ShiLiyan1
-
(School of Information Engineering and Artificial Intelligence, Henan Open University,
Zhengzhou Vocational University of Information and Technology, Zhengzhou 450046, China
)
Copyright © The Institute of Electronics and Information Engineers(IEIE)
Keywords
LabVIEW, Epidemics, Sudden events, Public security images, Image recognition
1. Introduction
In recent years, flexible geometric models and attribute assignment methods have been
used in security image recognition. At present, image recognition technology is widely
used in society, especially for image processing. For example, an image is analyzed
based on the frame, the grayscale used, and its color histogram, texture attributes,
and/or mixed attributes. The image is also displayed as a matrix of elements that
corresponds to the grayscale values in the image. Based on one-valued degradation
and principal component analysis, image features are extracted and classified. Because
of the randomness of the angle and the positions of human bodies, descriptions of
the objects in security images are different. In the anti-smuggling department, the
image sequence can be analyzed according to image size, materials present, and image
characteristics. With the introduction of new technology and safety control methods,
three-dimensional image recognition methods are gradually developed.
With the continuous development of science and technology in the fields of image recognition
and statistical learning theory, artificial intelligence and new scientific methods
are used. An automatic imaging method is recommended. However, in the application
of security image recognition, the positions of specific targets in security images
are not clear, and different targets and boundaries are fuzzy, so it is difficult
to realize automatic detection. Therefore, research on security image detection methods
is the key to future security image processing. It is necessary to construct an analysis
model for suspicious security images from public places in order to realize suspicious
activity detection and fusion of security images from public places. Improving recognition
abilities in security images from public places based on epidemics and sudden events
will improve the level of security monitoring.
The extraction of suspicious features from security images during epidemics and sudden
events is based on detection and cluster analysis of the suspicious information. In
traditional methods, extracting such suspicious features mainly includes irregular
triangulation methods, linear enhancement technology, and particle clustering analysis.
In [4], a method for extracting suspicious features from public security images during epidemics
and other sudden events via Laplace sharpening feature analysis was proposed. Maximum
cluster variance analysis is used to establish a suspicious feature segmentation model
for public security images, and image parameters can be detected and identified through
the suspicious segmentation results. However, this method has poor adaptability and
low detection accuracy when extracting suspicious features from public security images
taken during epidemics and sudden events. In [5], a suspicious feature extraction method based on a regional growth algorithm was
proposed. Fuzzy-information clustering and regional growth analysis are used to extract
suspicious features from public security images captured during epidemics and sudden
events, but environmental adaptability under this method is not high. The authors
in [6] put forward a suspicious feature extraction method for security images based on grayscale
parameter analysis combined with activities and shapes to realize feature extraction
from security images taken during epidemics and sudden events by using maximum pixel
recognition. The imaging level and degree of recognition from this method are not
high.
To solve the above problems, this paper puts forward an image recognition algorithm
based on LabVIEW for public security images, and constructs image recognition and
suspicious feature extraction based on cross-regional block fusion. It constructs
the model for suspicious information detection by adopting a fusion detection method
for human appearance feature parameters and dynamic gaits. This paper constructs a
matching model of suspicious dynamic information block features from security images,
decomposes suspicious background information, and constructs an edge contour detection
model for security images taken during epidemics and sudden events. According to the
suspicious edge detection results from public-place security images, the spatial structure
of the images is extracted, and the risk difference characteristics of human body
shapes in the images are captured to enhance the detection of suspicious information.
Finally, a simulation test shows that the proposed method has superior performance,
improving suspicious feature extraction abilities from public-place security images
based on epidemics and sudden events.
2. Suspicious Information Detection and Imaging Optimization
2.1 Detection of Suspicious Information
In order to extract suspicious features from security images taken during epidemics
and sudden events, it is necessary to construct a scale segmentation model of suspicious
information in security images. In this work, an image collection device with a flash
detector is adopted. When a target is hit by X-rays, the flash detector will detect
them passing through the target. The essence of detection is to convert X-rays into
electronic signals, which facilitates the subsequent image processing. This pavilion
can turn X-rays into visible rays in a short period of time. The obtained image can
quickly be converted into an electronic signal; the X-rays pass through the object
and are then radiated by the detector. The function of the photodiode is to convert
weak optical signals into codes that are convenient for analysis and processing, so
as to obtain suspicious information from public-place security images taken during
epidemics and sudden events. The label distribution is generated according to three
factors that are well packaged into a multi-factor distribution, thus realizing feature
embedding and feature extraction from data sets of suspicious images. The spatial
structure model is shown in Fig. 1.
Fig. 1. Schematic for security inspection of images from public places.
The suspicious information model for security inspection of images from public places
during epidemics and sudden events is shown in Fig. 1. The edge contour detection function is obtained by using maximum pixel parameter
identification, and the label identification model is expressed as follows:
where $\omega $ is the time point at which the actual occurrence of a public threat
is controlled, $v(t)$ is the degree of influence from the overall time distribution,
$p$ is the movement in which actual public security threat Y belongs, $x(t)$ is the
public security threat movement in which style I belongs, and $\varphi [p-x(t)]$ is
the public security threat parameter to capture different historical background colors
of public security threats. Based on the fuzzy-boundary feature detection method,
through orthogonal wavelet scale decomposition, a multi-level wavelet decomposition
structure model of security images taken during epidemics and sudden events is constructed
as follows:
where
where $\chi _{ws}(s,\tau )$ is the loss between the actual shape and the predicted
shape, $u(t)$ is the pixel error in the security images from public places taken during
epidemics and sudden events, and $u^{\ast }$ is the probability of the I-th kind of
suspicious category. A matching model of the suspicious dynamic information block
feature of security images from public places is constructed by using fusion detection
of human appearance feature parameters and dynamic gaits. The suspicious block matching
model is described by formula (5):
The scale spatial distribution function is constructed, and sparse background feature
segmentation of public-place security images taken during epidemics and sudden events
is carried out based on the matched filter detection method to obtain joint detection
results of image position and scale parameters. The extracted security images are
analyzed by information fusion using adaptive clustering, and the label distribution,
L, derived from the background information of factors threatening public security
(such as origin time, birthplace, and movement of threat factors) can assist in the
visual feature learning in a convolutional neural network. The fusion clustering model
is expressed as follows:
where, $u(x)$ is the neighborhood grayscale information of security images taken during
epidemics and sudden events, and $g(\cdot )$ is the rotation invariant feature quantity
of the image that satisfies $g\colon [0,1]\rightarrow [0,1]$. According to the above
analysis, a suspicious information detection model is constructed, and suspicious
features are extracted according to the information detection results.
2.2 Image Information Enhancement
Combining background and environmental factor detection, suspicious background information
feature decomposition is realized, the edge contour detection and image enhancement
model is constructed, and the background suspicious principal component feature quantity
is obtained as follows:
where $V_{c}(Y)$ is the subspace distribution of the security inspection image, $\beta
$ is the correlation map, and $\sum _{c\subset C}V_{c}(Y)$ is the template matching
the key-point direction information distribution, which minimizes the loss of style
and content in the final output, and obtains the factors threatening public security.
The model function of the drawing style learning framework is as follows:
wherein $V_{m}$ is the label of the factor threatening public security, and $V_{n}$
is the characteristic detection value of the factors threatening public security appearing
in adjacent order. According to the position and scale distribution of the extreme
points, the pixel values of the subspace noise reduction and information enhancement
output of the security image can be calculated as follows:
where $g(x,y)$ is the matching rate of grayscale data, $\overset{\wedge }{f_{Lee}}(x,y)$
is the frequency parameter of content feature matching, and $t$ is the sampling point.
Using comprehensive feature vector fusion and difference fusion, the two domains are
X and Y, respectively, and a feature detection model is constructed, which is expressed
as follows:
where $\mu _{pq}$ is the R-layer color distribution of the image, and $\mu _{00}$
is the G-layer color distribution of the security image. Semantic segmentation of
the security image is realized by using the difference mapping method, and the image
information is deeply enhanced, with the output expressed as follows:
where $u_{j}(k)$ is a local feature, and $\overset{\wedge }{x^{i}}$ is a public security
threat factor. Based on the above analysis, a noise reduction model is constructed,
and a suspicious clustering model with enhanced information output is obtained, as
shown in Fig. 2.
Fig. 2. Clustering model of suspicious features in security images from public places.
3. Feature Extraction Optimization
3.1 Suspicious Information Fusion
Based on suspicious edge detection, image enhancement is carried out by using a subspace
noise reduction method. Combined with a cross-regional block fusion model, multi-level
suspicious feature detection in security images from public places during epidemics
and sudden events is constructed as follows:
where $r$ is the spatial envelope feature. According to cross-regional block fusion,
several different types of image styles are defined, and the suspicious feature point
set is obtained:
where $a$ is the color feature, $f_{0}$ is the deep network feature, and $B$ is the
multi-view local color component. Using semantic segmentation, a fusion multilevel
feature distribution model of security images is constructed, and vector quantization
coding and the cross-regional block fusion output from security images captured during
epidemics and sudden events are obtained as follows:
where $PS$ is semantic information from the security inspection, and $d_{i}$ is the
sampling interval from the security inspection. Using a fine semantic segmentation
method, the grid parameter distribution of the security images is obtained as follows:
By using multi-scale feature decomposition, the detection of suspicious output from
security images taken during epidemics and sudden events can be obtained as follows:
where $C^{N}$ is the single-depth compression scale of fuzzy space, $D^{N}$ is the
characteristic response, and $s^{N}$ is the label inherited from the original image.
According to suspicious clustering segmentation, suspicious information fusion processing
is carried out using multi-level visual feature analysis to improve image detection
and recognition.
3.2 Feature Extraction in Security Images
Combined with the cross-regional block fusion model, the analytic rule function for
gradient fusion of suspicious information from security images taken during epidemics
and sudden events is constructed. Based on the cross-regional block fusion distribution
and the decomposition results of suspicious features in images, the risk factors from
the security images are calculated as follows:
where $\sigma $ represents the suspicious parameter from public security image fusion,
and $\Delta x$ represents the distribution set of candidate areas from public security
images captured during epidemics and sudden events. Based on the shallow feature information
method, the fusion method for public security images is constructed, and the filtering
detection model of public security images is established. The suspicious segmentation
results are as follows:
where c is the central pixel of the security image, and r is the feature map fusion
model of the security image. The suspicious detection output from a security image
can be obtained as follows:
where $x_{i+1}$ is the color difference component of the whole area in the security
image, and $y_{i}$ is the local color component of the whole area in the image. Based
on the cross-regional block fusion distribution and the feature decomposition results
of suspicious information from the image, the grayscale edge information decomposition
is adopted to obtain the image feature decomposition model as follows:
where $h(x,y)*f(x,y)$ is the fusion parameter of the suspicious information samples
based on epidemics and sudden events. The spatial structure of the images is extracted,
difference characteristics of human body shapes are captured to enhance suspicious
information detection during epidemics and sudden events, and the safety threatening
factors are as follows:
where $f(x,y)$ is the spatial spectrum feature of suspicious information in the images,
and $\eta (x,y)$ is the grayscale feature from security inspection colors based on
sudden events. In summary, grayscale edge information decomposition is adopted to
extract suspicious features from security images captured during epidemics and sudden
events. Visual simulation was carried out based on LabVIEW.
4. Simulation and Results Analysis
In order to verify the performance of the proposed method for extraction of suspicious
features from public security images taken during epidemics and sudden events, a simulation
was conducted and analysis carried out based on LabVIEW. Public security images used
included the Painting dataset, the OilPainting dataset, and the Pandora dataset. The
Painting91 dataset contains 4266 images from 91 artists, including 2338 paintings
from 50 artists done in 13 styles. The OilPainting dataset has 19,787 oil paintings
in 17 artistic styles, and the Pandora dataset contains 7724 images in 12 artistic
styles. In addition, these three datasets have different characteristics. Specifically,
compared with the Painting91 style dataset, the OilPainting dataset divides styles
in more detail. The Pandora dataset contains a wider range of categories from a long
time span. From them, a training set and a testing set were created according to the
settings of the data sets, containing 1250 images and 1088 images, respectively. The
training process for suspicious feature extraction from security images captured during
epidemics and sudden events was iterated 3000 times. The distribution of deep feature
information from security images during epidemics and sudden events was 12, and the
sampling delay was 0.25ms. Based on the above parameter settings, the suspicious feature
extraction model for security images was constructed, and the security images obtained
are shown in Fig. 3.
Taking the images in Fig. 3 as a sample, suspicious features of the images were detected. When the color of an
image reached or equaled 255, it was a special object. Conversely, pixels with a grayscale
value of 0 were removed to reflect the background or abnormal range of the object.
The second value in the figure is the grayscale value. When the whole picture is presented,
the influence of black and white is very significant; that is, in a single picture,
256 different light thresholds were selected. Secondly, the grayscale value was used
to reflect overall and local characteristics of the image. A binary image is a key
link in image processing, especially in a functional image. Therefore, a large amount
of binary image data was produced, which can help the image processing to continue.
It only contained 100 or 255 points, and was not affected by multiple pixel values.
Generally, to obtain the desired two digital images the uncovered area is defined
as the edge related to the closed boundary. The pixel beyond the aperture is a special
object, with a grayscale of 255. Otherwise, these pixels are removed from the O-level
object to show the background or a region of the object, thus obtaining the detection
results of suspicious feature points from public security images based on epidemics
and sudden events, as shown in Fig. 4. The central error distribution of video image detection is shown in Fig. 5.
By analyzing Figs. 4 and 5, we can know that in the process of forming images, there
are many groups of features or incorrect basic elements of one image in another image
due to noise and interference. At the same time, we can see from Fig. 5 that the method proposed in the study basically maintains a low error, and is superior
to DF and MIL in most cases, with high detection accuracy. In the original image,
the basic elements of the above features or errors are defect problems, which are
usually solved within the framework of standardization through different contract
terms. First, the HOPFIEK network was analyzed. When there are global constraints,
compatibility constraints and adjacent constraints, the energy effect of HOPFIEK is
fully demonstrated. The algorithm minimizing straight-line segments of two images
improves the reliability of registration. In addition, methods such as the least square
method and the voting method effectively eliminate false points and mismatches. In
this paper, the calibration accuracy of suspicious feature points from security images
during epidemics and sudden events is high, and the results from suspicious feature
extraction were tested by different methods. The proposed method had a high identification
rate for output features of suspicious feature extraction, and the time cost of suspicious
feature extraction was tested. The comparison results are in Table 1, which shows that the time cost for the proposed method in suspicious feature extraction
from security images during epidemics and sudden events is short.
Table 1. Time overhead test (unit: ms).
|
Iterations
|
Our method
|
SIFT algorithm
|
Harris algorithm
|
|
Dataset 1
|
3.554
|
12.184
|
21.135
|
|
Dataset 2
|
3.131
|
12.039
|
21.570
|
|
Dataset 3
|
3.820
|
12.838
|
21.044
|
|
Dataset 4
|
3.153
|
12.802
|
21.657
|
Fig. 3. Security images obtained.
Fig. 4. Feature detection in security images from public places during epidemics and sudden events.
Fig. 5. Diagrams for center error in a video sequence.
5. Conclusion
In this paper, suspicious features from public security images captured during epidemics
and sudden events are extracted according to the undirected weighted graph recognition
method. Based on cross-regional block fusion in suspicious feature extraction, the
orthogonal wavelet scale decomposition method was adopted to detect suspicious information
from public security images, and the semantic segmentation method was adopted to construct
a fusion multilevel feature distribution model for public security images taken during
epidemics and sudden events. Combined with multilevel visual feature analysis, suspicious
information fusion processing was carried out to enhance image detection and recognition.
This research shows that the proposed method has high output stability, a short time
cost, and good reliability when extracting suspicious features in security images
from public places during epidemics and sudden events, and it improves detection and
recognition from those security images.
REFERENCES
Khan S, Tufail M, Khan M T, et al. A novel framework for multiple ground target detection,
recognition and inspection in precision agriculture applications using a UAV. Unmanned
Systems, vol. 10, no. 1, pp. 45-56, 2022.
Jo H J, Choi W. A survey of attacks on controller area networks and corresponding
countermeasures. IEEE Transactions on Intelligent Transportation Systems, 2021, 23(7):
6123-6141, 10.1109/TITS. 2021.3078740
Marino. A, Sugimoto. M, Ouchi. K, Hajnsek. I, ``Validating a notch filter for detection
of targets at sea with ALOS-PALSAR data: Tokyo Bay,'' IEEE Journal of Selected Topics
in Applied Earth Observations and Remote Sensing, vol. 7, no. 12, pp. 74907-74918,
2014, 10.1109/JSTARS.2013.2273393
Daihong J, Sai Z, Lei D & Yueming D. Multi-scale generative adversarial network for
image super-resolution. Soft Computing, 2022, 26(8): 3631-3641. https://doi.org/10.5244/c.26.135
Krizhevsky. A, Sutskever. I, Hinton. GE, ``ImageNet classification with deep convolutional
neural-networks[C]//Proceedings of the 25th International Conference on Neural Information
Processing Systems,'' Red Hook: Curran Associates Inc., pp. 1097-1105, 2012.
Liu P, Zhou Y, Peng D~& Wu D. Global-Attention-Based Neural Networks for Vision Language
Intelligence. IEEE/CAA Journal of Automatica Sinica, 2021, 8(7): 1243-1252, 10.1109/JAS.2020.1003402
Wang. FS, Wang. J, Li. B, ``Deep attribute learning based traffic sign detection,''
Journal of Jilin University (Engineering and Technology Edition), vol. 48, no. 1,
pp. 319-329, 2018, 10.13229/j.cnki.jdxbgxb20161120
Quartulli. M, Datcu. M, ``Stochastic geometrical modeling for built-up area understanding
from a single SAR intensity image with meter resolution,'' IEEE Transactions on Geoscience
and Remote Sensing, vol. 42, no. 9, pp. 1996-2003, 2004, 10.1109/TGRS.2004.833391
Xu. XU, Fengli. Z, Guojun. W, Xiyou. F, Minmin. S, Zhikun. L, Xingdong. L, ``Building
height retrieval from dual-aspect SAR images based on match of strong backscattering
features,'' Remote Sensing Technology and Application, vol. 31, no. 1, pp. 149-156,
2016,
Yu H, Cheng X, Chen C,Heidari, A. A., Liu, J., Cai, Z., & Chen, H. Apple leaf disease
recognition method with improved residual network. Multimedia Tools and Applications,
2022, 81(6): 7759-7782. 10.1007/s11042-022-11915-2
Xu. C, Chen. Z, Hou. R, ``Deep learning classification method of Landsat 8 OLI images
based on inaccurate prior knowledge,'' Journal of Computer Applications, vol. 40,
no. 12, pp. 3550-3557, 2020,
Sulla-Menashe. D., Gray. JM, Abercrombie. SP, Friedl. MA, ``Hierarchical mapping of
annual global land cover 2001 to present: the MODIS Collection 6 Land Cover product,''
Remote Sensing of Environment, vol. 222, pp. 183-194, 2019,
Rodriguez-Galiano. VF, Ghimire. B, Rogan. J, Chica-Olmo. M, Rigol-Sanchez. JP, ``An
assessment of the effectiveness of a random forest classifier for landcover classification,''
ISPRS Journal of Photogrammetry and Remote Sensing, vol. 67, pp. 93-104, 2012,
Zhang. C, Sargent. I, Pan. X, Li. H, Gardiner. A, Hare. J, Atkinson. PM, ``An object-based
con-volutional neural Nntwork (OCNN) for urban land use classification,'' Remote Sensing
of Environment, vol. 216, pp. 57-70, 2018, 10.1016/j.rse.2018.06.034
Homer. C, Dewitz. J, Yang. L, Jin. S, Danielson. P, Xian. G, Megown. K, ``Completion
of the 2011 national land cover database for the conterminous united states-representing
a decade of land cover change information,'' Photogrammetric Engineering and Remote
Sensing, vol. 81, no. 5, pp. 345-354, 2015,
Author
Liyan Shi obtained her M.Sc. degree from Huazhong University of Science and Technology
(2008). Presently, she is working as an associate professor in Information Engineering
and Artificial Intelligence School, the open university of Henan. She has published
nearly 10 articles. Her areas of interest include Image processing, software engineering,
virtual reality technology.