Blind Image Extraction by Using Local Smooth Information.
ABSTRACT This paper proposes a new approach to blind image extraction. By using the property that most of near pixels are smooth, blind image extraction is formulated to a generalized eigen-decomposition problem. The key point of our method is the formulation of matrix pencil. Since the image is two dimensional, the values of a small patch are smooth. Based on this observation, two mixed signals are formulated in columnwise order or in rowwise order respectively. The matrix pencil is constructed by using these two mixed signals. The separation weight vector can be obtained by generalized eigen-decomposition. Compared with the 'Non-Negative ICA' algorithm, the original signals in our algorithm are not required to be well-grounded, which means that they have a non-zero pdf in the region of zeros. In contrast to many second order methods in recent literatures, the two dimensional signals are used. Simulation results on mixed images are employed to further illustrate the advantages of our approach.
- SourceAvailable from: psu.edu[show abstract] [hide abstract]
ABSTRACT: Optimization of a cost function J(W) under an orthogonality constraint WW = I is a common requirement for ICA methods.07/2004;
- [show abstract] [hide abstract]
ABSTRACT: This paper presents a method of blind source separation that jointly exploits the nonstationarity and temporal structure of sources . The method needs only multiple time-delayed correlation matrices of the observation data, each of which is evaluated at different timewindowed data frame, to estimate the demixing matrix. We show that the method is quite robust with respect to the spatially correlated but temporally white noise. We also discuss the extension of some existing second-order blind source separation methods. Extensive numerical experiments confirm the validity of the proposed method.05/2002;
- [show abstract] [hide abstract]
ABSTRACT: In order to operate properly, the superresolution methods based on orthogonal subspace decomposition, such as multiple signal classification (MUSIC) or estimation of signal parameters by rotational invariance techniques (ESPRIT), need accurate estimation of the signal subspace dimension, that is, of the number of harmonic components that are superimposed and corrupted by noise. This estimation is particularly difficult when the S/N ratio is low and the statistical properties of the noise are unknown. Moreover, in some applications such as radar imagery, it is very important to avoid underestimation of the number of harmonic components which are associated to the target scattering centers. In this paper, we propose an effective method for the estimation of the signal subspace dimension which is able to operate against colored noise with performances superior to those exhibited by the classical information theoretic criteria of Akaike and Rissanen. The capabilities of the new method are demonstrated through computer simulations and it is proved that compared to three other methods it carries out the best trade-off from four points of view, S/N ratio in white noise, frequency band of colored noise, dynamic range of the harmonic component amplitudes, and computing time.EURASIP Journal on Advances in Signal Processing. 01/2004;