CN109461164A - A kind of infrared small target detection method based on direction nuclear reconstitution - Google Patents

A kind of infrared small target detection method based on direction nuclear reconstitution Download PDF

Info

Publication number
CN109461164A
CN109461164A CN201811108681.8A CN201811108681A CN109461164A CN 109461164 A CN109461164 A CN 109461164A CN 201811108681 A CN201811108681 A CN 201811108681A CN 109461164 A CN109461164 A CN 109461164A
Authority
CN
China
Prior art keywords
image block
vector
image
infrared
small 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.)
Pending
Application number
CN201811108681.8A
Other languages
Chinese (zh)
Inventor
李彦胜
张永军
叶冬杰
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.)
Wuhan University WHU
Original Assignee
Wuhan University WHU
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 Wuhan University WHU filed Critical Wuhan University WHU
Priority to CN201811108681.8A priority Critical patent/CN109461164A/en
Publication of CN109461164A publication Critical patent/CN109461164A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10048—Infrared image

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

The present invention provides a kind of infrared small target detection method based on direction nuclear reconstitution, and raw video is divided into several image blocks first, calculates the local direction nuclear expression of each pixel, obtains the feature vector of each image block；Centered on each image block, the adjacent features vector set of N number of image block around is obtained, design factor vector, according to adjacent features vector set and coefficient vector, the reconstruction features vector for calculating each image block is indicated；To obtain the confidence map of entire infrared image；It is split according to threshold value, infrared small target is finally partitioned into from confidence map.The present invention encodes infrared image block using local feature kernel method, robustly calculates image internal characteristics structure and solves pixel-level image noise and ambiguity.Meanwhile the present invention utilizes adjacent features vector reconstruction features vector, enhances contrast by residual computations, achievees the purpose that prominent infrared image target while inhibiting ambient noise, can be used for improving infrared small target detection problem.

Description

A kind of infrared small target detection method based on direction nuclear reconstitution
Technical field
The invention belongs to image processing methods, and in particular to a kind of infrared small target detection based on direction nuclear reconstitution Method.
Background technique
Infrared small target detection plays a crucial role in infrared monitoring and early warning system.However it is pre- for early stage In alert situation, farther out due to distance, the pixel of target is too small in infrared image.Additionally due to infrared energy is with range attenuation Larger, this causes the noise of infrared target relatively low.Meanwhile there are certain picture noises and a variety of dry for infrared remote sensing image It disturbs.Therefore, infrared image is usually second-rate, and traditional characteristics of image descriptor is caused to be unable to satisfy infrared small target detection Demand.
In recent years, two kinds are generally divided into for the detection method of infrared small target at present: first is that by sequence infrared image, Target is confirmed according to the priori knowledge of the target characteristics of motion and intensity profile form；Second is that the spy for passing through single-frame images Sign, regards infrared small target as Gauss speck, inhibits background using suitable method.However due to the mesh of infrared sequence image Mark, is not available conventional feature descriptor and matching process usually, so that extracting time information is still from infrared sequence image It is so extremely challenging.And it is many to the Processing Algorithm of single-frame images at present, there are mean filter, median filtering, based on Mathematical Morphology Method, the method based on image entropy and the method based on extreme value theory.In addition, from the angle of vision attention, also Infrared small target detection method based on conspicuousness.However most of infrared small target detection methods based on significant property have height It computation complexity and is difficult to carry out parallel optimization.Currently, the method that existing certain methods use piecemeal, passes through simply vector Change original pixel intensities to calculate the feature descriptor of infrared image block.However, being described by the feature that original pixel intensities calculate Symbol is directly influenced by picture quality.Therefore, the infrared small target detection based on piecemeal needs to carry out robust to infrared image block Expression.
Summary of the invention
The present invention mainly solves second-rate infrared image present in the prior art and infrared target and background contrast Spend weaker, infrared image the problems such as there are certain picture noises；It provides a kind of based on regularization local direction nuclear reconstitution Infrared small target detection method can identify the target in infrared image simultaneously and inhibit ambient noise.
The technical scheme adopted by the invention is that: a kind of infrared small target detection method based on direction nuclear reconstitution, including Following steps:
Step 1, raw video is divided into each and every one several image blocks first, the pixel size for including in each image block is P；
Step 2, according to local direction core, the local direction nuclear expression of each pixel is calculated, each image block is obtained Feature vector；
Step 3, centered on each image block, the adjacent features vector set of N number of image block around, design factor are obtained Vector, according to adjacent features vector set and coefficient vector, the reconstruction features vector for calculating each image block is indicated；
Step 4, by the residual error of reconstruction features vector and original image block eigenvector, calculate each center image block with Local contrast between adjacent image block obtains the confidence level of each pixel in image block, finally obtains entire infrared image Confidence map；
Step 5, it is split according to threshold value, infrared small target is partitioned into from confidence map.
Further, the specific implementation that the feature vector of image block is calculated in step 2 is as follows,
Step 2.1, before calculating local direction core, the gradient in global field is normalized first；
Step 2.2, the feature vector for calculating image block is assessed according to local direction:
Wherein, with xiCentered on image block can indicate W (xi)={ x1,…,xi,…,xP, in xiCoordinate on position Intensity can be expressed as I (xi), xjAnd xiRefer to different location of pixels, xj∈W(xi), K (xi,xj) refer to xiCentered on The feature vector of image block, h are global smoothing parameter, covariance matrix CjBy with coordinate vectorCentered on neighbour Estimated by the set of near space gradient vector,
Wherein M indicates the number of pixels of proximity space, I1() and I2() is I () along the one of trunnion axis and vertical axis Order derivative.
Further, being implemented as follows for image block reconstruction features vector is calculated in step 3,
Step 3.1, according to step 2.2 feature vector calculated, with central feature vector fc∈RP×1Centered on it is N number of Adjacent feature vector set can be expressed as F=[f1,…,fN], w ∈ RN×1For linear combination coefficient, optimization object function is obtained To coefficient vector w=(FT·F+λ·I)-1·FT·fc,
Wherein λ is regularization term, is set as 10；
Step 3.2, according to adjacent features vector F=[f1,…,fN] and coefficient vector w, archicenter feature vector fcIt can To be redeveloped into fr,
fr=Fw=F (FT·F+λ·I)-1·FT·fc(formula three)
Wherein, FTIt is the transposition of F.
Further, the specific implementation of the confidence level of each pixel is as follows in step 4 calculating image block,
As unit of pixel, successively calculates local direction core image block characteristics centered on current pixel and its is neighbouring Local direction core image block characteristics, obtain the confidence level of each pixel:
ALCM-LSK(fc, F)=| | fc-fr||2(formula
=| | fc-F·(FT·F+λ·I)-1·FT·fc||2
Wantonly).
Further, step 5 is split using adaptive threshold according to the following formula,
Th=u+k σ (formula 5)
Wherein u is the mean value of confidence map, and σ is the standard deviation of confidence map, and k is an empirical, is set as 10.With it is existing Technology is compared, the advantages of the present invention: since the quality of infrared image is usually poor (for example, infrared image is frequent By infrared sensor noise pollution), we encode infrared image block using local feature kernel method, and local feature core can Robustly to calculate image internal characteristics structure and solve pixel-level image noise and ambiguity.Meanwhile the present invention is using neighbouring Feature vector reconstruction features vector enhances contrast by residual computations, reaches prominent infrared image target while inhibiting background The purpose of noise can be used for improving infrared small target detection problem.
Detailed description of the invention
Fig. 1: for the overview flow chart of the embodiment of the present invention；Wherein, Dark grey small circle includes infrared small target, light grey Great circle indicates ambient noise.1st, the 3rd and the 5th image show original image, confidence map and final detection result respectively.2nd, The 3-D view of the 1st, the 3rd and the 5th image is presented in 4th and the 6th image respectively.
Fig. 2: for the reconstruction local direction core feature schematic diagram of the embodiment of the present invention.
Specific embodiment
Understand for the ease of those of ordinary skill in the art and implement the present invention, with reference to the accompanying drawings and embodiments to this hair It is bright to be described in further detail, it should be understood that implementation example described herein is merely to illustrate and explain the present invention, not For limiting the present invention.
Referring to Fig.1, a kind of infrared small target detection method based on direction nuclear reconstitution provided by the invention, including following step It is rapid:
Step 1: raw video being divided into several image blocks first, the pixel size for including in each image block is P, such as Shown in Fig. 2, wherein s is tile size, and t is the overlapping dimension of image block.
The position of each pixel of infrared image is by coordinate vectorIt indicates, in xiCoordinate intensity on position It can be expressed as I (xi), with xiCentered on image block can indicate W (xi)={ x1,…,xi,…,xP}。
Step 2: calculating the local direction nuclear expression of each pixel according to local direction core see Fig. 2, obtain each The feature vector of a image block.
Step 2.1: before calculating local direction core, normalizing being carried out to the gradient in global field (i.e. entire image) first Change.
Step 2.2: the feature vector for calculating image block is assessed according to local direction:
Wherein with xiCentered on image block can indicate W (xi)={ x1,…,xi,…,xP, in xiCoordinate on position Intensity can be expressed as I (xi), xjAnd xiRefer to different location of pixels, xj∈W(xi), K (xi,xj) refer to xiCentered on The feature vector of image block, h are global smoothing parameter, are set as 0.2, covariance matrix CjBy with positionFor in Estimated by the set of the proximity space gradient vector of the heart,
Wherein M indicates the number of pixels of proximity space, I1() and I2() is I () along the one of trunnion axis and vertical axis Order derivative.Step 3: centered on each image block, obtaining the adjacent features vector set of N number of image block around, design factor Vector, according to adjacent features vector set and coefficient vector, the reconstruction features vector for calculating each image block is indicated.
Step 3.1: according to step 2.2 feature vector calculated, with fc∈RP×1Centered on N number of adjacent feature vector Set can be expressed as F=[f1,…,fN], w ∈ RN×1For linear combination coefficient, optimization object function obtains coefficient vector w= (FT·F+λ·I)-1·FT·fc:
Wherein λ is regularization term, it controls the tradeoff between estimated bias and the variance of model of fit.
Step 3.2: according to adjacent features vector F=[f1,…,fN] and coefficient vector w, archicenter feature vector fcIt can To be redeveloped into fr:
fr=Fw=F (FT·F+λ·I)-1·FT·fc(formula three)；
Wherein, FTIt is the transposition of F.
Step 4: by the residual error of reconstruction features vector and original image block eigenvector, calculate each center image block with Local contrast between adjacent image block obtains the confidence level of each pixel in image block, finally obtains entire infrared image Confidence map.
As unit of pixel, successively calculates local direction core image block characteristics centered on current pixel and its is neighbouring Local direction core image block characteristics, obtain the confidence level of each pixel:
ALCM-LSK(fc, F)=| | fc-fr||2(formula is wantonly)；
=| | fc-F·(FT·F+λ·I)-1·FT·fc||2
Step 5: being split according to threshold value, infrared small target is detected from confidence map.According to the following formula using adaptive Threshold value is split:
Th=u+k σ (formula 5)；
Wherein u is the mean value of confidence map, and σ is the standard deviation of confidence map, and k is an empirical, is set as 10.
It should be understood that the part that this specification does not elaborate belongs to the prior art.
It should be understood that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered to this The limitation of invention patent protection range, those skilled in the art under the inspiration of the present invention, are not departing from power of the present invention Benefit requires to make replacement or deformation under protected ambit, fall within the scope of protection of the present invention, this hair It is bright range is claimed to be determined by the appended claims.

Claims (5)

1. a kind of infrared small target detection method based on direction nuclear reconstitution, which comprises the steps of:
Step 1, raw video is divided into each and every one several image blocks first, the pixel size for including in each image block is P；
Step 2, according to local direction core, the local direction nuclear expression of each pixel is calculated, the spy of each image block is obtained Levy vector；
Step 3, centered on each image block, obtain the adjacent features vector set of N number of image block around, design factor to Amount, according to adjacent features vector set and coefficient vector, the reconstruction features vector for calculating each image block is indicated；
Step 4, by the residual error of reconstruction features vector and original image block eigenvector, calculate each center image block with it is neighbouring Local contrast between image block obtains the confidence level of each pixel in image block, finally obtains setting for entire infrared image Letter figure；
Step 5, it is split according to threshold value, infrared small target is partitioned into from confidence map.
2. a kind of infrared small target detection method based on direction nuclear reconstitution as described in claim 1, it is characterised in that: step The specific implementation that the feature vector of image block is calculated in 2 is as follows,
Step 2.1, before calculating local direction core, the gradient in global field is normalized first；
Step 2.2, the feature vector for calculating image block is assessed according to local direction:
Wherein, with xiCentered on image block can indicate W (xi)={ x1,…,xi,…,xP, in xiCoordinate intensity on position It can be expressed as I (xi), xjAnd xiRefer to different location of pixels, xj∈W(xi), K (xi,xj) refer to xiCentered on image The feature vector of block, h are global smoothing parameter, covariance matrix CjBy with coordinate vectorCentered on it is neighbouring empty Between gradient vector set estimated by,
Wherein M indicates the number of pixels of proximity space, I1() and I2() is that I () is led along the single order of trunnion axis and vertical axis Number.
3. a kind of infrared small target detection method based on direction nuclear reconstitution as claimed in claim 2, it is characterised in that: step Being implemented as follows for image block reconstruction features vector is calculated in 3,
Step 3.1, according to step 2.2 feature vector calculated, with central feature vector fc∈RP×1Centered on it is N number of adjacent Feature vector set can be expressed as F=[f1,…,fN], w ∈ RN×1For linear combination coefficient, optimization object function is Number vector w=(FT·F+λ·I)-1·FT·fc,
Wherein λ is regularization term；
Step 3.2, according to adjacent features vector F=[f1,…,fN] and coefficient vector w, archicenter feature vector fcIt can weigh It builds as fr,
fr=Fw=F (FT·F+λ·I)-1·FT·fc(formula three)
Wherein, FTIt is the transposition of F.
4. a kind of infrared small target detection method based on direction nuclear reconstitution as claimed in claim 3, it is characterised in that: step The specific implementation of the confidence level of each pixel is as follows in 4 calculating image blocks,
As unit of pixel, the local direction core image block characteristics centered on current pixel and its neighbouring office are successively calculated Portion direction core image block characteristics, obtain the confidence level of each pixel:
5. a kind of infrared small target detection method based on direction nuclear reconstitution as claimed in claim 4, it is characterised in that: step 5 are split using adaptive threshold according to the following formula,
Th=u+k σ (formula 5)
Wherein u is the mean value of confidence map, and σ is the standard deviation of confidence map, and k is an empirical.
CN201811108681.8A 2018-09-21 2018-09-21 A kind of infrared small target detection method based on direction nuclear reconstitution Pending CN109461164A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811108681.8A CN109461164A (en) 2018-09-21 2018-09-21 A kind of infrared small target detection method based on direction nuclear reconstitution

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811108681.8A CN109461164A (en) 2018-09-21 2018-09-21 A kind of infrared small target detection method based on direction nuclear reconstitution

Publications (1)

Publication Number Publication Date
CN109461164A true CN109461164A (en) 2019-03-12

Family

ID=65606795

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811108681.8A Pending CN109461164A (en) 2018-09-21 2018-09-21 A kind of infrared small target detection method based on direction nuclear reconstitution

Country Status (1)

Country Link
CN (1) CN109461164A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110796028A (en) * 2019-10-11 2020-02-14 武汉大学 Unmanned aerial vehicle image small target detection method and system based on local adaptive geometric transformation
CN112037145A (en) * 2020-08-31 2020-12-04 成都信息工程大学 Medical MRI (magnetic resonance imaging) image up-sampling method based on self-adaptive local steering nucleus

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107392885A (en) * 2017-06-08 2017-11-24 江苏科技大学 A kind of method for detecting infrared puniness target of view-based access control model contrast mechanism
CN108038856A (en) * 2017-12-22 2018-05-15 杭州电子科技大学 Based on the infrared small target detection method for improving Multi-scale Fractal enhancing
CN108062523A (en) * 2017-12-13 2018-05-22 苏州长风航空电子有限公司 A kind of infrared remote small target detecting method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107392885A (en) * 2017-06-08 2017-11-24 江苏科技大学 A kind of method for detecting infrared puniness target of view-based access control model contrast mechanism
CN108062523A (en) * 2017-12-13 2018-05-22 苏州长风航空电子有限公司 A kind of infrared remote small target detecting method
CN108038856A (en) * 2017-12-22 2018-05-15 杭州电子科技大学 Based on the infrared small target detection method for improving Multi-scale Fractal enhancing

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YANSHENG LI等: "《Robust infrared small target detection using local steering kernel reconstruction》", 《PATTERN RECOGNITION》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110796028A (en) * 2019-10-11 2020-02-14 武汉大学 Unmanned aerial vehicle image small target detection method and system based on local adaptive geometric transformation
CN110796028B (en) * 2019-10-11 2021-08-17 武汉大学 Unmanned aerial vehicle image small target detection method and system based on local adaptive geometric transformation
CN112037145A (en) * 2020-08-31 2020-12-04 成都信息工程大学 Medical MRI (magnetic resonance imaging) image up-sampling method based on self-adaptive local steering nucleus

Similar Documents

Publication Publication Date Title
CN107045629B (en) Multi-lane line detection method
Xia et al. Infrared small target detection based on multiscale local contrast measure using local energy factor
CN109816641B (en) Multi-scale morphological fusion-based weighted local entropy infrared small target detection method
CN106023185B (en) A kind of transmission facility method for diagnosing faults
CN110378308B (en) Improved port SAR image near-shore ship detection method based on fast R-CNN
CN107392885A (en) A kind of method for detecting infrared puniness target of view-based access control model contrast mechanism
CN109272489A (en) Inhibit the method for detecting infrared puniness target with multiple dimensioned local entropy based on background
CN109460764B (en) Satellite video ship monitoring method combining brightness characteristics and improved interframe difference method
CN107705288A (en) Hazardous gas spillage infrared video detection method under pseudo- target fast-moving strong interferers
Liu et al. A night pavement crack detection method based on image‐to‐image translation
CN110245600B (en) Unmanned aerial vehicle road detection method for self-adaptive initial quick stroke width
CN101976436A (en) Pixel-level multi-focus image fusion method based on correction of differential image
CN102096824A (en) Multi-spectral image ship detection method based on selective visual attention mechanism
CN110400274B (en) Infrared image enhancement method for vehicle-mounted infrared pedestrian detection
CN113920436A (en) Remote sensing image marine vessel recognition system and method based on improved YOLOv4 algorithm
CN103679694A (en) Ship small target detection method based on panoramic vision
CN110619647A (en) Method for positioning fuzzy region of image based on combination of edge point frequency domain and spatial domain characteristics
CN110751635A (en) Oral cavity detection method based on interframe difference and HSV color space
CN109461164A (en) A kind of infrared small target detection method based on direction nuclear reconstitution
CN109886336A (en) A kind of object detection method and system based on warship basic image
Fu et al. Infrared sea-sky line detection utilizing self-adaptive Laplacian of Gaussian filter and visual-saliency-based probabilistic Hough transform
CN111521245A (en) Liquid level control method based on visual analysis
CN115272876A (en) Remote sensing image ship target detection method based on deep learning
CN113205494B (en) Infrared small target detection method and system based on adaptive scale image block weighting difference measurement
CN117079117B (en) Underwater image processing and target identification method and device, storage medium and electronic equipment

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20190312

RJ01 Rejection of invention patent application after publication