Patch-Based Nonlocal Adaptive Gradient Regularization for Image Restoration in Sensor Networks

Mingzhu Shi

doi:10.1109/access.2019.2958941

Abstract

In this paper, we propose a novel patch-based adaptive nonlocal gradient regularization method for image restoration in sensor networks. It formulates the hyper-Laplacian distribution to regularize the global gradient distribution. The patch-based nonlocal gradient prior is utilized to regularize the nonlocal self-similarity of image gradients. Firstly, the L0-norm smoothing scheme is used innovatively as the preprocessing step to preserve strong edges, which are critical to improve the accuracy of clustering the similar image patches. Then, adaptive weights for each patches are developed from a set of clustered nonlocal self-similarity patches by learning the the expectation and variance for sparse gradient distribution at each pixel. Comparing with several recent state-of-the-art methods, experimental results show that the proposed method has better performance in alleviating block effects and preserving image details.

Highlights

With the rapid development of wireless sensor networks, there is an increasing demand for the quality of signal transmission, especially for the two-dimensional images, which are inevitably degraded in the process of image acquisition, transmission and processing
For the global regularization, the hyper-laplacian parameter p is determined from a large number of image statistics and is set to 0.5 based on research experience
In this paper, a patch-based adaptive nonlocal gradient regularization method is proposed for image restoration in sensor networks

Summary

Introduction

With the rapid development of wireless sensor networks, there is an increasing demand for the quality of signal transmission, especially for the two-dimensional images, which are inevitably degraded in the process of image acquisition, transmission and processing. In the spatially invariant system, the imaging process is often formulated as a common model y = H x + n. Y is the degraded image, x is the desired image and n is the additive Gaussian white noise with zero mean. H denotes the degraded process, which is the discrete point spread function (PSF) and usually modeled as a blurring matrix. The recovered results may be discontinuous because of observation errors, which lead to the high ill-posed problem. Regularizing such ill-posed problem is critical to obtain a stable solution and produce a desire image

Objectives

Results

Conclusion