融合对比度与背景先验的显著目标检测算法

发布时间：2018-11-09 10:58

【摘要】：针对已有的显著检测算法对背景复杂的图像检测效果较差的问题,提出融合对比度与背景先验的显著目标检测算法.首先将图像划分为感知均匀的像素块,再根据对比度先验定义图像的显著边缘、像素块的全局对比度及颜色相似像素块的空间分布,得到任一像素块与前景的相关性;然后根据背景先验将图像边界像素块定义为伪背景区域,通过计算像素块与伪背景区域的相似度得到像素块与背景的相关性;最后通过能量优化函数结合像素块与前景、背景的相关性,得到该像素块的显著值.实验结果表明,与同类算法相比,该算法能更好地使显著目标整体高亮,抑制背景噪声,得到较符合视觉感知的显著图.
[Abstract]:Aiming at the problem of poor performance of existing salient detection algorithms for images with complex background, a significant target detection algorithm based on fusion contrast and background priori is proposed. Firstly, the image is divided into perceptual uniform pixel blocks, then the significant edges of the image, the global contrast of the pixel blocks and the spatial distribution of the color similar pixel blocks are defined according to the contrast prior, and the correlation between any pixel block and the foreground is obtained. Then the image boundary pixel block is defined as pseudo background region according to background priori, and the correlation between pixel block and background is obtained by calculating the similarity between pixel block and pseudo background region. Finally, the significant value of the pixel block is obtained by combining the energy optimization function with the correlation between the pixel block and the foreground and background. The experimental results show that compared with the similar algorithms, the proposed algorithm can better highlight the whole salient target, suppress the background noise, and obtain the salient image which accords with the visual perception.
【作者单位】：上海交通大学计算机科学与工程系;东南大学信息科学与工程学院;中兴通讯股份有限公司云计算及IT研究院;
【基金】：国家“九七三”重点基础研究发展计划项目(2011CB302203) 国家自然科学基金(61133009;61502220;U1304616)
【分类号】：TP391.41

【相似文献】