NON-CONVEX LOW-RANK APPROXIMATION FOR HYPERSPECTRAL IMAGE
RECOVERY WITH WEIGHTED TOTAL VARAITION REGULARIZATION
Hanyang Li
1
, Peipei Sun
1
, Hongyi Liu
1
, Zebin Wu
2
, Zhihui Wei
2
1
School of Science, Nanjing University of Science and Technology, Nanjing, 210094, P.R.China
2
School of Computer Science and Engineering, Nanjing University of Science and Technology,
Nanjing, 210094, P.R.China
ABSTRACT
Low-rank representation has been widely used as a powerful
tool in hyperspectral image (HSI) recovery. The existing
studies involving low-rank problems are commonly under
the nuclear norm penalization. However, nuclear norm
minimization tends to over-shrink the components of rank,
which leads to modeling bias. In this paper, a new non-
convex penalty is introduced to obtain an unbiased low-rank
approximation. In Addition, local spatial neighborhood
weighted spectral-spatial total variation (TV) regularization
is introduced to preserve spatial structural information. And
sparse
-norm is used as a constraint to sparse noise.
Finally, a novel HSI non-convex low-rank relaxation
restoration model is proposed. A number of experiments
show that the proposed method can effectively remove the
mixed-noise, and result in an unbiased estimate with better
robustness.
Index Terms— Hyperspectral image(HSI), non-convex
relaxation, low- rank representation, total variation(TV).
1. INTRODUCTION
Hyperspectral image (HSI) captures significant information
regarding the Earth’s surface in hundreds of continuous
bands, which has great prospect for applications. However,
HSI is often contaminated by various noises, which
degrades image quality and affects the subsequent
processing. Therefore, HSI restoration is one of the most
important research topics [1-2].
Amongst various HSI restoration models, low-rank
representation (LRR), aims at learning low-dimensional
property of clean HSI, widely plays a role as the premise of
many HSI restoration methods. With an assumption that the
spectral of HSI lied in a low-rank subspace and spatial
spaces were to be piecewise smooth, a HSI restoration meth-
This work is supported in part by the National Natural Science
Foundation of China (61301215, 61471199, 61772274, 61701238,
11431015, 91538108); in part by the Fundamental Research Funds for the
Central Universities under Grant 30917015104; and in part by the Jiangsu
Provincial Natural Science Foundation of China under Grant BK20170858.
od based on low-rank matrix recovery (LRMR) was
proposed in [3]. Encouraged by the powerfulness of total
variation (TV) regularization in various image restoration
tasks, Zhang et al. integrated TV regularization in low-rank
matrix recovery (LRTV) model [4], and as a result enhanced
the capability of the LRR technique for HSI restoration.
Despite the effectiveness of LRTV, it is still a challenge as
for the definition of hyperspectral TV.
Typically, one of the most representative low-rank
approximations is nuclear norm. However, as it is convex,
the recovery performance will degrade in the presence of
measurement noise, and the solution can seriously deviate
from the original solution. Then many non-convex
approaches for low-rank approximations have been
proposed [5-10]. Including weighted nuclear norm
minimization (WNNM) [6], Schatten p-norm minimization
[7], weighted Schatten p-norm minimization (WSNM) [8]
and minimax concave penalty (MCP) function [9]. In [10],
Wang et al. developed MCP function to rank-minimization
problems and obtain a so-called γ-norm as a non-convex
relaxation of the matrix rank.
By combining the non-convex low-rank relaxation
penalty with weighted TV, we propose a novel HSI
restoration model. In the new model, the non-convex matrix
γ-norm is utilized to separate the clean HSI from mixed-
noise, and weighted TV, in which the spectral-spatial
gradient and weight function are computed in the local
spatial neighborhood, is developed to capture the spatial
piecewise smooth structure. Furthermore, the Alternating
Direction Method of Multipliers (ADMM) [11] iteration is
adopted to solve the resulting minimization problem.
2. PROPOSED HSI RECOVERY MEHTOD
2.1. Weighted HSI TV
It is known that TV norm can significantly preserve the
spatial structural information. Different TV, such as higher-
order TV, weighted TV, anisotropic TV are widely applied
for specific applications. For a 3-D HSI data
, weighted
hyperspectral TV (WHTV) is defined as follows:
(1)