Font Size: a A A

Theapplication Of Neural Network In The Characterization Of Display

Posted on:2013-09-01Degree:MasterType:Thesis
Country:ChinaCandidate:L X LiFull Text:PDF
GTID:2248330371464564Subject:Printing Engineering and Media Technology
Abstract/Summary:PDF Full Text Request
The developing of color management is came up with the wide applied of digital color devices, aimed to settle the problem of color matching among different devices. Color display is a basic part in pre-printing field, which plays a role of“soft proofing”. The device characteristic as well as color gamut of display has great influence to the“soft proofing”of printing system and the quality of presswork, so that the management of display has very important research value for us. There are several of methods to achieve the color management, which are different in the methods of characterization or conversion, and among this neural network has special advantage in the process of nonlinear.Basis on the researching and analyzing of the space gamut and uniformity for RGB, CMYK, XYZ and Lab, we obtain the judgment that RBF neural network is more suited to process the RGB-Lab conversion characteristic of display by comparing the performance of back-propagation(BP) neural network and radial basis function(RBF) neural network.During the comparison research of BP and RBF performance, we firstly make use of the Adobe Gamma software of CRT display to calibrate CRT display. And subsequently make the 729 color patches by dividing RGB space into nine gray scales in Photoshop, i.e.0, 32, 64, 96, 128, 160,192,224,255, and by the recording of the RGB-Lab data for each color patch, we get the training samples and testing samples. In the end, we stimulate the 729 data in MATLAB software by LCD. Based on the evaluating of color difference formula of CIE1976Lab in simulating samples optimizing, we improve that RBF model is advantage over BP model in the network performance, so we choose RBF neural network to characterize display.The performance of RBF neural network relates to its training center values. Based on the analyzing of the optimized algorithm, we apply Orthogonal Least Squares (OLS)algorithm to optimize RBF neural network. Our results of MATLAB simulate experiment show that the conversion error from RGB to Lab color space based on RBF neural network can be reduced in a certain extent by optimizing with OLS algorithm, and OLS-RBF model has good conversion effect in display characterization.
Keywords/Search Tags:display, characterization, back-propagation neural network, radial basis function neural network, orthogonal least squares
PDF Full Text Request
Related items