Multispectral Color Imaging And Processing Methods

The multispectral color imaging system measures the spectral reflectance across the surface of a sample with spatially-fine resolution,by acquiring spectral images at different bands.Compared with the conventional spectrophotometer,the multispectral color imaging system is contact-less and efficient.Besides,the system provides the general color appearance of the sample.However,the system suffers from many problems which prevent the accurate acquisition of the multispectral image and reduce the accuracy in spectral color measurement.This thesis deals with two key issues in the system which influence the measurement accuracy.Based on the acquired images with high color fidelity,this thesis proposes a method to recolor textile fabric images.The main contributions of this thesis are as follows:1.Due to the existence of multiple light reflections,the lighting condition inside the inte-grating sphere multispectral imaging system can be changed by the measured samples.A lighting deviation correction method is thus proposed which incorporates both system design and regression modeling.Under the assumption of ideal Lambertian reflection,a reference white patch is intro-duced according to the imaging model.The white patch characterizes the change of the lighting condition inside the sphere,and is able to compensate for the camera response.In practical mea-surements,there are many imperfect factors,including the direction dependence of the sample's reflection characteristics.A polynomial regression model is proposed to perform further correc-tion,where the training data are obtained by placing small color patches on different background samples.Experimental results illustrate that the proposed method eliminates the inconsistency of both camera responses and reconstructed spectral reflectances considerably,which consequently improves the accuracy in spectral color measurement under different lighting conditions.2.The multispectral imaging system acquires band images by rotating the filter-wheel,where misalignments may occur due to the possible mechanical vibration.A fast block-based multispec-tral image registration method is thus proposed.Considering the limited and spatially slowly vary-ing misalignments between band images,a band image is divided into blocks,on which matching processes are performed individually.The image blocks are selected according to the structural information of the image.The displacement of a selected block is computed using a similarity measure which is based on the gradient of the image.The parameters of the global affine transform between band images can then be estimated from corresponding point pairs efficiently.Experi-mental results validate that the proposed method makes band images in good alignment with each other,and hence effectively improves the accuracy in spectral color measurement.The proposed method is computationally efficient as the post-processing of multispectral image acquisition.3.To produce textile fabric images with modified colors for textile design,a recoloring method is proposed based on the high-fidelity images acquired by the system.Two steps,i.e.,fuzzy clus-tering and color modification,are involved.The clustering process incorporates the color of a pixel as well as the local spatial and color information.By introducing the concept of color credibility of the surrounding pixel,the fuzzy clustering performs well in regions with rich details.Based on the obtained membership matrix,two most relevant dominant colors of a pixel can be determined.The color of the pixel is then modeled as a linear combination of the two dominant colors,inside which the residual error term contains part of the texture information.By replacing a dominant color with a given color,the image is recolored through color combination.Experimental results demonstrate that the proposed method can produce natural and faithful color appearances on both printed and yarn-dyed fabric images,with preserved texture structure,controlled coverage scope,and high color fidelity.