A method for selecting training samples based on camera response. (5th August 2016)

• Record Type: Journal Article
• Authors: Zhang, Leihong; Li, Bei; Pan, Zilan; Liang, Dong; Kang, Yi; Zhang, Dawei; Ma, Xiuhua
• Is Part Of: Laser physics letters. Volume 13:Number 9(2016:Sep.)
• Publication Date: 2016-08-05
• Subjects: sample selection -- spectral reflectance reconstruction -- response value; Lasers -- Periodicals; Physics -- Periodicals; 621.366
• DOI: 10.1088/1612-2011/13/9/095201
• Languages: English
• ISSNs: 1612-2011

Abstract

Abstract: In the process of spectral reflectance reconstruction, sample selection plays an important role in the accuracy of the constructed model and in reconstruction effects. In this paper, a method for training sample selection based on camera response is proposed. It has been proved that the camera response value has a close correlation with the spectral reflectance. Consequently, in this paper we adopt the technique of drawing a sphere in camera response value space to select the training samples which have a higher correlation with the test samples. In addition, the Wiener estimation method is used to reconstruct the spectral reflectance. Finally, we find that the method of sample selection based on camera response value has the smallest color difference and root mean square error after reconstruction compared to the method using the full set of Munsell color charts, the Mohammadi training sample selection method, and the stratified sampling method. Moreover, the goodness of fit coefficient of this method is also the highest among the four sample selection methods. Taking all the factors mentioned above into consideration, the method of training sample selection based on camera response value enhances the reconstruction accuracy from both the colorimetric and spectral perspectives.