Research On Reflectance Reconstruction For Multispectral Imaging And Its Application

Along with the color image generally applied in many fields, color as the quality of image important standards makes acquirement and display of the accurate color particularly important. The color information get from different equipment are device-dependent, so these color information is related to the equipment. In order to obtain more precise color information which independent of equipment, propose the measure of multispectral reconstruction of color image- reconstruction the spectral reflectance of each pixel in the color image. The results of Reconstruction of multi image not only can contain more information than chromaticity space but also can effectively avoid the phenomenon of metamerism. The paper makes a systematic research to the process of spectral reconstruction and makes the multi-spectral imaging technology to be applied to HDR image synthesis.The multispectral reconstruction research includes two important parts, one is research on spectral reconstruction algorithm and the other is how to select the training sample. Previous researchers put forward a lot of effective spectral reconstruction algorithm, for example, the wiener estimation method and pseudo-inverse method for linear imaging system as well as polynomial regression method for nonlinear imaging system. Considering the noise in the imaging system have strong impact on the result of reconstruction of multispectral, this paper presents a kind of filter polynomial regression method combined filtering response value of camera with polynomial regression method. The experimental results show that this method can really improve the accuracy of the results of spectral reconstruction. The main idea of the training sample selection is to select some color sample which can cover a wider range of colors and represent main color. After the many years'exploration, researchers put forward many effective ways, including Hardeberg method using the singular value of the sample space and Westland Cheung method which depend on the Euclidean distance of judgment. Some of these methods based on color space and some based on high-dimension spectral reflectance space. In this paper, the methods introduced use a feedback type for the training samples selection which based on high-dimensional reflectance space, and its validity is proved by experiments.Finally, this paper combined multi-spectrum technology with HDR image synthesis technology. Through turning the HDR synthesis technology usually used in chromaticity space to spectral reflectance space. The multispectral image used contain more spectral information and can effectively avoid the phenomenon of metamerism which may result in non-precise phenomenon of color spectrum, making synthetic HDR image not only show high dynamic range but also can keep the original color of the real scene.