基于条件随机场的显著性检测算法

Saliency detection based on conditional random field

近年来,显著性检测与图像处理有着密不可分的关系,图像处理依赖于高质量的显著图才能得到较好的处理结果。因此为提高图像显著性检测的准确性,提出了一种新的基于条件随机场(CRF)的显著性融合算法。将显著性检测看做一个图像标注问题,运用多尺度对比,中央—周围直方图和颜色空间分布这三种不同的显著度计算得到显著图。通过CRF学习计算各个显著度的权重,采用最大似然估计方法获取模型参数估计,得到最优解。最后利用CRF检测测试图像。通过大量的实验结果表明,此算法可以对显著目标得到更加精确地检测。

In recent years, significant test has an inseparable relationship with the image processing,image processing technology depends on the quality of the saliency map to obitain better results. So to improve the accuracy of image saliency detection, this paper proposes a new saliency fusion algorithm based on conditional random field （CRF）. Saliency detection as an image annotation problem, using the multi-scale contrast, center-around histogram and color spatial distribution of the three different degrees calculated saliency map. Through the study of CRF to calculate weights of all the saliency degree, using maximum likelihood estimation method for model parameter estimation, gets the optimal solution. Finally using the CRF detection tests the images. A lot of the experimental results show that the algorithm can get more accurate detection of saliency goals.