对比度、颜色一致性和灰度像素保持的消色算法

目的 为了解决目前消色算法中不能同时保持原始图像的对比度,颜色一致性和灰度像素特征的问题,提出一种新的优化算法,最大限度地同时保留这些视觉特性。方法 为了保持原始图像的结构和局部对比度信息,用双高斯模型构建像素对之间的误差能量项;为了保持颜色一致性,采用局部线性嵌入模型构建能量项,确保原始图像中颜色一致的像素在结果图像中也拥有一样的灰度级;为了保持灰度像素特征,先标记出原始图像中的灰度像素,并强制规定这些像素的灰度值是已知的且在消色变换的过程中始终不变,然后用双高斯模型构建出灰度像素与其他像素之间的误差能量项。线性结合这3个能量项,得到目标能量函数,再通过迭代法求解出使总能量值达到最小的灰度值,从而得到了最终的消色结果。结果 实验结果表明,本文算法能够同时较好地保持原始图像中的对比度、颜色一致性和灰度像素特征。结论 本文算法基本符合人类对图像对比度变化的感知程度,而且能够很好地保持细节信息和全局结构,可应用于数字打印、模式识别等方面,具有很大的应用价值。

Decolorization with contrast, color consistency and grayscale pixel preservation

Objective In theory, color-to-gray transformation is a process of dimensionality reduction, thus making information loss inevitable. Therefore, the goal of decolorization is to use the limited gray level range to preserve as much information of the original color image as possible. Researchers have proposed many related algorithms. However, the algorithms fail in the simultaneous preservation of the local and global contrast, as well as the contrast, color consistency, and grayscale pixel, of the original image. To solve these problems, we propose a new approach that can maximally maintain the features of the original color image. Method To preserve the structure and local information, we use a bimodal Gaussian distribution, followed by the difference between the pixel and its neighbors, to construct the error energy function. For global color consistency, we use locally linear embedding to build the energy function, which causes the same color pixels to exhibit the same gray levels in the result. For grayscale feature preservation, we distinguish grayscale pixels and specify that the gray values of grayscale pixels are known quantities and are initially unchanged during conversion. We then construct the energy function between grayscale pixels and other pixels. Thereafter, we build the objective function with the linear combination of the three energy functions and the gray image is obtained by solving the objective function using the iterative method. Result Experimental results show that our algorithm can preserve salient structure and fine detail, as well ensure that the same color can be transformed to the same grayscale and that the gray value of the grayscale pixel would be unchanged after conversion. Conclusion Our algorithm conforms to the degree of perception about contrast change in image and can preserve detailed information and global structure. The algorithm can be applied to digital printing, pattern recognition, and so on, and thus has great application value.