基于自适应聚类和线性规划最优选点的彩色图像低成本打印方法

StippliNg Generation Method Based on Super Pixel and Color Knapsack Algorithm

彩色点画是一种从视觉上由大量小像素点构建图像的艺术技术,像素个数的多少直接影响着构图的成本。其优化选点构图方法为实现低成本打印提供了一个重要的方式。目前,点画生成存在着多通道采样点难以均匀分布,颜色层次难以兼顾等难点,并耗费大量的计算成本。对此,提出了一种基于超像素自适应聚类和线性规划最优选点的彩色点画生成方法,该方法在初步超像素划分图像的基础上,使用基于颜色密度峰值的自适应聚类方法得到最佳聚类个数,并进一步划分子聚类,然后根据每个子聚类的颜色均值作为子聚类内部选点的最佳间隔距离,在选点的同时依据SSIM指标,建立目标优化模型,通过数学优化器Gurobi实现模型选点,使点保留最少个数的目标基础上,同时保持聚类内部分布均匀和颜色渐变层次,以提高所生成的点画图像的可视化效果。实验结果表明,本文算法极大地降低了像素个数并在生成的点画的平均结构相似性(mean structural similarity index measure,SSIM)、峰值信噪比(peak signal to noise ratio,PSNR)等评价指标方面均优于当前方法。

Color dot painting is an art technique that visually constructs an image from a large number of small pixels,and the number of pixels directly affects the cost of composition.The optimized point selection and composition method provides an important way to achieve low-cost printing.At present,there are difficulties in point drawing generation,such as difficulty in evenly distributing multi-channel sampling points,difficulty in balancing color levels,and significant computational costs.This article proposes a method for generating color dots based on superpixel adaptive clustering and linear programming optimal points.This method uses an adaptive clustering method based on color density peak to obtain the optimal number of clusters on the basis of preliminary superpixel image segmentation,and further divides into sub clusters.Then,the color mean of each sub cluster is used as the optimal interval distance for selecting points within the sub clusters.At the same time,based on the SSIM(structural similarity index measure)index,a target optimization model was established.The mathematical optimizer Gurobi was used to achieve model point selection,which preserves the minimum number of points while maintaining uniform distribution and color gradient levels within the clusters,in order to improve the visualization effect of the generated dot image.The experimental results show that the algorithm proposed in this paper greatly reduces the number of pixels and outperforms the current method in terms of evaluation metrics such as mean SSIM and PSNR(peak signal to noise ratio)of generated dots.