CN111311657A - Infrared image homologous registration method based on improved corner main direction distribution - Google Patents
Infrared image homologous registration method based on improved corner main direction distribution Download PDFInfo
- Publication number
- CN111311657A CN111311657A CN202010169423.1A CN202010169423A CN111311657A CN 111311657 A CN111311657 A CN 111311657A CN 202010169423 A CN202010169423 A CN 202010169423A CN 111311657 A CN111311657 A CN 111311657A
- Authority
- CN
- China
- Prior art keywords
- image
- infrared image
- gradient
- matching
- main direction
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- 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
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- 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
- G06T2207/20112—Image segmentation details
- G06T2207/20164—Salient point detection; Corner detection
Abstract
The invention belongs to the technical field of image processing, and discloses an infrared image homologous registration method based on improved corner main direction distribution, which comprises the steps of calculating a plurality of corners in two infrared images to be registered respectively, improving a corner main direction distribution method, distributing a main direction for each corner, establishing a matching relation between a descriptor and the main direction so as to realize the rotation invariance of the images, further obtaining a local intensity feature invariant descriptor (PIIFD descriptor) corresponding to each corner, and finally registering the infrared image to be registered and a reference infrared image by adopting nearest neighbor priority BBF and combining a bilateral matching algorithm on the basis of extracting the PIIFD descriptors. The whole algorithm is simple and efficient in calculation and high in accuracy, and the problems of high registration difficulty and low registration accuracy caused by low infrared image resolution and large texture information loss are effectively solved.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to an infrared image homologous registration method based on improved corner main direction distribution.
Background
The automatic registration of images is indispensable in the field of binocular stereo vision technology and image fusion, for example, in the medical industry, a plurality of scanning images of the cranium or the eyeball are registered and fused to obtain the clearest diagnostic image, and in geological measurement, remote sensing images shot by an unmanned aerial vehicle are registered and spliced. David Lowe from Canada teaches that a Scale Invariant Feature Transformation (SIFT) registration algorithm is proposed in 1999, supports image registration with obvious Scale Transformation, exploits a new direction of an image registration direction, and greatly improves an application scene of image registration. Many students followed the thought of David to obtain more remarkable results, for example, Speed Up Robust Features (SURF) algorithm proposed by herbertby is also widely cited by those skilled in the art.
Because the parameters of the infrared cameras are inconsistent and the cost is inconsistent, two infrared images cannot be completely matched in the practical application process. The invention performs registration on two infrared images to obtain a measurement result with a smaller error. The infrared image has unclear texture and relatively large noise, so that the infrared image and the infrared image have high registration difficulty and low precision. The registration algorithm based on the gray level or the regional characteristics has low fault tolerance rate for solving the problems, so that the characteristic-based matching algorithm is more reasonable to select. In addition, in consideration of the particularity of the image scene of the electrical equipment, the method and the device finish the homologous image registration work by combining the corner extraction, the improved corner main direction distribution method and the bilateral matching method.
Disclosure of Invention
The invention provides an infrared image homologous registration method based on improved corner main direction distribution, which solves the problem of low registration precision of the existing infrared homologous image.
The invention can be realized by the following technical scheme:
an infrared image homologous registration method based on improved corner main direction distribution comprises the following steps:
step one, respectively detecting a plurality of angular points in two infrared images to be registered through an angular point detection algorithm.
Let I be the matrix of image pixel values, GuAnd GvRepresenting the gradients on the u and v axes, the matrix expression of the transverse and longitudinal gray scale gradients of the image is
Assuming that the window sliding displacement vector is (x, y), the gaussian displacement length is wx × wy, the gaussian displacement window is w ═ wx, wy, and the pixel difference value in the two windows after the displacement is recorded as (wx, wy)
Wherein
A. B, C respectively represent the convolution of the gradients on the u and v axes with a gaussian displacement window, M being the matrix formed by the convolution, Tr being the trace of the matrix M, and Det being the determinant value of the matrix M.
Let the corner response function R be Det-k Tr2K is a constant, and is generally 0.04 to 0.06.
Judging whether the pixel point is an angular point according to the response value R of the pixel point, wherein the criterion is
And obtaining candidate angular points in the image according to the criterion, and finally selecting the angular point with the maximum local R value as a final result by utilizing a maximum value inhibition mode.
And step two, distributing a main direction for each corner point by improving a corner point main direction distribution method, and establishing a matching relation between a descriptor and the main direction so as to realize the rotation invariance of the image.
Since the gradients of the infrared image for the same pixel point may be the same or opposite, the angle of the gradient vector must be limited to [0,180 °). Establishing a new gradient vector for the image I, and modifying the formula (1) into
The modified vector expression processes the initial gradient by a sign function, so that the angular point gradients opposite to the infrared image gradients are converted to the same direction.
And performing vector accumulation on the gradients in the pixel window in the neighborhood of the corner point, wherein the accumulation can cause opposite gradients to be mutually offset, and finally the gradient vector of the corner point is a 0 vector. To solve this problem, a gradient squared vector is introduced, denoted as
Wherein G iss,uAnd Gs,vThe improved squared gradient on the u and v axes is shown by further noting the mean squared gradient as
Wherein thereinAndrepresenting the improved mean squared gradient, ω, on the u and v axesσRefers to a gaussian convolution kernel with full width at half maximum σ. Because the window of the Gaussian convolution kernel is small and is sensitive to noise, the value of sigma is five pixels。
And finally, determining the main direction of each corner point according to the formula (7).
Thus determining for each corner point its principal direction as (u, v).
In the traditional registration algorithm, the main direction of an angular point is determined by adopting a neighborhood gradient histogram method, the calculation process is relatively dependent on the dimension of the histogram, and the results are discrete values. The principal direction determined by the formula (7) is a continuous value, and the principal directions of the same corner points in the image are ensured to be the same.
Step three, taking the main direction as a reference direction and the angular points as the central neighborhood gradient vectors, and further obtaining a local intensity feature invariant descriptor (PIIFD descriptor) corresponding to each angular point;
in order to ensure the rotational invariance of the descriptor, in the process of calculating the gradient direction of the neighborhood of the characteristic angular point, the main direction of the angular point is taken as the reference direction, namely the 0-degree direction. Taking a corner point 4 × 4 neighborhood, dividing the neighborhood into 16 sub-regions, and calculating 8 gradient directions for each sub-region. A gradient direction is represented by a gradient square column, for example, 8 gradient directions are taken as examples, as shown in fig. 2, 8 gradient vectors are obtained after gradient statistics of a white region, and then all sub-regions constitute a 128-dimensional vector.
However, there are a lot of gradient inversions in the infrared image of the power device, and the gradient direction of the neighborhood needs to be modified in order to ensure the local intensity invariance. Firstly, the gradient amplitude is subjected to sectional weighting treatment, the weight of the amplitude of the first 20 percent is given to 1, the weight of 20 to 40 percent is given to 0.75, the weight of 40 to 60 percent is given to 0.5, and the weight of the part of 20 percent with the minimum amplitude is 0 in the same way. Furthermore, to limit the gradient direction to within [0,180 °), the gradient magnitude is a 64-dimensional vector whose magnitudes differ by pi.
After the two-step processing method, the gradient vector of the ith row and jth column sub-area is recorded as HijGradient vector of the entire neighborhood is noted
It should be noted that even if the gradient direction is limited to [0,180 ° when calculating the main direction of the corner point, the two main directions calculated by equation (7) are still opposite to each other for the case where the image to be registered and the reference image are rotated by 180 ° from each other, and in order to solve this problem, an auxiliary matrix Q ═ rot (H,180 °) is introduced, and finally the PIIFD descriptor is obtained as
Hi=[Hi1Hi2Hi3Hi4]
Qi=[Qi1Qi2Qi3Qi4](10)
Wherein c is a scale factor for adjusting the magnitude of the PIIFD descriptor amplitude.
It is clear that the Des descriptor preserves the dimension size of the gradient vector, i.e. 4 × 4 × 4. In order to facilitate the subsequent bilateral matching work, the bilateral matching method converts the bilateral matching method into a one-dimensional row vector according to the sequence of the row vectors, and normalizes DES ═ Des/| Des |.
By introducing PIIFD descriptors, the situation that the gradients of the two infrared images are opposite is solved, and the descriptor vectors of each corner point can be matched in the next step.
Fourthly, on the basis of extracting a plurality of PIIFD descriptors, registering the infrared image to be registered and the reference infrared image by adopting nearest neighbor priority BBF and combining a bilateral matching algorithm to obtain a plurality of registration point sets;
suppose an image I to be registered1Corresponding description subsets are combined to F1Reference image I2Corresponding description subsets are combined to F2For aggregate is F1The ith element f in1iDefinition of f1iTo set F2A distance of
For f1iThe BBF algorithm refers to the search set D (f)1i,F2) F corresponding to the maximum value of2iAs its matching point. In this context, to ensure the suitability of the algorithm, use is made of2i-maxAnd f2i-smaxRespectively represent a set D (f)1i,F2) Maximum and sub-maximum of (d); only two satisfy the relationship f2i-smax<f2i-maxWhen t is less than t, the maximum value f is selected2iCorresponding f2iAnd taking the value of 0.8-0.9 as the matching point, otherwise, judging that the matching is failed, and taking the value as a proper value according to the empirical t value. The two descriptors (or the corners corresponding to the descriptors) successfully matched are recorded as a single-edge matching set M (I)1,I2). Obviously, a single-sided match can occur in many-to-one situations, i.e., F1May be matched to F2The same element in (1). To solve this problem, the following formula I2To register an image, I2For the reference image, a matching set M (I) is obtained2,I1). Retention of M (I)1,I2) And M (I)2,I1) The same element in (2) is used as the point of successful initial matching.
The above-mentioned bidirectional matching process for completing feature point descriptor matching is bilateral matching. The uniqueness of the matching points is ensured through bilateral matching, and the influence of many-to-one is removed.
The beneficial technical effects of the invention are as follows:
due to the difference of the sensors, the resolution of the infrared image is low, the texture information is more lost, most of the angular points extracted by the Harris angular point extraction algorithm are distributed on the outline of the infrared image, the registration error caused by the difference of the image is reduced, the main direction distribution of the angular points is improved, the dimension of a histogram is not depended, and the calculated amount is less than that of the traditional method. And finally, an auxiliary matrix is introduced to improve the PIIFD descriptor, so that the condition that the gradients of the two infrared images are opposite is solved, the registration precision is improved, and a better matching effect is ensured.
Drawings
FIG. 1 is a schematic overall flow diagram of the present invention;
FIG. 2 is a diagram of an example of gradient histogram statistics;
fig. 3 is a diagram illustrating the result of the present invention after registration.
Detailed Description
The following detailed description of the preferred embodiments will be made with reference to the accompanying drawings.
Because the parameters of the infrared cameras are inconsistent and the cost is inconsistent, two infrared images cannot be completely matched in the practical application process. The invention performs registration on two infrared images to obtain a measurement result with a smaller error. The infrared image has unclear texture and relatively large noise, so that the infrared image and the infrared image have high registration difficulty and low precision. The registration algorithm based on the gray level or the regional characteristics has low fault tolerance rate for solving the problems, so that the characteristic-based matching algorithm is more reasonable to select. In addition, in consideration of the particularity of the image scene of the electrical equipment, the method and the device finish the homologous image registration work by combining the corner extraction, the improved corner main direction distribution method and the bilateral matching method.
As shown in FIG. 1, the invention provides an infrared image homologous registration method based on improved corner main direction distribution, which comprises the steps of calculating a plurality of corners in two infrared images to be registered respectively, improving a corner main direction distribution method, distributing a main direction for each corner, establishing a matching relation between a descriptor and the main direction so as to realize the rotation invariance of the images, further obtaining a local intensity feature invariant descriptor (PIIFD descriptor) corresponding to each corner, and finally registering the infrared image to be registered and a reference infrared image by adopting nearest neighbor priority BBF and combining a bilateral matching algorithm on the basis of extracting the PIIFD descriptors. The specific implementation mode comprises the following steps:
step one, image preprocessing, and determining whether a filtering operation is needed according to the image quality. Firstly, for the redistribution of the gray value of the image, firstly, the gray value image is subjected to linear transformation, and then the gray value is stretched to 0-255, and the specific formula is as follows:
and step two, respectively calculating a plurality of corner points in the two infrared images to be registered, and taking a diagonal line where the corner point is positioned as a main direction of the two infrared images.
The corner point is a point with a sharp change of brightness in the image or a curvature maximum value on an edge curve of the image. Due to the fact that the detail textures of the infrared images are seriously lost, the most remarkable corner points are distributed on the outline of the object, and the corner point detection algorithm is used for respectively detecting the corner points in the two infrared images to be registered.
Let I be the matrix of image pixel values, GuAnd GvRepresenting the gradients on the u and v axes, the matrix expression of the transverse and longitudinal gray scale gradients of the image is
Assuming that the window sliding displacement vector is (x, y), the gaussian displacement length is wx × wy, the gaussian displacement window is w ═ wx, wy, and the pixel difference value in the two windows after the displacement is recorded as (wx, wy)
Wherein
A. B, C respectively represent the convolution of the gradients on the u and v axes with a gaussian displacement window, M being the matrix formed by the convolution, Tr being the trace of the matrix M, and Det being the determinant value of the matrix M.
Let the corner response function R be Det-k Tr2K is a constant, and is generally 0.04 to 0.06.
Judging whether the pixel point is an angular point according to the response value R of the pixel point, wherein the criterion is
And obtaining candidate angular points in the image according to the criterion, and finally selecting the angular point with the maximum local R value as a final result by utilizing a maximum value inhibition mode.
And step three, distributing a main direction for each corner point by improving a corner point main direction distribution method, and establishing a matching relation between the descriptor and the main direction so as to realize the rotation invariance of the image.
Since the gradients of the infrared image for the same pixel point may be the same or opposite, the angle of the gradient vector must be limited to [0,180 °). Establishing a new gradient vector for the image I, and modifying the formula (1) into
The modified vector expression processes the initial gradient by a sign function, so that the angular point gradients opposite to the infrared image gradients are converted to the same direction.
And performing vector accumulation on the gradients in the pixel window in the neighborhood of the corner point, wherein the accumulation can cause opposite gradients to be mutually offset, and finally the gradient vector of the corner point is a 0 vector. To solve this problem, a gradient squared vector is introduced, denoted as
Wherein G iss,uAnd Gs,vThe improved squared gradient on the u and v axes is shown by further noting the mean squared gradient as
Wherein thereinAndrepresenting the improved mean squared gradient, ω, on the u and v axesσRefers to a gaussian convolution kernel with full width at half maximum σ. Because the window of the gaussian convolution kernel is small and sensitive to noise, the value of σ is only five pixels.
And finally, determining the main direction of each corner point according to the formula (7).
Thus determining for each corner point its principal direction as (u, v).
In the traditional registration algorithm, the main direction of an angular point is determined by adopting a neighborhood gradient histogram method, the calculation process is relatively dependent on the dimension of the histogram, and the results are discrete values. The principal direction determined by the formula (7) is a continuous value, and the principal directions of the same corner points in the image are ensured to be the same.
Step four, taking the main direction as a reference direction and the angular points as the central neighborhood gradient vectors, and further obtaining a local intensity feature invariant descriptor (PIIFD descriptor) corresponding to each angular point;
in order to ensure the rotational invariance of the descriptor, in the process of calculating the gradient direction of the neighborhood of the characteristic angular point, the main direction of the angular point is taken as the reference direction, namely the 0-degree direction. Taking a corner point 4 × 4 neighborhood, dividing the neighborhood into 16 sub-regions, and calculating 8 gradient directions for each sub-region. A gradient direction is represented by a gradient square column, for example, 8 gradient directions are taken as examples, as shown in fig. 2, 8 gradient vectors are obtained after gradient statistics of a white region, and then all sub-regions constitute a 128-dimensional vector.
However, there are a lot of gradient inversions in the infrared image of the power device, and the gradient direction of the neighborhood needs to be modified in order to ensure the local intensity invariance. Firstly, the gradient amplitude is subjected to sectional weighting treatment, the weight of the amplitude of the first 20 percent is given to 1, the weight of 20 to 40 percent is given to 0.75, the weight of 40 to 60 percent is given to 0.5, and the weight of the part of 20 percent with the minimum amplitude is 0 in the same way. Furthermore, to limit the gradient direction to within [0,180 °), the gradient magnitude is a 64-dimensional vector whose magnitudes differ by pi.
After the two-step processing method, the gradient vector of the ith row and jth column sub-area is recorded as HijGradient vector of the entire neighborhood is noted
It should be noted that even if the gradient direction is limited to [0,180 ° when calculating the main direction of the corner point, the two main directions calculated by equation (7) are still opposite to each other for the case where the image to be registered and the reference image are rotated by 180 ° from each other, and in order to solve this problem, an auxiliary matrix Q ═ rot (H,180 °) is introduced, and finally the PIIFD descriptor is obtained as
Hi=[Hi1Hi2Hi3Hi4]
Qi=[Qi1Qi2Qi3Qi4](10)
Wherein c is a scale factor for adjusting the magnitude of the PIIFD descriptor amplitude.
It is clear that the Des descriptor preserves the dimension size of the gradient vector, i.e. 4 × 4 × 4. In order to facilitate the subsequent bilateral matching work, the bilateral matching method converts the bilateral matching method into a one-dimensional row vector according to the sequence of the row vectors, and normalizes DES ═ Des/| Des |.
By introducing PIIFD descriptors, the situation that the gradients of the two infrared images are opposite is solved, and the descriptor vectors of each corner point can be matched in the next step.
Fifthly, registering the infrared image to be registered and the reference infrared image by adopting nearest neighbor first BBF (base band function) combined with a bilateral matching algorithm on the basis of extracting a plurality of PIIFD descriptors to obtain a plurality of registration point sets;
suppose an image I to be registered1Corresponding description subsets are combined to F1Reference image I2Corresponding description subsets are combined to F2For aggregate is F1The ith element f in1iDefinition of f1iTo set F2A distance of
For f1iThe BBF algorithm refers to the search set D (f)1i,F2) F corresponding to the maximum value of2iAs its matching point. In this context, to ensure the suitability of the algorithm, use is made of2i-maxAnd f2i-smaxRespectively represent a set D (f)1i,F2) Maximum and sub-maximum of (d); only two satisfy the relationship f2i-smax<f2i-maxWhen t is less than t, the maximum value f is selected2iCorresponding f2iAnd taking the value of 0.8-0.9 as the matching point, otherwise, judging that the matching is failed, and taking the value as a proper value according to the empirical t value. The two descriptors (or the corners corresponding to the descriptors) successfully matched are recorded as a single-edge matching set M (I)1,I2). Obviously, a single-sided match can occur in many-to-one situations, i.e., F1May be matched to F2The same element in (1). To solve this problem, the following formula I2To register an image, I2For the reference image, a matching set M (I) is obtained2,I1). Retention of M (I)1,I2) And M (I)2,I1) The same element in (2) is used as the point of successful initial matching.
In order to verify the feasibility of the method, different algorithms are used for carrying out registration experiments on the self-built infrared image database, and the experimental contents comprise:
an experiment platform:
MATLAB 2017b,
evaluation indexes are as follows: root Mean Square Error (RMSE)
Wherein (x)i,yi) Registration point coordinates obtained by image registration (x'i,y’i) The coordinate of the theoretical registration point after the registration point passes through the theoretical perspective transformation matrix. The index can objectively reflect that the matching precision of the registration algorithm is high, and the smaller the value of the index is, the higher the registration precision is.
The method adopts three different traditional algorithms, namely SIFT, SURF and Harris-SIFT, to compare with the method, and compares an image in the infrared database of the self-built electrical equipment, as shown in figure 3. The indices are shown in the following table.
Compared with the results of the three traditional methods, the infrared image registration method has the minimum root mean square error. Most notably, the method of the present invention obtains more than twice the logarithm of the correct registration points in the registration process than other methods, and the distribution of the registration points on the image is most comprehensive. The experimental results show that the method has higher practicability and more accurate results compared with the traditional method.
TABLE 1 results obtained with the present invention and the results of the conventional three types of methods
Method of producing a composite material | RMSE | Ratio of correct registration points |
SIFT | 27.959 | 0.833 |
SURF | 25.091 | 0.500 |
Harris-SIFT | 8.154 | 0.643 |
The method of the invention | 7.437 | 0.922 |
Although specific embodiments of the present invention have been described above, it will be appreciated by those skilled in the art that these are merely examples and that many variations or modifications may be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is therefore defined by the appended claims.
Claims (5)
1. An infrared image homologous registration method based on improved corner main direction distribution is characterized by comprising the following steps:
step one, respectively detecting a plurality of angular points in two infrared images to be registered through an angular point detection algorithm.
And step two, distributing a main direction for each corner point by improving a corner point main direction distribution method, and establishing a matching relation between a descriptor and the main direction so as to realize the rotation invariance of the image.
Step three, taking the main direction as a reference direction and the corner points as a central neighborhood gradient vector, and further obtaining a local Intensity feature Invariant descriptor, namely a PIIFD (partial Intensity Invariant feature descriptor), corresponding to each corner point;
and fourthly, registering the infrared image to be registered and the reference infrared image by adopting nearest neighbor first BBF (base band function) combined with a bilateral matching algorithm on the basis of extracting a plurality of PIIFD descriptors to obtain a plurality of registration point sets.
2. The method for improving infrared image homologous registration based on corner point principal direction distribution as claimed in claim 1, wherein the method comprises the following steps:
aiming at the characteristics of low resolution and large texture information loss of an infrared image, the transverse and longitudinal gray gradient matrixes of the image are improved, and a new gradient vector is established for an image pixel matrix I as follows:
the improved vector expression processes the gradient in the y direction by a sign function, so that the gradient of a corner point opposite to the gradient of the reference infrared image to be registered is converted into the same direction.
3. The method for improving infrared image homologous registration based on corner point principal direction distribution as claimed in claim 1, wherein the method comprises the following steps:
if the gradient in the pixel window in the neighborhood of the corner point is directly subjected to vector accumulation, the accumulation can cause opposite gradients to be mutually offset, and finally the gradient vector of the corner point is a 0 vector. The present invention introduces a gradient square vector to solve this problem, as follows:
further note that the mean squared gradient is
Wherein wσRefers to a gaussian convolution kernel with full width at half maximum σ. Because the window of the gaussian convolution kernel is small and sensitive to noise, the value of σ is only five pixels.
Finally, the main direction of each angular point is determined, and the main direction of each angular point (u, v) is determined to be phi (u, v).
4. The method for infrared image homologous registration based on improved corner point principal direction distribution as claimed in claim 1, wherein the method comprises the following step three:
aiming at the condition that the image to be registered and the reference image are mutually rotated by 180 degrees, the two main directions obtained by calculation in the claim 3 are probably opposite, in order to solve the problem, the calculation method of the PIIFD descriptor is improved, an auxiliary matrix Q-rot (H,180 degrees) is introduced, and the PIIFD descriptor is finally obtained
Hi=[Hi1Hi2Hi3Hi4]
Qi=[Qi1Qi2Qi3Qi4](10)
Wherein c is a scale factor for adjusting the magnitude of the PIIFD descriptor amplitude.
It is clear that the Des descriptor preserves the dimension size of the gradient vector, i.e. 4 × 4 × 4. In order to facilitate the subsequent bilateral matching work, the bilateral matching method converts the bilateral matching method into a one-dimensional row vector according to the sequence of the row vectors, and normalizes DES ═ Des/| Des |.
5. The method for infrared image homologous registration based on improved corner point principal direction distribution as claimed in claim 1, wherein the method comprises the following four steps:
on the basis of extracting a plurality of PIIFD descriptors, the infrared image to be registered and the reference infrared image are registered by adopting nearest neighbor and preferentially combining a bilateral matching algorithm, so that the accuracy of a matching result is enhanced. Image to be registered I1Corresponding description subsets are combined to F1Reference image I2Corresponding description subsets are combined to F2For f1i∈F1Then f is1iTo set F2A distance of
For f1iThe BBF algorithm refers to the search set D (f)1i,F2) F corresponding to the maximum value of2iAs its matching point. In order to ensure the applicability of the algorithm in the present invention, use f2iAnd f2iRespectively represent a set D (f)1i,F2) Maximum and sub-maximum of (d); only two satisfy the relationship f2i<f2iWhen the value is less than i, the maximum value f is selected2iCorresponding f2iAnd taking the value of 0.8-0.9 as the matching point, otherwise, judging that the matching is failed, and taking the value as a proper value according to the empirical t value.
Recording the two descriptors (or the corner points corresponding to the descriptors) successfully matched as a single-edge matching set M (I)1,I2). Obviously, a single-sided match can occur in many-to-one situations, i.e., F1May be matched to F2The same element in (1). To solve this problem, the following formula I2To register an image, I1For the reference image, a matching set M (I) is obtained2,I1). Retention of M (I)1,I2) And M (I)2,I1) The same elements in the sequence are used as points for successful initial matching, so that the uniqueness of the matching points is ensured, and the influence of many-to-one is removed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010169423.1A CN111311657B (en) | 2020-03-12 | 2020-03-12 | Infrared image homologous registration method based on improved corner principal direction distribution |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010169423.1A CN111311657B (en) | 2020-03-12 | 2020-03-12 | Infrared image homologous registration method based on improved corner principal direction distribution |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111311657A true CN111311657A (en) | 2020-06-19 |
CN111311657B CN111311657B (en) | 2023-04-25 |
Family
ID=71145514
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010169423.1A Active CN111311657B (en) | 2020-03-12 | 2020-03-12 | Infrared image homologous registration method based on improved corner principal direction distribution |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111311657B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113095385A (en) * | 2021-03-31 | 2021-07-09 | 安徽工业大学 | Multimode image matching method based on global and local feature description |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2012058902A1 (en) * | 2010-11-02 | 2012-05-10 | 中兴通讯股份有限公司 | Method and apparatus for combining panoramic image |
WO2017049994A1 (en) * | 2015-09-25 | 2017-03-30 | 深圳大学 | Hyperspectral image corner detection method and system |
WO2018076137A1 (en) * | 2016-10-24 | 2018-05-03 | 深圳大学 | Method and device for obtaining hyper-spectral image feature descriptor |
CN109285110A (en) * | 2018-09-13 | 2019-01-29 | 武汉大学 | The infrared visible light image registration method and system with transformation are matched based on robust |
CN110223330A (en) * | 2019-06-12 | 2019-09-10 | 国网河北省电力有限公司沧州供电分公司 | A kind of method for registering and system of visible light and infrared image |
-
2020
- 2020-03-12 CN CN202010169423.1A patent/CN111311657B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2012058902A1 (en) * | 2010-11-02 | 2012-05-10 | 中兴通讯股份有限公司 | Method and apparatus for combining panoramic image |
WO2017049994A1 (en) * | 2015-09-25 | 2017-03-30 | 深圳大学 | Hyperspectral image corner detection method and system |
WO2018076137A1 (en) * | 2016-10-24 | 2018-05-03 | 深圳大学 | Method and device for obtaining hyper-spectral image feature descriptor |
CN109285110A (en) * | 2018-09-13 | 2019-01-29 | 武汉大学 | The infrared visible light image registration method and system with transformation are matched based on robust |
CN110223330A (en) * | 2019-06-12 | 2019-09-10 | 国网河北省电力有限公司沧州供电分公司 | A kind of method for registering and system of visible light and infrared image |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113095385A (en) * | 2021-03-31 | 2021-07-09 | 安徽工业大学 | Multimode image matching method based on global and local feature description |
CN113095385B (en) * | 2021-03-31 | 2023-04-18 | 安徽工业大学 | Multimode image matching method based on global and local feature description |
Also Published As
Publication number | Publication date |
---|---|
CN111311657B (en) | 2023-04-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111223133B (en) | Registration method of heterogeneous images | |
CN110097093B (en) | Method for accurately matching heterogeneous images | |
Ye et al. | A local descriptor based registration method for multispectral remote sensing images with non-linear intensity differences | |
CN111080529A (en) | Unmanned aerial vehicle aerial image splicing method for enhancing robustness | |
CN110223330B (en) | Registration method and system for visible light and infrared images | |
CN112085772B (en) | Remote sensing image registration method and device | |
Bouchiha et al. | Automatic remote-sensing image registration using SURF | |
CN111340701B (en) | Circuit board image splicing method for screening matching points based on clustering method | |
CN110490913B (en) | Image matching method based on feature description operator of corner and single line segment grouping | |
CN110569861B (en) | Image matching positioning method based on point feature and contour feature fusion | |
CN107066969A (en) | A kind of face identification method | |
CN107862319B (en) | Heterogeneous high-light optical image matching error eliminating method based on neighborhood voting | |
CN107240130B (en) | Remote sensing image registration method, device and system | |
CN103136525A (en) | Hetero-type expanded goal high-accuracy positioning method with generalized Hough transposition | |
CN109472770B (en) | Method for quickly matching image characteristic points in printed circuit board detection | |
CN107610166A (en) | A kind of plane map method for registering images based on linear character region segmentation | |
CN105631872A (en) | Remote sensing image registration method based on multiple feature points | |
CN111462198A (en) | Multi-mode image registration method with scale, rotation and radiation invariance | |
CN107016414A (en) | A kind of recognition methods of footprint | |
CN111311657A (en) | Infrared image homologous registration method based on improved corner main direction distribution | |
CN113095385A (en) | Multimode image matching method based on global and local feature description | |
CN115035281B (en) | Rapid infrared panoramic image stitching method | |
CN115775269A (en) | Train image accurate registration method based on line features | |
EP2884427B1 (en) | Method and system for describing an image | |
CN113792788B (en) | Infrared and visible light image matching method based on multi-feature similarity fusion |
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 |