------ Links: ------
A. Hashemi, Ngai-Man Cheung, Tony Q. S. Quek,
“Efficient Compressed Sensing for Diffusion
Fields”, preparing to submit to IEEE Transaction on Image
Many natural world environmental signals are spatiotemporal sources, where the signal depends on spatial location and changes over the time, e.g. temperature of a piece of land, an electromagnetic field, or reflection of light from a surface. Diffusion fields are an important subclass of such sources, where the source field satisfies diffusion partial differential equation (PDE). Sensor networks have been used as typical type of sampling devices to measure and reconstruct such sources for monitoring or automation purposes. Despite many benefits, sensor networks have their own deficiencies. First, there is a limit on the number of sensing units that can be deployed which limits the spatial sensing resolution. Moreover, battery life of a sensor node limits the power consumption and the temporal sampling rate. In this paper, the above limitations are tackled by means of compressed sensing (CS). We take advantage of the instinct property of such sources, i.e. satisfying a PDE, as side information to increase efficiency, we improve the reconstruction results by incorporating the side information from diffusion PDE into CS recovery. We demonstrate why diffusion compressed sensing (DCS) outperforms classic CS by treating PDE as a second source of knowledge in addition to sparsity. Experimental results are provided to demonstrate the effectiveness and usefulness of the proposed method. It is shown that DCS results in substantial data savings while producing estimates of higher accuracy, as compared to CS-base estimates.
Hashemi, M. Rostami, Ngai-Man Cheung, “2D
Diffusive Compressed Sensing for Solving Inverse Problems in Noisy
Environment”, preparing to submit to IEEE
Transaction on Signal Processing.|
The reconstruction of a diffusion ﬁeld using samples collected by a sensor network along with estimation the parameters of sources which induce the aforementioned field is a classical inverse problem with applications including temperature monitoring, pollution dispersion, EEG source localization and CPU thermal mapping. A classic work considers source models in noise free setting and use compressed sensing to recover the initial source distribution. While experiments demonstrate usefulness of the approach, it cannot handle noisy situations. Here, we consider uniform spatial sampling in a noisy environment setting and improve the classical approach by employing an extended version of CS which is suitable for this problem. Through experiments we demonstrate effectiveness of our approach.
Hashemi, R. Nikbakht “Reduced-Complexity Compressed
Spectrum Sensing in Low SNR Regimes Using Direct 2D Spectral
Correlation Function Reconstruction” preparing to submit to
IET Journal of Wireless Communication.|
In this paper, we explore a reduced-complexity compressive detection method which exploits the sparsity of the two-dimensional spectral correlation function (SCF). Due to the additional sparsity introduced in the SCF with respect to the power spectral density, sparsity pattern of 2D cyclic spectrum are shown to be remain even in low SNR scenarios. Utilizing this feature, we propose a detection method using entropy as the sparsity measure and working quit well in low SNR regimes. Furthermore, in compared to other trivial reconstruction approaches using the kronecker product in order to convert the 2D reconstruction problem in a 1D form, we present an algorithm which directly reconstructs the 2D cyclic spectrum without any vectorize operation. The results show its superiority with respect to tremendous amount of memory and computational cost in other approaches, which reduce the complexity by one order.
A.Hashemi, H. Nejati, Ngai-Man Cheung, H. Sosa, C.
I.; “DeepCAPTCHA: An Image CAPTCHA Based on Depth
Perception”, preparing to submit to IEEE Transaction on
Information Forensics and Security. |
In this paper, an adaptive image contrast enhancement algorithm based on an optimization problem in two dimensional histogram domain is presented. To reduce the unwanted effects of the histogram adjustment, through this optimization, similar to other common approach in this literature we find the 2D histogram of enhanced image in close proximity to input image histogram and uniform distribution, simultaneously. In addition, in contrast to other algorithms, by adaptive adjusting the components of a weight matrix, local information is counted. Experimental results on a wide range of images demonstrate the improved performance of the proposed method. Besides, applying the proposed method on variety of images results in 75% and 3% improvement in AMBEN and DEN measurements, respectively comparing to the reference method.
Hashemi, M. Rostami, Ngai-Man Cheung,
“Efficient Compressed Sensing Scheme for Reconstruction of
Spatiotemporal Field Arrays", submitted to IEEE
International Conference on Image Processing (ICIP) 2016. |
Diffusive Compressed Sensing for Thermal Monitoring 
Diffusion fields are an important subclass of spatiotemporal sources with a wide area of application. Sensor Networks are generally used for sampling and reconstruction of these fields for monitoring, surveillance, and automation purpose. Despite many benefits over traditional devices, sensor networks have their own limits. First, there is a limit on the number of sensing nodes considering the spatial sensing resolution. In addition, restrictions on power consumption of nodes limit the temporal sampling rate. We tackle these problems for the case of diffusion fields by using compressed sensing (CS) to process the collected samples more efficiently. In particular, we propose a Diffusive Compressive Sensing (DCS) framework to benefit from domain knowledge about diffusion fields. Experimental results are provided to demonstrate the effectiveness and efficiency of the proposed algorithm.
|2||A. Hashemi, M. Rostami, Ngai-Man
Cheung, “Efficient Environmental Temperature Monitoring
Using Compressed Sensing”, accepted to Data
Compression Conference (DCC) 2016, IEEE Signal Processing
Wireless Sensor Networks (WSN) have been used to collect data for environmental monitoring of physical quantities including temperature, humidity, and pressure. A WSN is composed of a network of sensing units which are deployed in the environment to collect local samples of the quantity of interest and then their outputs are fused for monitoring purpose on a planner area. Although WSN’s outperform traditional monitoring methods in terms of financial cost, they have their own design and resource constraints. First, although the sensing units are generally inexpensive but there is always a limit on the number of sensing units that can be used which depends on properties of the environment. On the other hand, resource constraints are imposed by power consumption of sensing nodes which limits the time sampling rate of WSN given battery life of sensing nodes. In this paper, we aim to use compressed sensing (CS) in order to tackles these problems for monitoring temperature by processing the collected temperature data more efficiently in a compressed domain. Moreover, we exploit an intrinsic property of diffusion fields as side information to improve the results. Experimental results are provided to demonstrate the effectiveness and usefulness the algorithm.
Hashemi, M. Rostami, Ngai-Man Cheung, “Source
Localization in Diffusion Field with Noisy
Measurements", preparing to submit to 42st IEEE
International Conference on Acoustics, Speech and Signal Processing,
ICASSP 2017. |
We study the problem of reconstructing unknown generating point sources of a diffusion field through spatiotemporal samples of the field, collected by a wireless sensor network. Our focus is to tackle this problem in noisy environments where a wireless sensor networks (WSN) is used for collecting samples. A typical problem with WSN is energy consumption of sensor nodes and limits on temporal sampling rate. These issues could increase ill-posedness of the problem. Compressed sensing (CS) is a technique that has been used in the literature to tackle these issues. However, the problem becomes very challenging with noisy measurements. In this paper, we propose to solve an additional inverse problem first before the source localization to reduce the ill-posedness of the localization problem. We take advantage of the signal structure governed by the diffusion field to increase the resolution of the data. To integrate the diffusion field prior, we employ a diffusive compressed sensing (DCS) algorithm in a uniform spatial sampling setting. We show that our proposed approach can significantly reduce the illposedness of the problem by decreasing the condition number of sampling operator. Experimental results demonstrate effectiveness and usefulness of our approach, and its competency over previous work in the literature.
|4||R. Nikbakht, H.
Aghayinia, A. Hashemi, M. Kazemi,
“Cyclostationary Features-Based Wideband Compressed Spectrum
Sensing for Cognitive Radio Networks” accepted in 2n
International Conference on Electrical, Mechanical, Computer and
Robust spectrum sensing is one of the most important parts of cognitive radio networks. The more bandwidth is available, the more spectrum opportunities are provided for secondary users. However, regarding the Nyquist sampling rate, the conventional algorithms for spectrum sensing which has been proposed so far, encounter some fundamental challenges such as expensive and costly computational implementation. In addition, the experiments show that a good portion of the frequency range is idle due to spectrum under-utilization, suggesting that the frequency spectrum of signal is highly sparse in Fourier domain. Therefore, compressed sensing theory can be used for reconstructing wideband spectrum. Since sparse reconstruction methods have poor performance in low SNRs, we cannot utilize the compressed sensing approach directly.
In order to solve this problem, direct reconstruction of cyclo-stationary features from compressed measurements has been proposed. In this paper, however, we modify the conventional wideband spectrum sensing to accommodate cyclostationary signals and formulate the problem in a simpler matrix form. Then, we convert the problem to the compressed sensing setting. According to simulations, the proposed scheme demonstrates good results even in low SNR regimes.
|5||R. Nikbakht, H.
Aghaeinia, M. Kazemi, A. Hashemi, “Wideband
Compressed Sensing based on 2D Sparse Reconstruction of Asymmetric
Cyclic Spectrum” accepted in 2n International Conference on
Electrical, Mechanical, Computer and Mechatronics
Robust wideband spectrum sensing is one of the most important parts of the cognitive radio networks. Regarding the Nyquist sampling rate challenges at wideband sensing and considering promising sparsity of frequency spectrum due to spectrum under-utilization, wideband spectrum can be reconstructed using compressed sensing methods. However, since sparse reconstruction methods have poor performance in low SNRs due to loss of sparsity property (e.g. due to increase in the noise power), we cannot utilize the compressed sensing approach directly. To address this problem efficiently, we here propose direct 2D sparse reconstruction of cyclostationary features from compressed measurements. In our approach we first reformulate obtaining the cyclic spectrum function in term of matrix form and then present an algorithm which directly reconstructs the 2D cyclic spectrum without any vector operation. This results in computationally efficiency, and significant reduction in the complexity, compared to previous methods that use Kronecker product method. In addition, we use blind detection method using eigenvalues of reconstructed cyclic spectrum. According to simulations, the proposed scheme demonstrates good results in noise uncertainty.