基于自适应LARK特征的图像显著性检测算法

Saliency Detection Based on Adaptive LARK Features

为了有效检测自然场景中的显著区域,提出一种简单高效的基于自适应LARK特征的图像显著性检测方法。首先自适应选取若干个有效的LARK特征分量,然后计算基于该特征的像素显著性值。为进一步增强图像像素的显著性,通过经典的超像素分割方法计算图像的超像素颜色奇异性值。最后将这两者线性融合,形成自然场景中像素的最终显著性值。在国际通用数据集上测试,结果表明,该方法优于其他视觉显著区域检测计算方法,并且可以产生均匀突出的显著性图谱,在正确率和召回率上都有明显提高。

A simple and efficient salient region detection method based on adaptive LARK features is proposed to effectively detect the salient region in the natural scene.Firstly,certain effective LARK features are adaptively selected,and then the salient values are computed based on these features.To further enhance the saliency of the image,color uniqueness values of the image are computed by the classical super pixel segment method.Finally,the final salient map is obtained by linearly fusing both feature values.Experimental results on the largest publicly available data sets demonstrate that the proposed method outperforms the state-of-the-art detection methods and can generate the uniformly highlighting salient region.Furthermore,the precision and recall rates are obviously improved.