Seyed Hamid Safavi, Farah Torkamani-Azar
In this paper, we propose a novel CS approach in which the acquisition of non-visible information is also avoided.
Seyed Hamid Safavi, Farah Torkamani-Azar
Recently, multidimensional signal reconstruction using a low number of measurements is of great interest. Therefore, an effective sampling scheme which should acquire the most information of signal using a low number of measurements is required. In this paper, we study a novel cube-based method for sampling and reconstruction of multidimensional signals. First, inspired by the block-based compressive sensing (BCS), we divide a group of pictures (GoP) in a video sequence into cubes. By this way, we can easily store the measurement matrix and also easily can generate the sparsifying basis. The reconstruction process also can be done in parallel. Second, along with the Kronecker structure of the sampling matrix, we design a weight matrix based on the human visuality system, i.e. perceptually. We will also benefit from different weighted $\ell_1$-minimization methods for reconstruction. Furthermore, conventional methods for BCS consider an equal number of samples for all blocks. However, the sparsity order of blocks in natural images could be different and, therefore, a various number of samples could be required for their reconstruction. Motivated by this point, we will adaptively allocate the samples for each cube in a video sequence. Our aim is to show that our simple linear sampling approach can be competitive with the other state-of-the-art methods.
Rui Liu, Hossein Nejati, Seyed Hamid Safavi et al.
We propose an algorithm to uncover the intrinsic low-rank component of a high-dimensional, graph-smooth and grossly-corrupted dataset, under the situations that the underlying graph is unknown. Based on a model with a low-rank component plus a sparse perturbation, and an initial graph estimation, our proposed algorithm simultaneously learns the low-rank component and refines the graph. Our evaluations using synthetic and real brain imaging data in unsupervised and supervised classification tasks demonstrate encouraging performance.