基于低秩矩阵二元分解的快速显著性目标检测算法

近年来，基于矩阵低秩表示模型的图像显著性目标检测受到了广泛关注。在传统模型中通常对秩最小化问题进行凸松弛，即引入最小化核范数将原始输入图像分解为低秩矩阵和稀疏矩阵。但是，这种方法在每次迭代中必须执行矩阵奇异值分解（SVD），计算复杂度较高。为此，本文提出了一种低秩矩阵双因子分解和结构化稀疏矩阵分解联合优化模型，并应用于显著性目标检测。算法不仅利用低秩矩阵双因子分解和交替方向法（ADM）来降低时间开销，而且引入分层稀疏正则化刻画稀疏矩阵中元素之间的空间关系。此外，所提算法能够无缝集成高层先验知识指导矩阵分解过程。实验结果表明，提出模型和算法的检测性能优于当前主流无监督显著性目标检测算法，且具有较低的时间复杂度。

Salient Object Detection via Efficient Low-rank Matrix Bi-Factorization

In recent years, salient object detection via low-rank recovery models have received a significant amount of attention in the field of object detection. In traditional models, a matrix is generally decomposed into a low-rank matrix and a sparse matrix by a convex relaxation of the rank minimization problem, i.e., minimizing the nuclear norm. But, these methods suffer from high computation complexity, since singular value decomposition (SVD) has to be performed in each iteration. To solve this issue, an efficient low-rank matrix bi-factorization model was proposed for salient object detection, which not only took advantage of low-rank matrix bi-factorization and alternating direction method (ADM) to reduce the computation cost, but utilized structured-sparsity regularization to exploit the spatial relations between the elements in the sparse matrix. Furthermore, high-level priors were integrated to jointly guide the matrix decomposition. Experimental results on five challenging datasets validate the efficiency and effectiveness of the proposed method in terms of six performance metrics.