摘要
自然图像抠图效果取决于待分离图像的α掩图的精度。基于改进的人工免疫调节网络的自然抠图算法是为未知区域中的象素点定义一个即反映颜色适应度也反映空间信息的权值函数,利用人工免疫网络在前景和背景的全局范围内搜索,以得到该像素点对应的最佳的样本对(F,B),从而得到α掩图。通过实验与其他算法比较,该算法对用户输入的依赖较少,对复杂背景图像的抠图精度高,稳定性好,且对于细长头发丝区域及毛发类的边界区域的划分特别有效。
The efficience of natural image matting depends on alpha matte of the image. The natural image matting algorithm based on the improved artificial immune networks defines a weight function for each pixel combining the color samples information and the space distance information, searches the best sample pairs on the known foreground area and known background area, thereby receive the alpha matte. The experiment result compared to other algorithms proves that the algorithm proposed reduce the dependence of the matte effect on the user input information, can attain a good accurate and stability matte to the matting of complex images, especially efficiency for the long thin hair area and the boundary of the area of hair.
出处
《现代计算机》
2013年第5期35-39,共5页
Modern Computer
关键词
自然抠图算法
人工免疫调节网络
权值函数
α掩图
Natural Image Matting
Artificial Immune Networks
Weight Function
Alpha Matte
