基于LS-SVR的图像矫正

最小二乘支持向量回归(the least squares support vector regression,LS-SVR)算法因其回归拟合度高广泛应用于各领域中.以目标物在不同光源下采集的图像呈现出不同的颜色值,从而导致图像与目标物出现视觉上的偏差为研究对象,并以潘通色卡为参照,利用LSSVR算法,结合将RGB颜色空间到sRGB颜色空间的转换模型,对测试图像进行矫正处理.实验结果表明:与多项式回归相比,LS-SVR算法能取得更小的色差,且矫正后的图像更接近于目标图像.

Image correction based on LS-SVR

The least squares support vector regression (LS-SVR) algorithm is widely applied to diverse research fields due to the advantage of higher fitting degree. Various color values are acquired from images gathered under diverse illuminants, which cause the deviation between the images and the objective. Pantone cards are regarded as reference and LS-SVR algorithm is employed to process image correction via the transformation model from RGB to sRGB color space. As illustrated in the experimental results, a better performance for the lower value of chromatic aberration and highly approximating to the object image after image correction is obtained by LS-SVR algorithm compared with polynomial regression.