Search

Superpixels can capture redundancy in the image and greatly reduce the complexity of subsequent image processing tasks. Any segmentation algorithm can extract superpixels, however, most of them produce highly irregular superpixels, and the number of superpixels is not easily decided. To solve this problem, a simple and linear complex algorithm is proposed to generate superpixels based on geodesic distance transform, which can generate desired number of superpixels with approximately uniform size and shape. Experimental results illustrate that the method proposed has the ability to generate superpixels with different complex images and results in exact and robust effects. [ABSTRACT FROM AUTHOR]

Medical image fusion can integrate the information of two different modal images, which can provide doctors with accurate diagnosis and treatment. The image features are extracted and fused by sparse representation. Firstly, all source images are combined into a joint-matrix. The over complete dictionary can be trained by K-singular value decomposition (K-SVD) algorithm and the sparse codes can be acquired by joint-matrix. Secondly, the sparse codes which are considered as image features are combined with the choosing max fusion rule. Finally, the fused image is reconstructed from the combined sparse codes and the over complete dictionary. Compared with three state-of-the-art algorithms, the results show that the proposed method has better fusion performance in both noiseless and noisy situations. [ABSTRACT FROM AUTHOR]

Three main factors which affect the accuracy of time delay estimation (TDE) for narrowband radio frequency (RE) signals are mentioned: intermediate frequency difference, signal bandwidth and SNR. A phase spectrum TDE method based on frequency difference compensation (FDC) is proposed. The intermediate frequencies (IF) of both received signals are estimated respectively using frequency doubling method, and FDC method is used to eliminate the side effect of IF difference to TDE result. Then, chirp Z transform (CZT) is applied to increasing the available frequency bins within a limited bandwidth. Meanwhile, Kalman filter is also adopted because of its excellent anti-noise characteristics to improve precision of TDE. Both theoretical analyses and computer simulation results show the effectiveness and advantage of the proposed method. [ABSTRACT FROM AUTHOR]

Based on the fractional lower order statistics (FLOS) and stochastic resonance (SR) technique, a novel robust time delay estimation approach, stochastic resonance covariation time delay estimation (SRCTDE) algorithm is proposed to detect the latency delay in the evoked potential (EP) with the presence of the a-stable noise under the lower mixed-signal-to-noise ratio (MSNR) condition. In the SRCTDE, MSNRs of the reference and detected EP signals are improved via the SR effect between noisy EP signals and bistable nonlinear dynamical systems. The SR effect is realized by tuning system parameters to their optimal values under the maximum symmetric covariation coefficient criterion. Then, the covariation function between outputs of two SR systems is computed. The EP latency delay is estimated by obtaining the location of this covariation function peak. The performance of the SRCTDE algorithm is verified through computer simulations compared with the correlation algorithm and covariation algorithm for various values of MSNR. The SRCTDE algorithm gives an improved approach to estimation of the latency delay in the weak EP with α-stable noises. [ABSTRACT FROM AUTHOR]

Non-Gaussian and non-stationary random signal processing is an interesting and important topic in the field of signal processing, and it has important academic significance and practical value. Since the conventional algorithms based on the recursive least square (RLS) algorithm degenerate severely under the α-stable distribution noise conditions, an autoregressive (AR) model with time-varying coefficients is used to track the non-stationary characteristics of the signal, and its coefficients are estimated with the least p-norm algorithm (LPN). The experimental results show that the LPN algorithm is applicable not only under Gauss noise but also under the conditions of α-stable distribution noise. Furthermore, the LPN algorithm has better flexibility compared with RLS algorithm which is applicable only under Gauss noise. [ABSTRACT FROM AUTHOR]

In the application of essential Elman network, the Kalman filter algorithm has high convergence speed but low precision; on the contrary, the BP algorithm with gradient descend technique has high precision to map the output with input nonlinearly, but usually with low convergence speed at extreme point. Therefore, a new algorithm based on Kalman filter is proposed to accomplish essential Elman network training. The advantages of Kalman filter and BP algorithm are combined to optimize the new algorithm. The outputs of the hidden layer of the essential Elman network are set as the state varibles of a nonlinear system. Kalman filter algorithm is used for rapid tracking of the state variables and BP algorithm is used for more precise weights modification. Furthermore, some new learning information is added to improve, the precision of the results during the learning process. The validity of the new algorithm is supported by the simulation results. [ABSTRACT FROM AUTHOR]

In order to eliminate the effect of artifacts in electroencephalograph (EEG) signals, such as electrocardiogram (ECG) and electrooculogram (EOG), etc. , a new ICA (independent component analysis) method with multi-references is proposed based on the fixed-point algorithm and the ICA algorithm with references. After the proportions of artifacts in observation signals are calculated respectively, the comparatively pure EEGs are separated from the artifacts. Compared with fast ICA, the new method has less computational complexity and doesn't need to dispose the independent sources. In addition, it can overcome the defect of ICA-R algorithm which just extracts single signal. Computer simulation demonstrates that the new method can effectively remove several artifacts from brain signals. [ABSTRACT FROM AUTHOR]

Bandwidth of non-Gaussian signals and noises in practical applications is narrow. If simply using correlation algorithm in traditional Gaussian model to estimate the multipath time delays, peak values will overlap each other. So it will bring on large estimate errors. To solve this problem, according to the characteristics of signal and noise and the theory of the fractional lower order statistics, a super resolution multipath time delay estimation algorithm referred to P-EM is proposed. It is based on the EM (expectation-maximization) algorithm using the pth-order correlation and can be applied under α-stable distribution noise. It is shown that the proposed approach is robust through the theoretical analysis and the computer simulations. [ABSTRACT FROM AUTHOR]

Since there exists non-Gaussian impulsive noise in real environment, a-stable distribution is taken as the noise model. And under this assumption, the problem of multi-source time delay estimation in passive systems is studied. Considering the effect of a-stable distributed noises for the classical second order statistics, combining fractional lower order statistics theory and chirp Z transformation (CZT), a novel multi-source time delay estimation method in impulsive noise environment is proposed. Simulations show that the proposed method is a high resolution method suited for a-stable distributed noises condition and its performance is better than that of the common method named covariation method. [ABSTRACT FROM AUTHOR]