摘要
针对人体肤色在显著性检测过程中容易被忽视,影响后期图像人体分割效果这一问题,本文提出了一种改进的显著性检测方法.首先利用超像素图分割法对图像进行分割,利用人脸检测提取肤色信息,接着将肤色信息融入到颜色独特性和颜色空间分布计算中,最后将得到的超像素显著值分配给每个像素,生成像素级显著图.在公开的数据集上选取100幅包含人物的图片进行测试,该算法取得最高95%的精确度,优于其他显著性算法,在图像人体分割实验中,可以很好地把人体和背景分开.实验结果表明,本文方法在精确率、召回率和F-测量上都优于其他方法,并且能很好地分割人体图像.
Aiming at the problem that human skin is easily overlooked in salient detection process and it will affect the later human body segmentation,we propose an improved salient detection method. Firstly,the algorithm uses the superpixel segmentation method for image segmentation and uses face detection to extract skin color information. Secondly,the skin color information will be integrated into both color uniqueness and color distribution. Finally,we assign the superpixel saliency values to the input image to get a pixel-accurate saliency map. When evaluated using 100 people pictures from publicly available datasets,our method outperforms several existing salient object detection methods with an achieved accuracy of 95%. Through the human body segmentation experiment,the human body can be well separated from the background. The experimental result shows our method is superior to other methods on precision,recall rates and F-measure,and can be a good human body segmentation.
出处
《小型微型计算机系统》
CSCD
北大核心
2016年第3期608-611,共4页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61202267)资助
关键词
图像人体分割
显著性检测
人脸检测
图像分割
F-测量
human body segmentation
salient detection
face detection
image segmentation
F-measure
