IEEE Consumer Electronics Magazine - March/April 2023 - 19

FIGURE 5. Histogram of the smooth layer.
FIGURE 6. Histogram of the texture layer.
The smooth layer can be understood as the
The ideal smoothing layer Fsðx; yÞ should
remain the similar geometric shapes of the original
image, whiling removing the texture structure
as much as possible. Although the texture
structures of different images vary with the contents
of the images, the image generally has
more high-frequency information and a corresponding
larger gradient in the texture area. The
solution of smooth layer Fsðx; yÞ is transformed
into a Bayesian estimation problem based on
gradient regularization constraint
Fbs x; yðÞ¼ arg min ðÞ2
XM
x¼0
XN
y¼0
þ b rFs x; yðÞ
jj1:
(9)
The first term on the right of (9) is the data-fitting
term with the function of keeping the geometric
structure of the smoothing layer consistent
with the original image, jrFsðx; yÞj1 is the L1-norm
regularization of image gradient rFsðx; yÞ. Similar
to the solution of SR reconstruction, the solution
of
Fx; yðÞFs x; yðÞ
original image without texture, its distribution of
gray values is basically consistent with the original
image. To avoid excessive enhancement,
brightness-preserving histogram equalization
algorithm39 is used for smooth layer. The core
idea of the algorithm is that the histogram is
divided into two regions with the average gray
level as the dividing line, and each region is equalized
separately, as shown in Figure. 5. The gray
level of the smooth layer is basically kept at the
median after enhancement, which effectively
avoids excessive enhancement.
2) Equalization of the texture layer
The histogram distribution of the texture
F^sðx; yÞ is also an optimization problem. The
steepest descent method are applied to determine
the optimal estimation value of smoothing layer.
Since the image Fðx; yÞ is known, the texture
layer Ftðx; yÞ can be easily solved by subtraction
once ^Fsðx; yÞ is determined
Ft x; yðÞ¼ FFx; yðÞ bFs x; yðÞ:
(10)
To avoid over-enhancement of background
parts and insufficient enhancement of the targets
areas, different histogram equalization strategies
are used for the smooth layer and the texture layer
1) Equalization of the smooth layer
layer is quite different from that of the original
image and the smooth layer, as shown in
Figure. 6. The main reason is that the gray level
of most areas in the smoothing layer are very
close to that of the original image, as a result,
the differential image Ftðx; yÞ has low gray values
in these areas. It is also reflected in the histogram
that the values of Ftðx; yÞ are almost
concentrated in the low gray level area.
Histogram equalization make these pixels with
low gray level to get a higher gray value to
approach the average gray value, whichmeans the
differences between texture part and smooth part
become smaller. Obviously, this goes against our
original intention to enhance the visual saliency of
the infrared targets. To solve this problem, the texture
layer is further decomposed into two subsets
according to the average gray valueFtðx; yÞ
Ft1 x; yðÞ ¼ Ft x; yðÞj FtFt x; yðÞ
Ft2 x; yðÞ ¼ Ft x; yðÞj Ft >Ft x; yðÞ
(11)
March/April 2023
19
IEEE Consumer Electronics Magazine - March/April 2023

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - March/April 2023
