CN106530271B - A kind of infrared image conspicuousness detection method - Google Patents
A kind of infrared image conspicuousness detection method Download PDFInfo
- Publication number
- CN106530271B CN106530271B CN201610873518.5A CN201610873518A CN106530271B CN 106530271 B CN106530271 B CN 106530271B CN 201610873518 A CN201610873518 A CN 201610873518A CN 106530271 B CN106530271 B CN 106530271B
- Authority
- CN
- China
- Prior art keywords
- block
- image
- notable
- local
- infrared image
- 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
Links
Classifications
-
- 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
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Abstract
The invention discloses a kind of infrared image conspicuousness detection method, steps are as follows: firstly, carrying out multiple dimensioned local rarefaction representation to infrared image, obtains the image saliency map based on multiple dimensioned local rarefaction representation;Secondly, obtaining the image saliency map measured based on improved local contrast to the local contrast measurement processing that infrared image improves;Finally, above two notable figure is merged according to certain weight, obtain the final notable figure of original infrared image.The present invention has the advantages that can guarantee that the salient region of infrared image is completely highlighted, while background is adequately suppressed, and is compared with other methods, the verification and measurement ratio of conspicuousness is higher, and false alarm rate is lower.
Description
Technical field
It is the invention belongs to technical field of image processing, in particular to a kind of based on multiple dimensioned local rarefaction representation and improved
The infrared image conspicuousness detection method of local contrast measurement.
Background technique
The conspicuousness detection of infrared image be intended to allow computer rapidly identify in infrared image containing useful information and
The salient region paid close attention to the most in human vision.Conspicuousness detection the result is that highlighting marking area completely
Come, while background is inhibited well, subsequent in order to image is further processed.
In present image procossing research, the serial of methods of conspicuousness detection is successfully applied to visible images.Example
Such as, a kind of image significance detection method based on edge of publication number CN102609911A, passes through the inspection of significant point and marginal point
It surveys to determine the marking area in image, but this method does not make full use of each scale feature of image, therefore, what is detected is aobvious
Write property region there are it is imperfect situations such as.
A kind of conspicuousness detection method based on contrast Yu angle point Minimum Convex Closure of publication number CN103927758A, first root
Global notable figure is calculated according to the global contrast of super-pixel, then calculates local notable figure, last root using center-surrounding operator
It is filled into subregional interference in the middle part of background according to the position and size of the Minimum Convex Closure estimation main target of Harris angle point, so that
Main target is protruded.The method can equably protrude target, but it excessively relies on the accuracy of convex closure, in prominent target
While, a big chunk zone errors are also revealed in convex closure, therefore reduce the accuracy of conspicuousness detection.
Though all coming with some shortcomings, being embodied in, schemed in short, present target conspicuousness detection method is more
The interference of background as in, algorithm is more sensitive to noise, is difficult to take into account the scale feature of the local feature of target and target, significantly
Property Detection accuracy it is low, false alarm rate is high.In addition, infrared image is directed to, if directlying adopt the existing image for visible light
Conspicuousness detection algorithm carries out the detection of salient region, due to not accounting for infrared image special imaging mechanism and image
The special nature of middle salient region, causes algorithm performance to substantially reduce.
Summary of the invention
Goal of the invention: aiming at the problems existing in the prior art, the present invention provides one kind towards infrared image, can guarantee red
The salient region of outer image is completely highlighted, while background is adequately suppressed, strong antijamming capability, and is detected quasi-
The high infrared image conspicuousness detection method of true rate.
Technical solution: in order to solve the above technical problems, the present invention provides a kind of infrared image conspicuousness detection method, including
Following steps:
Step 1: local rarefaction representation is carried out to original infrared image, local search frame is set as a × a, and wherein a is positive
Then integer obtains a kind of notable figure;
Step 2: carrying out a certain proportion of diminution for original infrared image, the image after being reduced, then by local search
Frame size is set as b × b, and wherein b is positive integer, then carries out local rarefaction representation, obtains two class notable figures, and two classes are shown
The size of work figure is reduced into consistent with original infrared image size;
Step 3: by a kind of notable figure that step 1 obtains and the two class notable figures that step 2 obtains according to certain weight
Ratio is weighted fusion, obtains the notable figure based on multiple dimensioned local rarefaction representation;
Step 4: the local contrast measurement processing that original infrared image is improved is obtained based on improved part
The notable figure of contrast measurement;
Step 5: by step 3 obtain based on multiple dimensioned local rarefaction representation notable figure and step 4 obtain based on changing
Into local contrast measurement notable figure according to certain weight ratio be weighted fusion, obtain final notable figure.
Wherein, in step 3, the two class notable figures that a kind of notable figure and step 2 that step 1 obtains obtain are according to 1:1's
Weight ratio is weighted fusion.
Wherein, in step 5, what what step 3 obtained obtained based on multiple dimensioned local rarefaction representation notable figure and step 4
Fusion is weighted according to the weight ratio of 1:4 based on improved local contrast measurement notable figure.
Further, the concrete operations of image saliency map are obtained using local sparse representation method in step 1 and step 2
Process is as follows:
Step a: given one with element x ∈ R2Centered on input picture, whereinIt indicates centered on x simultaneously
And include the block of n element, set a size asLocal search frame, local search frame centered on x,
K is the number of non-central element in local search frame,For the element centered on non-central element each in search box
Block, wherein i=1,2...k are therefore, as follows with the non-central piece of expression formula come linear expression central block:
Wherein,It is the matrix of non-central element block,It is linear
Combined coefficient vector;
Step b: the expression formula of central block in step a is expressed as following equatioies:
WhereinFor the coefficient vector comprising a small amount of nonzero element, solved by orthogonal matching pursuit comprising a small amount of non-zero entry
The coefficient vector of elementCentral blockThe linear combination obtained after rarefaction representation is as follows:
Step c: saliency can be obtained by the residual error of reconstructed image and original image, with following formula come simple table
Show saliency:
Wherein | | | |2Indicate L2Norm;
Step d: local rarefaction representation is carried out to whole image, i.e., from top to bottom by search box, from left to right search is entire
Image, and rarefaction representation is carried out to each element in local search frame, finally ask the residual error of equation in step c to obtain whole
The notable figure of a image.
Further, the solution procedure that improved local contrast measures in step 4, concrete operations are as follows:
Step 4.1: the processing of dimensional Gaussian differential filtering being carried out to target, the very noisy in wiping out background, so that background is high
Luminance area gray value becomes larger;
Step 4.2: set a size as the local search frame of W × W to infrared image from top to bottom, from left to right carry out
Search, carries out piecemeal for whole image, obtains a series of sub-block, the moving step length of search box is set when carrying out piecemeal,
Then these blocks being performed corresponding processing, each of given piece of value is the gray value of all pixels in block, it is defined as follows:
Wherein, block (s, t) is a sub-block, and u is its value, and m × n is the size of block, and pix (i, j) is to belong to block
Block (s, t) inner pixel, also, f (pix (i, j)) is the corresponding pixel value of pix (i, j).
Step 4.3: defining the value of block according to step 4.2, then operate whole image with block, by whole image
The block of the formation matrix new as one, is defined as U for this matrix, and is defined as follows by the new matrix that block forms:
U (i, j)=u (block (i, j)), i=1,2...M, j=1,2...N
Wherein, M, N respectively indicate the number of whole image both vertically as well as horizontally block, i.e., the line number of this new matrix and
The size of columns, after piecemeal is handled, whole image will carry out the processing between block, and more traditional local contrast, which measures, to be calculated
It is handled between the pixel that method carries out, efficiency will be higher.
Step 4.4: in the new matrix of sub-block composition, the average gray for defining a sub-block block (s, t) is
u0, InFor the maximum value of block (s, t) gray value, following relational expression is then formed:
In=max (f (pix (i, j))), pix (i, j) ∈ block (s, t)
Again it is the image block of sub-block block (s, t) three times with a size, block centered on block (s, t), and looks for
To 8 adjacent blocks of block (s, t) in this image block and their corresponding average value u in U matrix1~u8, by
This, improved local contrast measurement is defined as foloows:
When the central block to be detected is object block, max (u at this timei) < In, ILCM > u at this time0, target will be added at this time
By force;When the central block to be detected is background block, max (u at this timei)≥In, ILCM≤u at this time0, background parts are suppressed at this time;
By this improved local contrast measurement operation, salient region is reinforced, and background is suppressed, and is finally preferably shown
Write figure.
Wherein, the filtering of difference of Gaussian described in step 4.1 is equivalent to one and can remove is retained in original image
The bandpass filter of every other frequency information except the frequency come, can be with the very noisy in wiping out background, so that background is high
Luminance area gray value becomes larger.
Compared with the prior art, the advantages of the present invention are as follows:
Even if this method is still in infrared image under the very strong situation of brightness irregularities or background interference of salient region
Can completely, accurately highlight salient region, inhibit background clutter, guarantee that conspicuousness verification and measurement ratio is high, false alarm rate is low, and existing
There is method to compare, is a good method for the infrared image conspicuousness detection under complex background.
Detailed description of the invention
Fig. 1 is structural block diagram of the invention;
Fig. 2 is the structural schematic diagram in specific embodiment;
Fig. 3 is the structural block diagram that notable figure is realized in improved local contrast measurement in specific embodiment;
Fig. 4 is that notable figure merges schematic diagram in specific embodiment;
Fig. 5 is the final notable figure obtained in specific embodiment.
Specific embodiment
With reference to the accompanying drawings and detailed description, the present invention is furture elucidated.
Infrared image proposed by the present invention based on multiple dimensioned local rarefaction representation and the measurement of improved local contrast is aobvious
Work property new detecting method obtains firstly, carrying out multiple dimensioned local rarefaction representation to infrared image based on multiple dimensioned local sparse table
The image saliency map shown;Secondly, obtaining the local contrast measurement processing that infrared image improves based on improved part
The image saliency map of contrast measurement;Finally, merging above two notable figure according to certain weight, finally shown
Write figure.
Wherein, described that multiple dimensioned local rarefaction representation is carried out to infrared image, i.e., not only original size image is carried out
Local rarefaction representation also carries out local rarefaction representation to the original image reduced according to a certain percentage, and at the same time, part is dilute
Dredging indicates that search box also carries out size change over according to a certain percentage, then by the original image of different size ratio, corresponds to different
The local rarefaction representation result that the local search frame of size carries out is weighted fusion (the ratio between weight is 1:1), obtains base
In the image saliency map of multiple dimensioned local rarefaction representation.
The local contrast measurement processing that the infrared image improves is to carry out piecemeal to whole image, utilizes
Image block associated information calculation local contrast.Compared with the existing method measured based on local contrast, carrying out significantly
Property calculate when, the maximum brightness information of image block is not only utilized in improved local contrast measurement method, but also utilizes simultaneously
The dimension information of picture centre block, so that the significantly more efficient image for obtaining measuring based on improved local contrast is significant
Figure.
Finally, two kinds of specific images that multiple dimensioned local rarefaction representation and improved local contrast measurement are obtained carry out
Fusion.Selection uses Weighted Fusion in the present invention, and the ratio between weight of above-mentioned two classes notable figure is 1:4, is guaranteeing to obtain in this way
While complete salient region, background can also be inhibited well.
Infrared image conspicuousness detection method based on multiple dimensioned local rarefaction representation and the measurement of improved local contrast,
Integrated operation process is as shown in Fig. 2, specific implementation step is as follows:
1) local rarefaction representation is carried out to original infrared image, local search frame is set as 12 × 12, obtains notable figure 1;
2) original infrared image is subjected to a certain proportion of diminution (ratio that reduces is 0.7), the image after being reduced, then
Local search frame size is set as 8 × 8, carries out local rarefaction representation, obtains (the notable figure size reduction at this time of notable figure 2
It is consistent with original infrared image size);
3) notable figure 1 and notable figure 2 are weighted fusion, and the ratio between their weight is 1:1, is obtained based on multiple dimensioned
The notable figure of local rarefaction representation;
4) the local contrast measurement processing for improving original infrared image is obtained based on improved local contrast
The notable figure of measurement;
5) added based on multiple dimensioned local rarefaction representation notable figure with based on improved local contrast measurement notable figure
Power fusion, the ratio between weight are 1:4, obtain final notable figure.
The specific operation process for obtaining image saliency map using multiple dimensioned local sparse representation method is as follows:
1) local rarefaction representation principle: given one with element x ∈ R2Centered on input picture, it is currentCome
Indicate centered on x and the block comprising n element, set a size asLocal search frame, equally
This search box is centered on x, and k is the number of non-central element in this local search frame,For with search box
In centered on each non-central element element block, wherein i=1,2...k therefore can be with non-central pieces come linear expression
Our central block, expression formula are as follows:
Wherein,It is the matrix of non-central element block,It is linear
Combined coefficient vector.
In order to enable coefficient can more preferable more efficient linear expression input picture, it is only a small number of in Required coefficient vector
Expression formula (1) is therefore expressed as following equation by nonzero element:
To accurately solve the coefficient vector comprising a small amount of nonzero element of equation (2)Pass through orthogonal matching pursuit
To solve.Central blockThe linear combination obtained after rarefaction representation is as follows:
Saliency can be obtained by the residual error of reconstructed image and original image.Therefore, with following formula come simple table
Show saliency:
Wherein | | | |2Indicate L2Norm.
Local rarefaction representation is carried out to whole image and that is, from top to bottom by search box from left to right searches for whole image, and
And rarefaction representation is carried out to each element in local search frame, finally ask the residual error of equation (4) to obtain the aobvious of whole image
Write figure.
2) multiple dimensioned local rarefaction representation: since the profile that local rarefaction representation generally highlights salient region is stronger,
And inside display is weaker, better effect, the present invention use multiple dimensioned local rarefaction representation in order to obtain: not only to original size
Image carries out local rarefaction representation, also carries out local rarefaction representation to the original image reduced according to a certain percentage, same with this
When, local rarefaction representation search box also carries out size change over according to a certain percentage, then by the original image of different size ratio,
The local rarefaction representation result that the local search frame of corresponding different size ratio carries out is merged, and is obtained based on multiple dimensioned part
The image saliency map of rarefaction representation.Not only profile is fine for the notable figure, and inside also can be highlighted out well.By multiple
Experiment obtains: 1. original image size is constant, and local search frame is 12 × 12, and 2. image down is the 0.7 of original image ratio, part
Search box is 8 × 8, and the two merges the notable figure that can obtain best effects, so that the profile of salient region and inside are all
It can be highlighted out well.
Based on the image saliency map of improved local contrast measurement, solution procedure is as shown in figure 3, it uses block and block
Between operation can rapidly and effectively obtain being based on improved office using image block associated information calculation local contrast
The image saliency map of portion's contrast measurement.Concrete operations are as follows:
1) processing of dimensional Gaussian differential filtering is carried out to target, difference of Gaussian filtering, which is equivalent to one, can remove in addition to that
The bandpass filter of every other frequency information except the frequency being retained in original image a bit.Therefore, it can filter
Except the very noisy in background, so that background high-brightness region gray value becomes larger.
2) set a size as the local search frame of W × W to infrared image from top to bottom, from left to right scan for,
Whole image is subjected to piecemeal, obtains a series of sub-block, the moving step length of search box is set when carrying out piecemeal.To this
A little blocks perform corresponding processing, and each of given piece of value is the gray value of all pixels in block, are defined as follows:
Wherein, block (s, t) is a sub-block, and u is its value, and m × n is the size of block, and pix (i, j) is to belong to block
Block (s, t) inner pixel, also, f (pix (i, j)) is the corresponding pixel value of pix (i, j).
3) value of block is defined, so that it may operate whole image with block, the block that whole image is formed is as one
This matrix is defined as U by a new matrix, and is defined as follows by the new matrix that block forms:
U (i, j)=u (block (i, j)), i=1,2...M, j=1,2...N (6)
Wherein, M, N respectively indicate the number of whole image both vertically as well as horizontally block, i.e., the line number of this new matrix and
The size of columns.After piecemeal is handled, whole image will carry out the processing between block, and more traditional local contrast, which measures, to be calculated
It is handled between the pixel that method carries out, efficiency will be higher.
4) when piecemeal, may include in a block be entirely target, entirely background or is had.It is original infrared
Salient region is usually brighter than the background in image, can be highlighted by given threshold;But works as and encounter than conspicuousness area
The brighter background in domain, will cause huge error.Therefore, present invention employs following improved local contrast measurement methods:
In the new matrix of sub-block composition, the average gray for defining a sub-block block (s, t) is u0, InFor
The maximum value of block (s, t) gray value.Following relational expression can be formed:
In=max (f (pix (i, j))), pix (i, j) ∈ block (s, t) (7)
Again it is the image block of sub-block block (s, t) three times with a size, block centered on block (s, t), and looks for
To 8 adjacent blocks of block (s, t) in this image block and their corresponding average value u in U matrix1~u8, by
This, improved local contrast measurement is defined as foloows:
Generally, it is measured according to local contrast it is found that when the central block to be detected is object block, at this time max (ui) < In,
So can determine ILCM > u at this time0, target will be reinforced at this time.When the central block to be detected is background block, max at this time
(ui)≥In, so can determine ILCM≤u at this time0, background parts are suppressed at this time.By this improved local contrast
Degree measurement operation, salient region are reinforced, and background is suppressed, and finally preferably obtains notable figure.
After above two method is respectively processed image, respectively obtain based on multiple dimensioned local rarefaction representation
Both notable figures are merged (as shown in Figure 4), are obtained by notable figure and the notable figure measured based on improved local contrast
To final infrared image notable figure.Concrete operations are as follows:
1) by the notable figure based on multiple dimensioned local rarefaction representation and the notable figure based on the measurement of improved local contrast
It is denoted as S respectivelyNAnd SILCM。
2) fusion is weighted to above-mentioned two classes notable figure.When fusion, the ratio between weight of above-mentioned two classes notable figure is 1:4,
Obtain the final notable figure S of infrared image.
S=w1SN+w2SILCM (9)
Wherein, w1And w2It respectively indicates and obtains by multiple dimensioned local rarefaction representation and improved local contrast measurement
The weight of notable figure.And w1: w2When for 1:4, obtained notable figure effect is best.
In conjunction with simulated conditions, the present invention will be further described with result
1) simulated conditions
This experiment is in PC machine (Intel Core, dominant frequency 2.6GHZ, memory 4GB) in, Matlab2013b programmed environment
In to the hot pedestrian's database of Oregon State University (OSU) and oneself collect the figure chosen in the infrared picture data library of production
Piece is tested.Take when multiple dimensioned local rarefaction representation processing: 1. original image size is constant, and local search frame is 12 ×
12,2. image down is the 0.7 of original image ratio, and the scale that local search frame is 8 × 8 is tested.It will be based on multiple dimensioned office
When the notable figure that portion's rarefaction representation and improved local contrast measurement respectively obtain is weighted fusion, the ratio between weight is set as
1:4。
2) emulation content and interpretation of result
It is based on multiple dimensioned local rarefaction representation using proposed by the present invention as shown in figure 5, being provided by an example and changes
Into local contrast measurement infrared image conspicuousness testing result, to utilize multiple dimensioned local rarefaction representation in pilot process
The notable figure respectively obtained with improved local contrast measurement processing.
In this experiment, three kinds of different types of pictures are picked, are single body respectively, more human bodies, non-human picture, from
From the point of view of experimental result, based on the notable figure that multiple dimensioned local rarefaction representation obtains, salient region can be detected completely
Come, but when background luminance is uneven or too bright, there are stronger background interferences in notable figure;On the contrary, using improved
When local contrast measurement method carries out notable figure solution, background clutter can be inhibited well, but salient region exists
In the stronger situation of background luminance, what is highlighted is imperfect;Finally, the present invention is by multiple dimensioned local rarefaction representation and improved part
Contrast measurement, which combines, carries out conspicuousness detection, and no matter original infrared image has relatively simple background or complicated back
Scape, background luminance are uniform or non-uniform, can completely protrude salient region, while can be good at inhibiting back
Scape guarantees that conspicuousness Detection accuracy is high, false alarm rate is low.
In conclusion combining local rarefaction representation by the way that original infrared image is carried out multiple dimensioned local rarefaction representation
The profile of infrared image salient region can be highlighted well and multiple dimensioned can preferably highlight infrared image conspicuousness
The internal feature in region guarantees completely to highlight salient region;Meanwhile the above-mentioned improved local contrast of fusion is surveyed
Amount can be good at inhibiting the background interference in infrared image, so that strong antijamming capability when conspicuousness detects.The present invention can
Independent, complete, anti-background interference infrared image conspicuousness testing result is obtained, and with conspicuousness observed by human eye
Region is more nearly.
It is sub that the above description is only an embodiment of the present invention, is not intended to restrict the invention, the present invention is for numerical value meter
The accurately display of research object is especially suitable in calculation.It is all within principle of the invention, made equivalent replacement should be included in
Within protection scope of the present invention.The content that the present invention is not elaborated, which belongs to well known to this professional domain technical staff, to be had
Technology.
Claims (6)
1. a kind of infrared image conspicuousness detection method, which comprises the steps of:
Step 1: carrying out local rarefaction representation to original infrared image, and local search frame is set as a × a, and wherein a is positive integer,
Then a kind of notable figure is obtained;
Step 2: carrying out a certain proportion of diminution for original infrared image, the image after being reduced, then local search frame is big
Small to be set as b × b, wherein b is positive integer, then carries out local rarefaction representation, obtains two class notable figures, and by two class notable figures
Size be reduced into it is consistent with original infrared image size;
Step 3: by a kind of notable figure that step 1 obtains and the two class notable figures that step 2 obtains according to certain weight ratio
It is weighted fusion, obtains the notable figure based on multiple dimensioned local rarefaction representation;
Step 4: the local contrast measurement processing that original infrared image is improved is obtained based on improved local contrast
Spend the notable figure of measurement;
Step 5: by step 3 obtain based on multiple dimensioned local rarefaction representation notable figure and step 4 obtain based on improved
Local contrast measures notable figure and is weighted fusion according to certain weight ratio, obtains final notable figure.
2. a kind of infrared image conspicuousness detection method according to claim 1, it is characterised in that: in the step 3,
The two class notable figures that a kind of notable figure and step 2 that step 1 obtains obtain are weighted fusion according to the weight ratio of 1:1.
3. a kind of infrared image conspicuousness detection method according to claim 1, it is characterised in that: in the step 5,
What step 3 obtained is surveyed based on what multiple dimensioned local rarefaction representation notable figure and step 4 obtained based on improved local contrast
It measures notable figure and is weighted fusion according to the weight ratio of 1:4.
4. according to claim 1 to a kind of infrared image conspicuousness detection method described in one of 3, it is characterised in that: the step
Rapid one and step 2 in obtain image saliency map using local sparse representation method specific operation process it is as follows:
Step a: giving an input picture, usesIt indicates centered on x and the image block comprising n element, sets
One size isLocal search frame, local search frame centered on x, k be local search frame in it is non-central
The number of pixel,For with non-central pixel y each in search boxiCentered on element block, wherein i=1,2 ..., k,
Therefore, with non-central pieceCarry out linear expression central blockExpression formula it is as follows:
Wherein,It is the matrix of non-central element block,It is linear combination
Coefficient vector;
Step b: the expression formula of central block in step a is expressed as following equatioies:
WhereinFor the coefficient vector comprising a small amount of nonzero element, solved by orthogonal matching pursuit comprising a small amount of nonzero element
Coefficient vectorCentral blockThe linear combination obtained after rarefaction representation is as follows:
Step c: saliency is obtained by the residual error of reconstructed image and original image, simply indicates image with following formula
Conspicuousness:
Wherein | | | |2Indicate L2Norm;
Step d: carrying out local rarefaction representation to whole image and that is, from top to bottom by search box from left to right search for whole image,
And rarefaction representation is carried out to each element in local search frame, the residual error of equation in step c is finally asked entirely to be schemed
The notable figure of picture.
5. a kind of infrared image conspicuousness detection method according to claim 4, it is characterised in that: improved in step 4
The solution procedure of local contrast measurement, concrete operations are as follows:
Step 4.1: the processing of dimensional Gaussian differential filtering being carried out to target, the very noisy in wiping out background, so that background high brightness
Area grayscale value becomes larger;
Step 4.2: set a size as the local search frame of W × W to infrared image from top to bottom, from left to right searched
Whole image is carried out piecemeal, obtains a series of sub-block by rope, the moving step length of search box is set when carrying out piecemeal, so
These blocks are performed corresponding processing afterwards, each of given piece of value is the gray value of all pixels in block, it is defined as follows:
Wherein, block (s, t) is a sub-block, s=1,2...M, t=1,2...N, M, N respectively indicate whole image it is vertical and
The number of horizontal direction block, u are its values, and v × w is the size of block, and pix (i, j) is to belong to the inner pixel of block block (s, t),
(i, j) indicates the coordinate of pixel, and f (pix (i, j)) is the corresponding pixel value of pix (i, j);
Step 4.3: defining the value of block according to step 4.2, whole image is operated with block then, whole image is formed
The block matrix new as one, this matrix is defined as U, and is defined as follows by the new matrix that block forms:
U (s, t)=u (block (s, t)), s=1,2...M, t=1,2...N
Wherein, M, N respectively indicate the number of whole image both vertically as well as horizontally block, the i.e. line number and columns of this new matrix
Size;
Step 4.4: in the new matrix of sub-block composition, the average gray for defining a sub-block block (s, t) is u0, InFor
The maximum value of block (s, t) gray value, then forms following relational expression:
In=max (f (pix (i, j))), pix (i, j) ∈ block (s, t)
Again it is the image block of sub-block block (s, t) three times with a size, block centered on block (s, t), and finds
8 adjacent blocks of block (s, t) and their corresponding average value u in U matrix in this image block1~u8, as a result,
Improved local contrast measurement is defined as foloows:
When the central block to be detected is object block, max (u at this timez) < In, ILCM > u at this time0, target will be reinforced at this time;
When the central block to be detected is background block, max (u at this timez)≥In, ILCM≤u at this time0, background parts are suppressed at this time.
6. a kind of infrared image conspicuousness detection method according to claim 5, it is characterised in that: in the step 4.1
Difference of Gaussian filtering be equivalent to one can remove it is every other except the frequency being retained in original image
The bandpass filter of frequency information, can be with the very noisy in wiping out background, so that background high-brightness region gray value becomes larger.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610873518.5A CN106530271B (en) | 2016-09-30 | 2016-09-30 | A kind of infrared image conspicuousness detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610873518.5A CN106530271B (en) | 2016-09-30 | 2016-09-30 | A kind of infrared image conspicuousness detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106530271A CN106530271A (en) | 2017-03-22 |
CN106530271B true CN106530271B (en) | 2019-03-12 |
Family
ID=58331430
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610873518.5A Active CN106530271B (en) | 2016-09-30 | 2016-09-30 | A kind of infrared image conspicuousness detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106530271B (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107563370B (en) * | 2017-07-07 | 2020-04-03 | 西北工业大学 | Visual attention mechanism-based marine infrared target detection method |
CN107451595A (en) * | 2017-08-04 | 2017-12-08 | 河海大学 | Infrared image salient region detection method based on hybrid algorithm |
CN107992874B (en) * | 2017-12-20 | 2020-01-07 | 武汉大学 | Image salient target region extraction method and system based on iterative sparse representation |
CN108364273B (en) * | 2018-01-30 | 2022-02-25 | 中南大学 | Method for multi-focus image fusion in spatial domain |
CN108647693B (en) * | 2018-04-20 | 2021-07-02 | 华中科技大学 | Sea surface infrared target detection method based on binary significance characteristics |
CN108805139B (en) * | 2018-05-07 | 2022-02-18 | 南京理工大学 | Image similarity calculation method based on frequency domain visual saliency analysis |
CN109934224B (en) * | 2019-03-07 | 2023-06-09 | 绵阳慧视光电技术有限责任公司 | Small target detection method based on Markov random field and visual contrast mechanism |
CN111768376A (en) * | 2020-06-24 | 2020-10-13 | 山东科技大学 | Eddy current thermal imaging edge detection method and system, storage medium and application |
CN112163606B (en) * | 2020-09-17 | 2022-09-27 | 哈尔滨工业大学(威海) | Infrared small target detection method based on block contrast weighting |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104392463A (en) * | 2014-12-16 | 2015-03-04 | 西安电子科技大学 | Image salient region detection method based on joint sparse multi-scale fusion |
WO2015171084A1 (en) * | 2014-05-08 | 2015-11-12 | Aselsan Elektronik Sanayi Ve Ticaret Anonim Şirketi | A real- time, semi-automatic method for target tracking window initialization in thermal imagery |
-
2016
- 2016-09-30 CN CN201610873518.5A patent/CN106530271B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2015171084A1 (en) * | 2014-05-08 | 2015-11-12 | Aselsan Elektronik Sanayi Ve Ticaret Anonim Şirketi | A real- time, semi-automatic method for target tracking window initialization in thermal imagery |
CN104392463A (en) * | 2014-12-16 | 2015-03-04 | 西安电子科技大学 | Image salient region detection method based on joint sparse multi-scale fusion |
Non-Patent Citations (1)
Title |
---|
基于局部对比和全局稀有度的显著性检测;贺良杰 等;《计算机应用研究》;20140930;第31卷(第9期);2832-3840 |
Also Published As
Publication number | Publication date |
---|---|
CN106530271A (en) | 2017-03-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106530271B (en) | A kind of infrared image conspicuousness detection method | |
CN105447512B (en) | A kind of detection method and device for the beauty defect that essence slightly combines | |
CN109409190A (en) | Pedestrian detection method based on histogram of gradients and Canny edge detector | |
CN102819740B (en) | A kind of Single Infrared Image Frame Dim targets detection and localization method | |
CN105913415A (en) | Image sub-pixel edge extraction method having extensive adaptability | |
CN110488368B (en) | Contraband identification method and device based on dual-energy X-ray security inspection machine | |
Wang et al. | Fire smoke detection based on texture features and optical flow vector of contour | |
CN110189290A (en) | Metal surface fine defects detection method and device based on deep learning | |
CN104424632B (en) | A kind of video contrast's method for detecting abnormality and device | |
CN110930390A (en) | Chip pin missing detection method based on semi-supervised deep learning | |
CN110232387A (en) | A kind of heterologous image matching method based on KAZE-HOG algorithm | |
CN104933738A (en) | Visual saliency map generation method based on local structure detection and contrast | |
CN110189375A (en) | A kind of images steganalysis method based on monocular vision measurement | |
CN105427324A (en) | Magneto-optic imaging defect detection method based on automatic search of binary threshold | |
CN105869148A (en) | Target detection method and device | |
CN110706208A (en) | Infrared dim target detection method based on tensor mean square minimum error | |
CN103065320A (en) | Synthetic aperture radar (SAR) image change detection method based on constant false alarm threshold value | |
CN110033434A (en) | A kind of detection method of surface flaw based on texture conspicuousness | |
Lin et al. | Surface defect detection of machined parts based on machining texture direction | |
Lin et al. | A new prediction method for edge detection based on human visual feature | |
CN110321869A (en) | Personnel's detection and extracting method based on Multiscale Fusion network | |
Kumar et al. | Comparative analysis for edge detection techniques | |
CN107369163B (en) | Rapid SAR image target detection method based on optimal entropy dual-threshold segmentation | |
CN109978855A (en) | A kind of method for detecting change of remote sensing image and device | |
CN105359165B (en) | Determine the barcode position in document |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | 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 |