روش تولید رنگ توسط رگرسیون بردار پشتیبانی برای بینایی کامپیوتر رنگی

In the color computer vision system, the nonlinearity of the camera and computer screen may result in different colors between the screen and the actual color of objects, which requires for color calibration. In this paper, support vector regression (SVR) method was introduced to reproduce the colors of the nonlinear imaging system. Firstly, successive 3σ method was used to eliminate the large errors found in the color measurement. Then, based on the training set measured in advance, SVR model of RBF kernel was applied to map the nonlinear imaging system. In this step, two important parameters (C, γ) were optimized by the Least Mean Squared Validating Errors algorithm to get the best SVR model. Finally, this optimized model could predict the real values displayed on the screen. Compared with quadratic polynomial regression, BP neural network and relevance vector machine, the optimized SVR model has better ability in color reproduction performance and generalization.

In recent years, ICC (International Color Consortium) color management is gradually being received and adopted. The core of its management is to characterize the behavior of color description of each device in imaging systems, namely, the establishment of a function between RGB or CMYK of device control signal values and the tristimulus values. This function is often described in different ways, such as Look-up Table (LUT) combined with the interpolation [1], multiple regression [2] and neural networks [3] and [4], etc. Under normal circumstances, the Look-up Table method provides a precision higher than other methods, but it requires a lot of calibration samples. To reduce the data dimension of calibration samples, Wang et al. used the color correction technology for the domain partition of the multi-channel printer color correction [5]. Multiple regression works by means of polynomial approximation to the nonlinear characteristics of device color, featured in a simple conversion relationship and the lower calibration accuracy. Furthermore, the polynomial as a global function may lead to the local distortion to be extended to the whole color space. An effective way is to narrow the range of correction, that is, correction partition [6] and [7]. Theoretically, the neural network can approximate any nonlinear relationship, so it has a high applicability when used for color correction. One concern is the difficulty to determine the internal structure of neural networks, such as the hidden layer. In recent years, support vector machine based on the statistical learning theory has been playing a big role in terms of pattern recognition, image classification, function approximation, etc. And it also finds its way to be applied in the field of color correction [8]. However, research in this area is also relatively less. This paper presented an attempt to introduce a support vector machine model for establishing color correction, with correction of experimental data being used to test the accuracy of the model.

These above results may lead to the following conclusions: (1) Successive 3σ filter algorithm can be used to eliminate the larger errors found in the measurement data, and also able to improve the generalization capability of correction algorithm; (2) The Least Mean Squared Validating Errors algorithm can be used to optimize the SVR parameters, while reducing the amount of fitting and over fitting; (3) Compared with other methods in color correction, the SVR method presented a more accurate prediction of unknown samples, that is, chromaticity coordinates of its projections showed the more accurate estimation on the actual chromaticity coordinates.