Abstract:The reflectance spectrum has a high dimension and has nothing to do with illumination and observation. It can truly and objectively describe the color information of an object. It is determined by the characteristics of the object itself, so it is called the "fingerprint" of the object. However, the amount of data of reflectance spectra is more than ten times that of the traditional three-color system, and these huge spectral data cause huge burdens in terms of storage, data processing, and data transfer, and spend too much computing time. If the high-dimensional spectrum can be mapped to the low-dimensional space through mathematical transformation methods, and ensure that the low-dimensional space data can better represent the information covered by the original spectrum, the multi-spectral data can be effectively compressed and the processing efficiency of spectrum-based color reproduction can be improved. PCA treats all wavelengths in the visible range equally, and the reconstructed spectrum is only a mathematical approximation to the original spectrum, which often leads to the problem of small spectral reconstruction error and large colorimetric reconstruction error. A multispectral dimension reduction algorithm based on the weight function of spectral color difference is proposed in this paper. The dimensionality of Munsell was reduced to one dimension by PCA, and then restored to 31 dimensions, and the average spectral color difference between Munsell's original spectrum and its reconstructed spectrum was used as a weight function. Taking NCS as the training sample and NCS, Munsell and 3 multispectral images as the test samples respectively, the performance of the proposed method of this paper and the classical PCA and the other four weighted PCA are analyzed and compared. CIELAB color difference under the conditions of multiple Lighting and viewing (D65/2° and A/2°) and root mean square error (RMSE) evaluate the colorimetric reconstruction accuracy and spectral reconstruction accuracy between the original spectra and the reconstructed spectra of the test sample respectively. The experimental results show that: compared with PCA, the proposed method has greatly improved colorimetric reconstruction accuracy at the expense of a small amount of spectral reconstruction accuracy. The improvement of colorimetric reconstruction accuracy is very important for spectral color reproduction. The results also show that the colorimetric reconstruction accuracy of the proposed method is better than that of the other four existing weighted PCA.
Key words:Multispectral dimensionality reduction; Multispectral compression; Principal component analysis; Weight function; Spectral color difference