Saturation-Value Blind Color Image Deblurring with Geometric Spatial-Feature Prior

Blind deblurring for color images has long been a challenging computer vision task.The intrinsic color structures within image channels have typically been disregarded in many excellent works.We investigate employing regularizations in the hue,saturation,and value(HSV)color space via the quaternion framework in order to better retain the internal relationship among the multiple channels and reduce color distortions and color artifacts.We observe that a geometric spatial-feature prior utilized in the intermediate latent image successfully enhances the kernel accuracy for the blind deblurring variational models,preserving the salient edges while decreasing the unfavorable structures.Motivated by this,we develop a saturation-value geometric spatial-feature prior in the HSV color space via the quaternion framework for blind color image deblurring,which facilitates blur kernel estimation.An alternating optimization strategy combined with a primal-dual projected gradient method can effectively solve this novel proposed model.Extensive experimental results show that our model outperforms state-of-the-art methods in blind color image deblurring by a wide margin,demonstrating the effectiveness of the proposed model.

Fast MRI Reconstruction via Edge Attention

Fast and accurate MRI reconstruction is a key concern in modern clinical practice.Recently,numerous Deep-Learning methods have been proposed for MRI reconstruction,however,they usually fail to reconstruct sharp details from the subsampled k-space data.To solve this problem,we propose a lightweight and accurate Edge Attention MRI Reconstruction Network(EAMRI)to reconstruct images with edge guidance.Specifically,we design an efficient Edge Prediction Network to directly predict accurate edges from the blurred image.Meanwhile,we propose a novel Edge Attention Module(EAM)to guide the image reconstruction utilizing the extracted edge priors,as inspired by the popular self-attention mechanism.EAM first projects the input image and edges into Q_(image),K_(edge),and V_(image),respectively.Then EAM pairs the Q_(image)with K_(edge)along the channel dimension,such that 1)it can search globally for the high-frequency image features that are activated by the edge priors;2)the overall computation burdens are largely reduced compared with the traditional spatial-wise attention.With the help of EAM,the predicted edge priors can effectively guide the model to reconstruct high-quality MR images with accurate edges.Extensive experiments show that our proposed EAMRI outperforms other methods with fewer parameters and can recover more accurate edges.

EFFICIENT BOX-CONSTRAINED TV-TYPE-l^1 ALGORITHMS FOR RESTORING IMAGES WITH IMPULSE NOISE

In this paper, we study the restoration of images simultaneously corrupted by blur and impulse noise via variational approach with a box constraint on the pixel values of an image. In the literature, the TV-l^1 variational model which contains a total variation （TV） regularization term and an l^1 data-fidelity term, has been proposed and developed. Several numerical methods have been studied and experimental results have shown that these methods lead to very promising results. However, these numerical methods are designed based on approximation or penalty approaches, and do not consider the box constraint. The addition of the box constraint makes the problem more difficult to handle. The main contribution of this paper is to develop numerical algorithms based on the derivation of exact total variation and the use of proximal operators. Both one-phase and two-phase methods are considered, and both TV and nonlocal TV versions are designed. The box constraint [0, 1] on the pixel values of an image can be efficiently handled by the proposed algorithms. The numerical experiments demonstrate that the proposed methods are efficient in computational time and effective in restoring images with impulse noise.

Alternating Minimization Method for Total Variation Based Wavelet Shrinkage Model

In this paper,we introduce a novel hybrid variational model which generalizes the classical total variation method and the wavelet shrinkage method.An alternating minimization direction algorithm is then employed.We also prove that it converges strongly to the minimizer of the proposed hybrid model.Finally,some numerical examples illustrate clearly that the new model outperforms the standard total variation method and wavelet shrinkage method as it recovers better image details and avoids the Gibbs oscillations.

VARIATIONAL IMAGE FUSION WITH FIRST AND SECOND-ORDER GRADIENT INFORMATION

Image fusion is important in computer vision where the main goal is to integrate several sources images of the same scene into a more informative image. In this paper, we propose a variational image fusion method based on the first and second-order gradient information. Firstly, we select the target first-order and second-order gradient information from the source images by a new and simple salience criterion. Then we build our model by requiring that the first-order and second-order gradient information of the fused image match with the target gradient information, and meanwhile the fused image is close to the source images. Theoretically, we can prove that our variational model has a unique minimizer. In the numerical implementation, we take use of the split Bregman method to get an efficient algorithm. Moreover, four-direction difference scheme is proposed to discrete gradient operator, which can dramatically enhance the fusion quality. A number of experiments and comparisons with some popular existing methods demonstrate that the proposed model is promising in various image fusion applications.

Incorporating the Maximum Entropy on the Mean Framework with Kernel Error for Robust Non-Blind Image Deblurring

Non-blind deblurring is crucial in image restoration.While most previous works assume that the exact blurring kernel is known,this is often not the case in prac-tice.The blurring kernel is most likely estimated by a blind deblurring method and is not error-free.In this work,we incorporate a kernel error term into an advanced non-blind deblurring method to recover the clear image with an inaccurately estimated kernel.Based on the celebrated principle of Maximum Entropy on the Mean(MEM),the regularization at the level of the probability distribution of images is carefully com-bined with the classical total variation regularizer at the level of image/kernel.Exten-sive experiments show clearly the effectiveness of the proposed method in the pres-ence of kernel error.As a traditional method,the proposed method is even better than some of the state-of-the-art deep-learning-based methods.We also demonstrate the potential of combining the MEM framework with classical regularization approaches in image deblurring,which is extremely inspiring for other related works.

The Convex Relaxation Method on Deconvolution Model with Multiplicative Noise

In this paper,we consider variational approaches to handle the multiplicative noise removal and deblurring problem.Based on rather reasonable physical blurring-noisy assumptions,we derive a new variational model for this issue.After the study of the basic properties,we propose to approximate it by a convex relaxation model which is a balance between the previous non-convex model and a convex model.The relaxed model is solved by an alternating minimization approach.Numerical examples are presented to illustrate the effectiveness and efficiency of the proposed method.

Low Rank and Total Variation Based Two-Phase Method for Image Deblurring with Salt-and-Pepper Impulse Noise

Although there are many effective methods for removing impulse noise in image restoration,there is still much room for improvement.In this paper,we propose a new two-phase method for solving such a problem,which combines the nuclear norm and the total variation regularization with box constraint.The popular alternating direction method of multipliers and the proximal alternating direction method of multipliers are employed to solve this problem.Compared with other algorithms,the obtained algorithm has an explicit solution at each step.Numerical experiments demonstrate that the proposed method performs better than the stateof-the-art methods in terms of both subjective and objective evaluations.