IEEE Consumer Electronics Magazine - March/April 2023 - 18

Special Section on Security, Trust and Privacy Solutions for Intelligent Internet of Vehicular Things
FIGURE 3. Influence of proposed edge-preserving
regularization terms on SR results.
It can be derived from (6) that the expression for
RdX(i,j) has the following properties:
lim
d!0
RdXði; jÞ¼
1;rXði; jÞ 6¼ 0
0;rXði; jÞ¼ 0
:
(7)
It can be seen from (7) that the proposed Lpnorm
is approximately equal to a L0-norm, when
the value of d is small. The edge-preserving regularization
term realize the gradient constraint of
image by approximating L0-norm, and adjust the
edge of image in the direction of greater gradient.
To verify the effectiveness of the proposed edgepreserving
regularization term in the aspect of
enhancing the edge saliency, the experimental
results with different weight coefficients2 are
shown in Figure. 3.
GRADIENT-CONSTRAINT IMAGE
ENHANCEMENT
Due to the existence of molecular thermal
motion phenomenon, different objects continue
to carry out heat transfer between each other,
leading to little temperature difference of different
objects in the same field of view. As a result,
the contrast of infrared image is relatively low
and the edges of targets inside the infrared
image seem not obvious, which brings inconvenience
to infrared target detection.
In order to further solve this problem, an infrared
image enhancement method is proposed on
the basis of SR result, hoping to increase the
saliency of infrared target submerged in the
18
FIGURE 4. Workflow of the proposed gradientconstraint
image enhancement algorithm.
background and provide further help for the subsequent
infrared target detection. Most areas of
an infrared image are composed of background
parts, since the background temperature is
approximately equal to the environment temperature,
the distributions of gray values in these
areas are very concentrated. Traditional global
enhancement algorithms may lead to the overenhancement
of background parts and insufficient
enhancement of the targets areas, when applied to
infrared images. To avoid this, we propose a multilayer
enhancement strategy based on gradientconstraint
image decomposition, the overall workflowof
the algorithm is shown in Figure. 4.
The workflow shown in Figure. 4 can be
expressed as follows:
step 1. Obtain the SR result using the SR algorithm
proposed in Section III-A.
step 2. Decompose the SR image to the smooth
layer and the texture layer with L1-norm
regularization of image gradient.
step 3. Enhance the smooth layer and the texture
layer separately through different
histogram equalization strategies.
step 4. Fuse the results of each layer according
to different weights, and output the
final enhanced image.
For the infrared image Fðx; yÞ with the resolution
of M N, suppose that it consists of the
smooth layers Fsðx; yÞ and the texture layer
Ftðx; yÞ
FFx; yðÞ¼ Fs x; yðÞþFt x; yðÞ:
(8)
IEEE Consumer Electronics Magazine

IEEE Consumer Electronics Magazine - March/April 2023

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