstruction, reconstruction by interpolation and learning-based reconstruction. Thena six channel spectral imaging system is established. Three applicable methods aresimulated by the six-channel system. Experiments show that the R-matrix method canbring better results than other methods because it can consider both the spectral andchromatic information in the digital image. Therefore, it can be used in spectralreproductionsystemtoreconstructthespectralreflectance.Then aiming at the issue that the high dimension of the multispectral imagesincreases the complexityof the spectral color reproduction, two spectral data dimensionreduction methods are proposed. To solve the problem that the reconstructed spectralreflectance maybe negative by using existing dimension reduction methods, anonnegative constrained principal component analysis method is proposed to construct a low-dimensional space by which the conversion between it and the multi-spectralspace can be achieved. Experiments show that the new method can make thereconstructed spectral reflectance data in the interval of [0, 1]. That maintains thephysical significance of the spectral reflectance. Aiming at the problem that theavailable dimension reduction methods cannot consider simultaneously the spectralaccuracy, chromatic accuracy and color difference stability under various illuminant, aWSPCAplus method is proposed. By introducing the characteristic of human visualsystem,thenewmethodimprovesthechromaticprecisionandcolordifferencestability.By compensating the spectral loss caused by heightening the chromatic accuracy, themethod reduces the spectral error. By combining the results from the first two stepsaccording to the application requirements, the method obtains the final dimensionreduction result. Experiments show the new method outperforms the existing methodsin the colorimetric accuracy, spectral accuracy and color difference stability undervariousilluminant.When multispectral images are reproduced using hardcopy devices, it is necessaryto transform the image data to the device color space and accomplish the non-linearcalibration. Among the existing multispectral input/output devices, only the multi-inkprinter is put into practical application. In addition, it is the device that has the mostcomplex non-linear characteristic. Therefore, this paper takes the multi-ink printer asthe research object, and presents two spectral color calibration schemes. By using thetraditional color correction technologies for the reference, a spectral color calibrationmethod of multi-ink printer using look-up table is proposed. Through combining thedimension reduction and look-up table technology, the method can handle the highdimension of the spectral image and take advantage of the look-up table with goodprecision and high speed. This enhances the accuracy and efficiency of spectral colorcorrection.Sinceestablishingdeviceanalytical model just needs measureafewsamples,a spectral color calibration method based on device spectral model is proposed. On thebasis of Yule-Nielsen Spectral Modified Neugebauer equation, the method takes boththe spectral and chromatic precision as the objective and uses the non-linearoptimization to accomplish the color calibration. Experiments show that the newmethod outperforms the existing model methods in chromatic and spectral accuracy.Meanwhile,ithastheadvantageofgooderrordistributionandcolorseparationresults.Finally, a multispectral image color management system is constructed accordingto the studies above. It utilizes the spectral CMM+Profile mode and is compatible withthe traditional colorimetric color management system. The spectral CMM of the system is designed first and the detail workflow of CMM is given. Then two data generationmethods of spectral target Profile are proposed. Lastly, the validity of the system isvalidated by objective and subjective metric. Experiments show that the two spectralcolor reproduction workflows presented in this dissertation have individual emphases.The workflow using look-up table has the advantage of high accuracy and efficiency,while the workflow using spectral output model has the advantage of low cost. Theyshould be chose in the light of applications. Meanwhile, the subjective observation alsoshow that the system can provide the faithful color reproduction of the scenes undervarious illuminant, which overcomes the inherent defect of the tradition colorreproductionsystem.