A two-stage parametric subspace model for efficient contrast-preserving decolorization

rgb2gray转换模型是目前最为经典和流行的彩色图像灰度化方法。最新的研究表明,对一阶线性模型的三个加权参数进行自适应离散搜索具有良好的潜力。在本文中,我们提出一种两步法方式,将参数搜索策略扩充到二阶多变量多项式模型。该模型划分为三个子空间的和,本文论证了第一个子空间是最为重要的,并且第二和三个子空间可以看作为对其进行加细的部分。在模型的第一步,将梯度相似性测度(gradient correlation similarity,Gcs)用于第一个子空间,得到一个初步的灰度化图像。然后,再次使用Gcs对初步灰度图像和第二、三子空间所引出形成的候选图像进行最优解搜索。通过数值实验,在定量评价、定性视觉评价和算法复杂度方面表明了本方法的有效性。

The RGB2GRAY conversion model is the most popular and classical tool for image decolorization. A recent study showed that adapting the three weighting parameters in this first-order linear model with a discrete searching solver has a great potential in its c6nversion ability. In this paper, we present a two-step strategy to efficiently extend the parameter searching solver to a two-order multivariance polynomial model, as a sum of three subspaces. We show that the first subspace in the two-order model is the most important and the second one can be seen as a refinement. In the first stage of our model, the gradient correlation similarity （Gcs） measure is used on the first subspace to obtain an immediate grayed image. Then, Gcs is applied again to select the optimal result from the immettiate grayed image plus the second subspace-induced candidate images. Experimental results show the advantages of the proposed approach in terms of quantitative evaluation, qualitative evaluation, and algorithm complexity.