基于元胞自动机的低秩加权协同显著性检测

Low Rank Weighted Co-saliency Detection Via Cellular Automata

针对图像协同显著性检测问题,提出了一种基于元胞自动机的低秩加权检测方法.利用3个显著性检测算法生成多个显著图,结合超像素分割提取协同显著性区域.在协同显著区域的特征矩阵上施加秩约束,以获得稀疏误差矩阵,基于稀疏误差矩阵为每个显著图计算自适应权重.将显著图与其对应的权重相乘以获得融合的协同显著图,利用元胞自动机更新显著性区域.在MSRC和i Coseg数据集上的实验结果表明,相对于其他的算法,本文算法可以在有效地突出显著目标的同时抑制背景噪声.

Aiming at the problem of image co-saliency detection,a low rank weighted detection method based on cellular automata is proposed. Generate three saliency maps using multiple saliency detection algorithms,and then combined with superpixel segmentation to extract co-saliency regions. Then,the rank constraint is applied to the feature matrix of the co-saliency region to obtain the sparse error matrix,and the adaptive weights is computed for each saliency map based on the sparse error matrix. The final co-saliency map is obtained by multiplying the saliency map by its corresponding weight. Finally,the cellular automaton is utilized to update the salient regions. The experimental results on the MSRC and iCoseg datasets show that compared to other algorithms,the proposed algorithm can effectively suppress the background noise while highlighting the salient object.