CN109033969B - Infrared target detection method based on Bayesian saliency map calculation model - Google Patents

Infrared target detection method based on Bayesian saliency map calculation model Download PDF

Info

Publication number
CN109033969B
CN109033969B CN201810669568.0A CN201810669568A CN109033969B CN 109033969 B CN109033969 B CN 109033969B CN 201810669568 A CN201810669568 A CN 201810669568A CN 109033969 B CN109033969 B CN 109033969B
Authority
CN
China
Prior art keywords
probability
saliency map
target detection
bayesian
infrared target
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810669568.0A
Other languages
Chinese (zh)
Other versions
CN109033969A (en
Inventor
宋勇
李旭
赵宇飞
郭拯坤
杨昕
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Institute of Technology BIT
Original Assignee
Beijing Institute of Technology BIT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Institute of Technology BIT filed Critical Beijing Institute of Technology BIT
Priority to CN201810669568.0A priority Critical patent/CN109033969B/en
Publication of CN109033969A publication Critical patent/CN109033969A/en
Application granted granted Critical
Publication of CN109033969B publication Critical patent/CN109033969B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/29—Graphical models, e.g. Bayesian networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an infrared target detection method based on a Bayesian saliency map calculation model, particularly relates to an infrared target detection method based on a visual attention mechanism, and belongs to the technical field of computer vision. The realization method of the invention is as follows: establishing a saliency map calculation model based on Bayes 'theorem by calculating the prior probability and the likelihood probability of the infrared image on the basis of Bayes' theorem; solving prior probability based on a meanshift segmentation algorithm and side suppression; solving likelihood probability based on pixel point statistics; and calculating a final saliency map based on the solved prior probability and likelihood probability to obtain an infrared target detection result. The invention realizes the detection of the infrared target based on the Bayesian saliency map calculation model, and has the following advantages: the background noise suppression capability is strong, and the outline of the salient region is clear and complete.

Description

Infrared target detection method based on Bayesian saliency map calculation model
Technical Field
The invention relates to a detection method of an infrared target, in particular to an infrared target detection method based on a visual attention mechanism, belongs to the technical field of computer vision, and is the early work of target identification and target tracking.
Background
The infrared target detection is a key technology in an infrared detection system, has important application in the field of infrared target searching and early warning, and has become a research hotspot in recent years. Under the interference influence of an infrared detector, cloud and fog background and the like, an infrared image is fuzzy generally and has stronger background clutter.
In recent years, the information processing mechanism of the human brain visual system has unique advantages in the fields of target detection, identification and tracking. The human brain visual attention mechanism has the characteristics of emphasizing special properties and highlighting a remarkable target, so that an observer can quickly find an interested target in a certain scene to acquire useful information. The current visual attention models for object detection are mainly divided into feature-driven bottom-up visual attention models and task-driven top-down visual attention models. The top-down attention model is generally divided into two parts, namely feature learning and significance detection, and attracts attention in recent years due to being more consistent with the mechanism of human brain visual attention mechanism, but most attention models only stay in a qualitative description stage, cannot be completely realized, need to learn features and are complex in calculation. The bottom-up attention model is simple in calculation and high in operation speed, and is widely applied.
At present, an infrared target detection algorithm based on a visual attention mechanism has a better detection effect compared with a conventional algorithm, but under the condition of a complex background, the algorithms have the problems of too complex calculation process, limited background suppression capability and the like in different degrees. In 1998, itti et al propose a visual attention model based on saliency, extract three kinds of color features, brightness features and direction features of an image, and finally generate a saliency map, so that the visual attention model becomes a first relatively complete and systematic calculation model in the field and lays a foundation for the research of the visual attention model; in 2007, hou et al propose a visual attention model based on spectral residual errors, explore background characteristics, and obtain a saliency map of a time-space domain by using background information and saliency region information. The method has high operation speed, but is greatly influenced by noise and has low detection precision; in 2012, sun et al proposed a visual attention model with super-gaussian distribution as a salient feature on CVPR, and analyzed visual statistics points to obtain the characteristic that the salient has super-gaussian distribution. In 2016, li et al proposed a space-time saliency calculation method, in which static saliency and dynamic saliency were fused to obtain a saliency map of an enhanced dynamic target, and the algorithm improved the detection rate of the dynamic target, but was more complex. Therefore, infrared target detection in a complex background based on visual attention mechanism is a challenging research direction.
Disclosure of Invention
In order to solve the problems of weak noise suppression capability and low detection precision brought by an infrared target detection method based on a visual attention saliency map calculation model in the prior art and the requirement of high detection precision of infrared target detection under a complex condition, the infrared target detection method based on a Bayesian saliency map calculation model disclosed by the invention aims to solve the technical problems that: the infrared target detection is realized based on the Bayesian saliency map calculation model, and the method has the following advantages: the background noise suppression capability is strong, and the outline of the salient region is clear and complete.
The invention is realized by the following technical scheme.
The invention discloses an infrared target detection method based on a Bayesian saliency map calculation model, which is based on Bayesian theorem and establishes a saliency map calculation model based on the Bayesian theorem by calculating the prior probability and the likelihood probability of an infrared image. Solving prior probability based on a meanshift segmentation algorithm and side suppression. And solving likelihood probability based on pixel point statistics. And calculating a final saliency map based on the solved prior probability and likelihood probability to obtain an infrared target detection result.
The invention discloses an infrared target detection method based on a Bayesian saliency map calculation model, which comprises the following steps:
step 1: and establishing a saliency map calculation model based on Bayesian theorem.
And deducing the image significance by using Bayes theorem, and converting the image significance problem into a probability solving problem by using a Bayes formula. Bayesian formulation is derived as shown in equation (1):
Figure BDA0001707721250000021
wherein x represents a pixel point x, T represents a target, and the saliency value of the pixel point x is defined as P (T | x), which refers to the probability that the pixel point x is known as the target. From the total probability formula, P (x) is derived as follows:
P(x)＝P(T)·P(x|T)+P(B)·P(x|B) (2)
where B is defined as background, the calculation formula of P (x) is converted into equation (2) because event T and event B are mutually exclusive. Finally, establishing a saliency map calculation model based on Bayesian theorem as follows:
Figure BDA0001707721250000022
step 2: and solving the prior probability based on a meanshift algorithm and adaptive side suppression.
P (T) and P (B) are defined as prior probabilities, i.e. estimates of the saliency values of each pixel of the image, and then likelihood probabilities are calculated given the foreground and background. The prior probability is obtained by combining a meanshift algorithm and an adaptive side suppression method, firstly, PCA compression processing is carried out on an image, and the following purposes are achieved: reducing the calculation amount and improving the calculation speed; and (b) the detection precision is more accurate, and the robustness is high.
And step 3: and solving likelihood probability based on pixel point statistics.
Likelihood probability is defined as the probability that a pixel point x is foreground or background, given the foreground or background. The calculation formula is as follows:
Figure BDA0001707721250000031
Figure BDA0001707721250000032
g (x) represents the gray value of pixel point x, N T (g (x)) represents the number of pixels having a gray scale value g (x) in the foreground obtained in step 2, N T Representing the total number of pixels in the foreground. In the same way, N B (g (x)) represents the number of pixels with a background gray scale value g (x), N B Representing the total number of pixels in the background.
And 4, step 4: and calculating a final saliency map based on the prior probability solved in the step 2 and the likelihood probability solved in the step 3, and obtaining an infrared target detection result.
And (3) substituting the prior probability and the likelihood probability obtained by the formulas (4) and (5) into a saliency map calculation formula shown in the formula (3), calculating to obtain a final saliency map, and obtaining an infrared target detection result.
Has the advantages that:
1. the speed is fast. The infrared target detection method based on the Bayesian saliency map calculation model disclosed by the invention realizes the detection of the infrared target based on the Bayesian saliency map calculation model, introduces a human brain visual attention mechanism into the infrared target detection process through the saliency map calculation model, utilizes the characteristic that the visual attention mechanism highlights the interested target, accelerates the operation speed, and simultaneously performs PCA compression processing on the infrared image, thereby reducing the operation amount and improving the operation speed.
2. The noise suppression capability is strong. The infrared target detection method based on the Bayes saliency map calculation model disclosed by the invention is based on the Bayes theorem, the prior probability and the likelihood probability are calculated to obtain the saliency map, and the noise can be effectively inhibited and a clear saliency region can be obtained by the calculation method of the prior probability and the likelihood probability.
3. The salient region is clear in outline. The infrared target detection method based on the Bayesian saliency map calculation model disclosed by the invention is based on Bayesian theorem, the prior probability and the likelihood probability are calculated to obtain the saliency map, and the prior probability is solved based on a meanshift method, so that the complete outline of the saliency region can be kept.
Drawings
FIG. 1 is a flow chart of an infrared target detection method based on a Bayesian saliency map calculation model disclosed by the invention;
FIG. 2 is a schematic block diagram of an infrared target detection method based on a Bayesian saliency map calculation model disclosed by the invention;
FIG. 3 is a prior probability of an image obtained by solving in step 1 according to an embodiment of the present invention;
fig. 4 is a foreground region and a background region in step 2 of the method of the present invention, where fig. 4 (a) is a foreground region diagram and fig. 4 (b) is a background region diagram.
FIG. 5 is a graph comparing the results of the method of the present invention with other methods of detecting significance.
Detailed Description
For better illustrating the objects and advantages of the present invention, the following description is provided in conjunction with the accompanying drawings and examples.
Example (b):
in this embodiment, a MATLAB 2013a platform is used on a PC of Intel (R) Core (TM) i3 CPU 3.07ghz,4.00g to test a single frame image of an infrared target under a complex background, thereby completing simulation.
As shown in fig. 1, the overall process of the infrared target detection method based on the bayesian saliency map calculation model disclosed in this embodiment is shown in fig. 1, and specifically includes the following steps:
step 1: and solving the prior probability based on a meanshift algorithm and side suppression.
Inputting an infrared image yuan _ gou, and carrying out PCA compression processing on the image. And (3) carrying out suppression processing on the compressed image by utilizing a meanshift algorithm and an adaptive side to obtain an approximately salient region of the image, namely the prior probability of the image. As shown in fig. 3.
Step 2: and solving likelihood probability based on pixel point statistics.
Likelihood probability is defined as the probability that a pixel point x is foreground or background, given the foreground or background. The calculation formula is as follows:
Figure BDA0001707721250000041
Figure BDA0001707721250000042
g (x) represents the gray value of pixel point x, N T (g (x)) represents the number of pixels with a gray value g (x) in the foreground of the prior probability, N T Representing the total number of pixels in the foreground. In the same way, N B (g (x)) represents the number of pixels with a background gray scale value g (x) of the prior probability, N B Representing the total number of pixels in the background.
The specific calculation method of the likelihood probability is as follows: and on the premise of knowing the foreground or background area, calculating the number of pixels with the same gray value of the pixel point x in the foreground or background area and the probability of occupying the total number of the pixels in the foreground or background area. The foreground region and the background region are shown in fig. 4, where fig. 4 (a) shows the known foreground region and fig. 4 (b) shows the known background region.
And step 3: and calculating a final saliency map based on the prior probability solved in the step 2 and the likelihood probability solved in the step 3, and obtaining an infrared target detection result.
The saliency map calculation formula is:
Figure BDA0001707721250000051
and substituting the obtained prior probability, namely P (T) and P (B), and the likelihood probability, namely P (x | T) and P (x | B), into an image significance calculation formula (8), calculating a final significance map, and obtaining an infrared target detection result. As shown in fig. 5: from left to right, the first column is an original infrared target image, the second column is an infrared target detection result of the Itti method, the third column is an infrared target detection result of the SR method, and the fourth column is an infrared target detection result of the method, and as can be seen from the figure, the Itti method is unclear in target detection, and only one bright area is detected; the outline of the significant region of the SR method is unclear, and the noise interference is large; the infrared target detection method provided by the embodiment has the advantages that the outline of the salient region is clear, the background clutter is suppressed, and the detection result is superior to other comparison methods.
The above detailed description is intended to illustrate the objects, aspects and advantages of the present invention, and it should be understood that the above detailed description is only exemplary of the present invention and is not intended to limit the scope of the present invention, and any modifications, equivalents, improvements and the like within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (1)

1. The infrared target detection method based on the Bayesian saliency map calculation model is characterized by comprising the following steps of: comprises the following steps of (a) carrying out,
step 1: establishing a saliency map calculation model based on Bayesian theorem;
deducing image significance by using Bayes theorem, and converting an image significance problem into a probability solving problem by using a Bayes formula; the Bayesian formula derivation is shown in formula (1):
Figure FDA0003818894000000011
wherein x represents a pixel point x, T represents a target, the significant value of the pixel point x is defined as P (T | x), and the P (T | x) refers to the probability that the known pixel point x is the target; from the total probability formula, P (x) is derived as follows:
P(x)＝P(T)·P(x|T)+P(B)·P(x|B) (2)
where B is defined as background, the calculation formula of P (x) is converted into equation (2) because event T and event B are mutually exclusive; finally, establishing a saliency map calculation model based on Bayesian theorem as follows:
Figure FDA0003818894000000012
step 2: solving prior probability based on a meanshift algorithm and self-adaptive side suppression; the specific process is as follows:
1) Carrying out PCA compression processing on the input image to obtain a compressed image;
2) Processing the compressed image by utilizing a meanshift algorithm and self-adaptive side suppression respectively in the two parallel processing channels to obtain corresponding processing results;
3) Multiplying the results obtained after the two methods are processed to obtain approximate salient regions of the target in the input image, namely the prior probabilities P (T) and P (B);
and step 3: solving likelihood probability based on pixel point statistics;
the likelihood probability is defined as the probability that the pixel point x is the foreground or the background under the condition of the known foreground or the background; the calculation formula is as follows:
Figure FDA0003818894000000013
Figure FDA0003818894000000014
g (x) represents the gray value of pixel point x, N T (g (x)) represents the number of pixels in the foreground having a gray scale value g (x), N T Representing the total number of pixels in the foreground; in the same way, N B (g (x)) represents the number of pixels having a gray scale value g (x) in the background, N B Representing the total number of pixels in the background;
and 4, step 4: calculating a final saliency map based on the prior probability solved in the step 2 and the likelihood probability solved in the step 3 to obtain an infrared target detection result;
and (3) substituting the prior probability and the likelihood probability obtained by the formulas (4) and (5) into a saliency map calculation formula shown in the formula (3), calculating to obtain a final saliency map, and obtaining an infrared target detection result.
CN201810669568.0A 2018-06-26 2018-06-26 Infrared target detection method based on Bayesian saliency map calculation model Active CN109033969B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810669568.0A CN109033969B (en) 2018-06-26 2018-06-26 Infrared target detection method based on Bayesian saliency map calculation model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810669568.0A CN109033969B (en) 2018-06-26 2018-06-26 Infrared target detection method based on Bayesian saliency map calculation model

Publications (2)

Publication Number Publication Date
CN109033969A CN109033969A (en) 2018-12-18
CN109033969B true CN109033969B (en) 2022-12-02

Family

ID=64611270

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810669568.0A Active CN109033969B (en) 2018-06-26 2018-06-26 Infrared target detection method based on Bayesian saliency map calculation model

Country Status (1)

Country Link
CN (1) CN109033969B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110084833A (en) * 2019-04-25 2019-08-02 北京计算机技术及应用研究所 A kind of infrared motion target detection method based on adaptive neighborhood Technology of Judgment

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107967492A (en) * 2017-12-18 2018-04-27 西北工业大学 Bayes's conspicuousness detection method that a kind of combination is detected like physical property

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107967492A (en) * 2017-12-18 2018-04-27 西北工业大学 Bayes's conspicuousness detection method that a kind of combination is detected like physical property

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于自适应侧抑制网络的红外弱小目标检测;史漫丽 等;《强激光与粒子束》;20110430;第23卷(第4期);第906-910页 *

Also Published As

Publication number Publication date
CN109033969A (en) 2018-12-18

Similar Documents

Publication Publication Date Title
US9025875B2 (en) People counting device, people counting method and people counting program
WO2021129569A1 (en) Human action recognition method
WO2018032660A1 (en) Moving target detection method and system
CN103258332B (en) A kind of detection method of the moving target of resisting illumination variation
Jia et al. A two-step approach to see-through bad weather for surveillance video quality enhancement
CN104463870A (en) Image salient region detection method
CN105574515B (en) A kind of pedestrian recognition methods again under non-overlapping visual field
CN107564022A (en) Saliency detection method based on Bayesian Fusion
CN106991686B (en) A kind of level set contour tracing method based on super-pixel optical flow field
US11042986B2 (en) Method for thinning and connection in linear object extraction from an image
Lian et al. A novel method on moving-objects detection based on background subtraction and three frames differencing
CN112784835A (en) Method and device for identifying authenticity of circular seal, electronic equipment and storage medium
CN109033969B (en) Infrared target detection method based on Bayesian saliency map calculation model
CN102013101A (en) Blind detection method of permuted and tampered images subjected to fuzzy postprocessing
Zhang et al. Dehazing with improved heterogeneous atmosphere light estimation and a nonlinear color attenuation prior model
Qin et al. A shadow removal algorithm for ViBe in HSV color space
Gao et al. Single image haze removal algorithm using pixel-based airlight constraints
Sanmiguel et al. Pixel-based colour contrast for abandoned and stolen object discrimination in video surveillance
Yang et al. Contrast limited adaptive histogram equalization for an advanced stereo visual slam system
CN115331008A (en) End-to-end target detection method based on target probability density graph
CN108205814B (en) Method for generating black and white contour of color image
CN114581475A (en) Laser stripe segmentation method based on multi-scale saliency features
CN111797761B (en) Three-stage smoke detection system, method and readable medium
Wang et al. Combined improved Frequency-Tuned with GMM algorithm for moving target detection
CN110503061B (en) Multi-feature-fused multi-factor video occlusion area detection method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant