Hu He

University at Buffalo, The State University of New York, Buffalo, NY, United States

Are you Hu He?

Claim your profile

Publications (8)7.83 Total impact

  • Source
    Hu He, L.P. Kondi
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we propose an image super-resolution (resolution enhancement) algorithm that takes into account inaccurate estimates of the registration parameters and the point spread function. These inaccurate estimates, along with the additive Gaussian noise in the low-resolution (LR) image sequence, result in different noise level for each frame. In the proposed algorithm, the LR frames are adaptively weighted according to their reliability and the regularization parameter is simultaneously estimated. A translational motion model is assumed. The convergence property of the proposed algorithm is analyzed in detail. Our experimental results using both real and synthetic data show the effectiveness of the proposed algorithm.
    IEEE Transactions on Image Processing 04/2006; · 3.20 Impact Factor
  • Source
    Hu He, L.P. Kondi
    [Show abstract] [Hide abstract]
    ABSTRACT: In traditional digital image restoration, the blurring process of the optic is assumed known. Many previous research efforts have been trying to reconstruct the degraded image or video sequence with either partially known or totally unknown point spread function (PSF) of the optical lens, which is commonly called the blind deconvolution problem. Many methods have been proposed in the application to image restoration. However, little work has been done in the super-resolution scenario. In this paper, we propose a generalized framework of regularized image/video iterative blind deconvolution / super-resolution (IBD-SR) algorithm, using some information from the more matured blind deconvolution techniques from image restoration. The initial estimates for the image restoration help the iterative image/video super-resolution algorithm converge faster and be stable. Experimental results are presented and conclusions are drawn.
    Image Processing, 2005. ICIP 2005. IEEE International Conference on; 10/2005
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Image registration is a crucial part of the success of the super-resolution algorithms. In real applications, atmospheric turbulence is an important factor that brings further degradation to the low-resolution image sequence (video frames), besides other degradations such as global motion due to movement of the optical device, blurring due to the point spread function of the lens, and blurring due to the finite size of the detector array. In this paper, the degradation of the atmospheric turbulence to the low-resolution images is modeled as per-pixel motion in the high-resolution plane and is assumed to be spatially local and temporally quasi-periodic. The registration is a two-stage process: first, the global motion between frames is estimated using the phase-correlation method to remove "jitter" and stabilize the sequence; then, an optical flow method with quasi-periodic constraint is used to estimate the per-pixel motion. A threshold is used to separate the relatively larger object movement from per-pixel atmospheric turbulence. After registration, the shift map of each frame is obtained, along with a prototype of the high-resolution image. A maximum a posteriori (MAP) based super-resolution algorithm is therefore applied to reconstruct the high-resolution image. Experiments using synthetic images are conducted to verify the validity of the proposed method. Finally, conclusions are drawn.
    Proc SPIE 05/2005;
  • Source
    Hu He, L.P. Kondi
    [Show abstract] [Hide abstract]
    ABSTRACT: In many imaging systems, the resolution of the detector array of the camera is not sufficiently high for a particular application. Furthermore, the capturing process introduces additive noise and the point spread function of the lens and the effects of the finite size of the photo-detectors further degrade the acquired video frames. The goal of resolution enhancement is to estimate a high-resolution image from a sequence of low-resolution images while also compensating for the degradations. We propose a technique for image resolution enhancement with adaptively weighted low-resolution images (channels) and simultaneous estimation of the regularization parameter. The weight coefficients work as the cross-channel fidelity to each low-resolution image, while the regularization parameter acts as the within-channel balance between data and prior model for each channel. Experimental results are presented and conclusions are drawn.
    Acoustics, Speech, and Signal Processing, 2004. Proceedings. (ICASSP '04). IEEE International Conference on; 06/2004
  • Source
    Hu He, L.P. Kondi
    [Show abstract] [Hide abstract]
    ABSTRACT: The MAP (maximum a posteriori) -based resolution enhancement technique with Huber-Markov random field (HMRF) as the image prior has been proposed in the literature, and better preserves image discontinuities when compared with a Gaussian prior model. The reconstruction relies on the choice of Huber function parameter, or threshold T. There is no explicit selection of T in the previous studies. In this paper, we propose a method for choosing the threshold of the HMRF image prior in MAP based resolution enhancement. The method is based on the fact that the threshold T of the Huber function in the HMRF image priors is physically the trade-off between high-frequency components and low-frequency components for imagery data. High-pass filtering using the discrete Laplacian kernel along with the Huber function is used as the smoothness measure. When the high-passed value is less than T, the measure is a parabola function, while when the value is larger than T, the smoothness measure becomes a linear function. We hence define two different sets and derive the MAP estimator as a function of T Experimental results are presented and conclusions are drawn.
    Electrical and Computer Engineering, 2004. Canadian Conference on; 06/2004
  • [Show abstract] [Hide abstract]
    ABSTRACT: MAP (Maximum A Posteriori) -based resolution enhancement technique with Huber-Markov random field (HMRF) as the image prior has been proposed in the literature, and better preserves image discontinuities when compared with a Gaussian prior model. The reconstruction relies on the choice of Huber function parameter, or threshold T. There is no explicit selection of T in the previous studies. In this paper, we propose a method for choosing the threshold of the HMRF image prior in MAP based resolution enhancement. The method is based on the fact that the threshold T of the Huber function in the HMRF image priors is physically the trade-off between high-frequency components and lowfrequency components for imagery data. High-pass filtering using the discrete Laplacian kernel along with the Huber function is used as the smoothness measure. When the high-passed value is less than T, the measure is a parabola function, while when the value is larger than T, the smoothness measure becomes a linear function. We hence define two different sets and derive the MAP estimator as a function of T. Experimental results are presented and conclusions are drawn.
    05/2004;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we propose a resolution enhancement algorithm based on the Expectation Maximization (EM) frame-work. The objective of resolution enhancement (super-resolution) is to reconstruct a high-resolution image from a sequence of low-resolution (LR) images, under the assumption that there exists subpixel motion between the low-resolution frames. EM based image restoration has been studied in previous work, while its results can not be directly applied to the resolution enhancement scenario because of the subsampling (decimation) process in the acquisition model. This leads to a non-square degradation matrix which is not circulant as in the restoration case and hence cannot be diagonalized and operated on by the EM algorithm in the frequency domain. To overcome this difficulty, we propose to reorganize and interlace the low-resolution frames to construct an interlaced image using the registration parameters. This interlaced image is equivalent to a uniform blur process of the PSF blurred image. Now the resolution enhancement problem reduces to a restoration problem with two low-pass filters to deblur: one is the blur due to the point spread function (PSF) of the optical lens, and the other the uniform blur due to the decimation matrix. EM based restoration algorithm is thus computed efficiently in the frequency domain, considering the inaccurate estimate of the PSF and unknown power spectrum of both the high-resolution image and noise. Simulations using synthetic images are implemented to verify the proposed algorithm and conclusions are drawn.
    01/2004;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we extend our previous resolution enhancement results by proposing a technique for the estimation of the regularization parameter based on the assumption that it should satisfy the following properties: It should be a function of the regularized noise power of the data and its choice should yield a convex functional whose minimization would give the desired high-resolution image. Experimental results are presented and conclusions are drawn.
    Proc SPIE 01/2004; 13:586-596.