CN106980862B - Circular image recognition feature extraction method and system - Google Patents

Circular image recognition feature extraction method and system Download PDF

Info

Publication number
CN106980862B
CN106980862B CN201710164058.3A CN201710164058A CN106980862B CN 106980862 B CN106980862 B CN 106980862B CN 201710164058 A CN201710164058 A CN 201710164058A CN 106980862 B CN106980862 B CN 106980862B
Authority
CN
China
Prior art keywords
circular image
points
support vector
vector machine
code value
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201710164058.3A
Other languages
Chinese (zh)
Other versions
CN106980862A (en
Inventor
张东波
陈红磊
文登伟
寇涛
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.)
Xiangtan University
Original Assignee
Xiangtan University
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 Xiangtan University filed Critical Xiangtan University
Priority to CN201710164058.3A priority Critical patent/CN106980862B/en
Publication of CN106980862A publication Critical patent/CN106980862A/en
Application granted granted Critical
Publication of CN106980862B publication Critical patent/CN106980862B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/88—Image or video recognition using optical means, e.g. reference filters, holographic masks, frequency domain filters or spatial domain filters

Abstract

The invention provides a circular image recognition feature extraction method. The method comprises the following steps: after the circular image is obtained, the circular image is subjected to normalization processing and is divided into an annular space containing a plurality of annular areas by using a mask template; obtaining a pattern pair of points symmetrical about a central point of each annular region based on a method for calculating a code value; obtaining a histogram of a pattern pair of points symmetrical with respect to the central point for each of the ring-shaped regions based on the pattern pair; the histograms are combined to obtain the identified features. According to the invention, the problem of angle rotation of the circular image can be solved by extracting the histograms of the mode pairs of the points symmetrical about the central point, so that the identification characteristics of the circular image are extracted, and the circular pattern is identified with high accuracy.

Description

Circular image recognition feature extraction method and system
Technical Field
The invention relates to the technical field of image recognition, in particular to a method and a system for extracting circular image recognition features.
Background
In real life, a large number of circular image target recognition problems exist, such as car logos, icons, picture logos, trademarks, labels, coins, circular patterns and patterns on decorative articles and the like. In many occasions, the printing and placing angles of the circular images are not fixed, and the patterns are frequently subjected to angular rotation. If the acquired original image is directly identified, a large number of samples of prototype patterns of various angles need to be stored in the system, and the sample collection difficulty is high and is difficult to realize. Therefore, in order to solve the problem of identifying a circular image object in a real scene, it is first necessary to process an acquired original image and extract an identification feature. Before extracting the identification features, the problem of angular rotation of the image is solved. In the prior art, a method for extracting and identifying features after angle rotation of a circular image is not well overcome.
Disclosure of Invention
In view of the above, the present invention provides a method and a system for extracting circular image recognition features, which can overcome the problem of angular rotation of a circular image and achieve the purpose of extracting the recognition features of the circular image.
In order to achieve the purpose, the invention provides the following technical scheme:
a circular image recognition feature extraction method comprises the following steps:
carrying out size normalization on a first circular image to obtain a second circular image, wherein the first circular image is an obtained original circular image;
dividing the second circular image into an annular space consisting of a plurality of annular areas by using a mask template;
extracting a pattern pair of points of the annular region that are symmetric about a center point;
obtaining a histogram of the pattern pair of the points of the ring-shaped region symmetric about the central point based on the pattern pair;
and assembling the histograms to obtain identification features.
Preferably, the method further comprises the following steps:
and sending the identification features to a support vector machine classifier, and identifying the identification features by using the support vector machine classifier to obtain an identification result.
Preferably, the support vector machine classifier uses a linear kernel function as its kernel function.
Preferably, the identifying result obtained by identifying the identifying features by using the support vector machine classifier is specifically:
and the support vector machine classifier identifies the identification features of the three channels to obtain the identification result.
A circular image recognition feature extraction system, comprising:
the normalization module is used for carrying out size normalization on the first circular image to obtain a second circular image, wherein the first circular image is an acquired original circular image;
the partitioning module is used for dividing the second circular image into an annular space formed by a plurality of annular areas by using a mask template;
a calculation module for extracting a pattern pair of points of the annular region that are symmetric about a center point;
a statistical module to derive a histogram of the pattern pair of the points of the ring-shaped region symmetric about a center point based on the pattern pair;
and the assembling module is used for assembling the histogram to obtain the identification features.
Preferably, the method further comprises the following steps:
and sending the identification features to a support vector machine classifier, and identifying the identification features by using the support vector machine classifier to obtain an identification result.
Preferably, the support vector machine classifier uses a linear kernel function as its kernel function.
Preferably, the identification result obtained by identifying the identification features by using the support vector machine classifier is specifically as follows:
and the support vector machine classifier identifies the characteristics of the three channels to obtain an identification result.
According to the technical scheme, the invention provides the circular image identification feature extraction method, after the circular image is obtained, the circular image is subjected to normalization processing, and is divided into an annular space comprising a plurality of annular areas by using the mask template; obtaining a pattern pair of points symmetrical about a central point of each annular region based on a method for calculating a code value; obtaining a histogram of a pattern pair of points symmetrical with respect to the central point for each of the ring-shaped regions based on the pattern pair; the histograms are combined to obtain identification features, the problem of angle rotation of the circular image can be solved by extracting the histograms of the mode pairs of the points which are symmetrical about the central point, the identification features of the circular image are extracted, and then the circular pattern is identified with high accuracy.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart of an embodiment 1 of a circular image recognition feature extraction method disclosed in the present invention;
FIG. 2 is a schematic diagram of calculating an encoded value;
FIG. 3 is a flowchart of an embodiment 3 of a circular image recognition feature extraction method disclosed in the present invention;
FIG. 4 is a diagram of sector division when approximating a code value;
FIG. 5 is a schematic illustration of the location of point A, B, C;
FIG. 6 is a flow chart of calculating the encoded value of point A, B, C;
fig. 7 is a schematic structural diagram of an embodiment 7 of the circular image recognition feature extraction system disclosed in the present invention;
FIG. 8 is a schematic structural diagram of an embodiment 8 of a circular image recognition feature extraction system disclosed in the present invention;
FIG. 9 is a schematic diagram of an annular space and points of symmetry about a center point.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, a flowchart of an embodiment 1 of the circular image recognition feature extraction method provided by the present invention is shown:
s101, carrying out size normalization on the first circular image to obtain a second circular image, wherein the first circular image is an obtained original circular image;
normalizing the first circular image size to a second circular image of M, i.e. IROIAnd the value of M is in direct proportion to the size of the first circular image. For example, the first circular image is divided into a 200 × 200 area, where 200 is 200 unit lengths.
S102, dividing the second circular image into annular spaces formed by a plurality of annular areas by using a mask template;
generating a plurality of ring mask templates with increasing radius and partially overlapped in the same circle center, namely generating n rings with equal area in an overlapped mode by taking a point (M/2 ) as a circle center O in a 200X 200 area, wherein the mask template is psimFor example, taking n-26, i.e. the radius of each ring is,
Figure GDA0002451173720000041
using a mask template to align the second circular image IROIMasking to obtain several annular regions Im，Im＝IROI*ψm
S103, extracting a mode pair of points which are symmetrical about a central point of the annular region;
first, define the code value, as shown in FIG. 2, let O be the ring region ImP is ImAnd (3) extracting the coding value of the pixel point P from any one pixel point: and establishing a local coordinate system, wherein the direction along OP is called as a radial coordinate axis r, the direction vertical to OP is called as a tangential coordinate axis t, and P is a new local coordinate system origin. Respectively finding 4 symmetrical adjacent points in the radial direction r and the tangential direction t, and respectively recording the adjacent points as P in a counterclockwise direction1、P2、P3、P4In which P is1On the coordinate axis r, and P2Is located on the opposite side of the coordinate axis r relative to O, and the distances between the 4 neighborhood points and the P point are recorded as d, where d may be 1, where 1 is a unit length same as that adopted in the above 200Unit length of (1) according to P1、P2、P3、P4The pixel value size is compared with the P point according to the following formula:
Figure GDA0002451173720000042
obtaining binary number code T ═ T (T)1T2T3T4) I (P) is the pixel value of point P, I (P)i) Is the taken point PiThe pixel value of (a) is obtained as a binary number code T ═ T (T)1T2T3T4)。
And then by the formula: F-8T1+4*T2+2*T3+T4And (4) converting the binary number code into a decimal number code, wherein F is the code value of the point P.
In order to increase the calculation speed and avoid interpolation calculation when calculating the code value of the P point, an approximation algorithm may be used when calculating the code value.
As shown in fig. 4, i.e., within the second circular image, the circular area is divided into 8 sectors in units of 45 degrees in the direction pointing from the center of the image to point P.
As shown in fig. 5 and 6, the position diagrams of the three arbitrarily selected points A, B and C and the flow chart of calculating the code value are shown. Wherein A, B, C forms an angle theta with the x-axis in the horizontal directionA＜22.5°,22.5°＜θB＜67.5°,67.5°＜θC＜112.5°。
Fig. 9 is a schematic diagram of the annular space and the point symmetrical with respect to the center point. Taking points (i, j) and (i ', j') in the annular region, wherein the two points are symmetrical about the central point O, and the coordinate relationship between the two points is as follows:
Figure GDA0002451173720000051
calculating the code value in the neighborhood (including (i, j)) of k multiplied by k (k ranges from 3 to 11) of (i, j), counting the code value with the most occurrence times, taking the code value as the representative mode of the point (i, j), and marking the code value as S1. The same can be obtained(i ', j') represents the pattern S2If S is1>S2Then, the pattern pair of points symmetric about the center point is represented as (S)1，S2) Otherwise, it is represented as (S)2，S1)。
S104, obtaining a histogram of the mode pair of points of the annular region symmetrical about the central point based on the mode pair;
counting each annular region ImAll the mode pairs in the annular region I can be obtained according to the frequency of the mode pairs in the annular region ImThe histogram description of all mode pairs in (1) is denoted as hmAccording to the combination relationship, the pattern pair has 16 × 16-256 combinations, so hmIs a 256-dimensional feature vector. Considering that (i, j) and (i ', j') are symmetric about the central point O, the actual calculation requires that the point (i, j) is sampled only in the upper half of the ring region, thereby avoiding unnecessary repeated statistics.
And S105, assembling the histogram to obtain the identification features.
All the annular spaces ImHistogram of (a)mAssembling is performed, for example, when n is 26, the identification feature H is (H)1,h2,……，h26). The assembling mode can be sequentially assembled from inside to outside.
In summary, the present embodiment provides a method for extracting circular image recognition features, after a circular image is obtained, performing normalization on the circular image, and dividing the circular image into an annular space including a plurality of annular regions by using a mask template; obtaining a pattern pair of points symmetrical about a central point of each annular region based on a method for calculating a code value; obtaining a histogram of a pattern pair of points symmetrical with respect to the central point for each of the ring-shaped regions based on the pattern pair; the histograms are combined to obtain identification features, the problem of angle rotation of the circular image can be solved by extracting the histograms of the mode pairs of the points which are symmetrical about the central point, the identification features of the circular image are extracted, and then the circular pattern is identified with high accuracy.
As shown in fig. 3, a flowchart of a circular image recognition feature extraction method embodiment 3 provided by the present invention is shown:
s301, carrying out size normalization on the first circular image to obtain a second circular image, wherein the first circular image is an obtained original circular image;
normalizing the first circular image size to a second circular image of M, i.e. IROIAnd the value of M is in direct proportion to the size of the first circular image. For example, the first circular image is divided into a 200 × 200 area, where 200 is 200 unit lengths.
S302, dividing the second circular image into an annular space formed by a plurality of annular areas by using a mask template;
generating a plurality of ring mask templates with increasing radius and partially overlapped in the same circle center, namely generating n rings with equal area in an overlapped mode by taking a point (M/2 ) as a circle center O in a 200X 200 area, wherein the mask template is psimFor example, taking n-26, i.e. the radius of each ring is,
Figure GDA0002451173720000061
using a mask template to align the second circular image IROIMasking to obtain several annular regions Im，Im＝IROI*ψm
S303, extracting a mode pair of points which are symmetrical about a central point of the annular region;
first, define the code value, as shown in FIG. 2, let C be the ring region ImP is ImAnd (3) extracting the coding value of the pixel point P from any one pixel point: and establishing a local coordinate system, wherein the direction along the CP is called a radial coordinate axis r, the direction vertical to the CP is called a tangential coordinate axis t, and P is the origin of the new local coordinate system. Respectively finding 4 symmetrical adjacent points in the radial direction r and the tangential direction t, and respectively recording the adjacent points as P in a counterclockwise direction1、P2、P3、P4In which P is1On the coordinate axis r, and P1On the opposite side of coordinate axis t with respect to C, and the distances between the 4 neighborhood points and point P are denoted as d, where d may be 1, where 1 is a unit length the same as 200, and is according to P1、P2、P3、P4The pixel value size is compared with the P point according to the following formula:
Figure GDA0002451173720000071
obtaining binary number code T ═ T (T)1T2T3T4) I (P) is the pixel value of point P, I (P)i) Is the taken point PiThe pixel value of (a) is obtained as a binary number code T ═ T (T)1T2T3T4)。
And then by the formula: F-8T1+4*T2+2*T3+T4And (4) converting the binary number code into a decimal number code, wherein F is the code value of the point P.
In order to increase the calculation speed and avoid interpolation calculation when calculating the code value of the P point, an approximation algorithm may be used when calculating the code value.
As shown in fig. 4, i.e., within the second circular image, the circular area is divided into 8 sectors in units of 45 degrees in the direction pointing from the center of the image to point P.
As shown in fig. 5 and 6, the position diagrams of the three arbitrarily selected points A, B and C and the flow chart of calculating the code value are shown. Wherein A, B, C forms an angle theta with the x-axis in the horizontal directionA＜22.5°,22.5°＜θB＜67.5°,67.5°＜θC＜112.5°。
Fig. 9 is a schematic diagram of the annular space and the point symmetrical with respect to the center point. Taking points (i, j) and (i ', j') in the annular region, wherein the two points are symmetrical about the central point O, and the coordinate relationship between the two points is as follows:
Figure GDA0002451173720000072
calculating the code value in the neighborhood (including (i, j)) of k multiplied by k (k ranges from 3 to 11) of (i, j), counting the code value with the most occurrence times, taking the code value as the representative mode of the point (i, j), and marking the code value as S1. Similarly, the representative pattern S of (i ', j') can be obtained2If S is1>S2Then, the pattern pair of points symmetric about the center point is represented as (S)1，S2) Otherwise, it is represented as (S)2，S1)。
S304, obtaining a histogram of the mode pair of points of the annular region which are symmetrical about the central point based on the mode pair;
counting each annular region ImAll the mode pairs in the annular region I can be obtained according to the frequency of the mode pairs in the annular region ImThe histogram description of all mode pairs in (1) is denoted as hmAccording to the combination relationship, the pattern pair has 16 × 16-256 combinations, so hmIs a 256-dimensional feature vector. Considering that (i, j) and (i ', j') are symmetric about the central point O, the actual calculation requires that the point (i, j) is sampled only in the upper half of the ring region, thereby avoiding unnecessary repeated statistics.
S305, assembling the histograms to obtain the identification features.
All the annular spaces ImHistogram of (a)mAssembling is performed, for example, when n is 26, the identification feature H is (H)1,h2,......，h26). The assembling mode can be sequentially assembled from inside to outside.
In summary, the present embodiment provides a method for extracting circular image recognition features, after a circular image is obtained, performing normalization on the circular image, and dividing the circular image into an annular space including a plurality of annular regions by using a mask template; obtaining a pattern pair of points symmetrical about a central point of each annular region based on a method for calculating a code value; obtaining a histogram of a pattern pair of points symmetrical with respect to the central point for each of the ring-shaped regions based on the pattern pair; the histograms are combined to obtain identification features, the problem of angle rotation of the circular image can be solved by extracting the histograms of the mode pairs of the points which are symmetrical about the central point, the identification features of the circular image are extracted, and then the circular pattern is identified with high accuracy.
In order to further optimize the scheme, the method further comprises the following steps:
s306, the identification features are sent to a support vector machine classifier, and the support vector machine classifier is used for identifying the identification features to obtain an identification result.
The support vector machine method is based on VC (virtual c-dimensional) theory of statistical learning theory and the principle of minimum structural risk, and seeks the best compromise between the complexity of the model (i.e. learning precision of a specific training sample) and the learning ability (i.e. ability of identifying any sample without error) according to limited sample information so as to obtain the best popularization ability.
The support vector machine classifier used here is a support vector machine classifier that is trained by using a plurality of samples and can perform image recognition according to recognition features. And after the transmission module sends the identification image to the support vector machine classifier, the original image can be identified.
To further optimize the present solution, the support vector machine classifier uses a linear kernel as its kernel.
In order to further optimize the scheme, the support vector machine classifier identifies the characteristics of the three channels to obtain an identification result. For example for the second circular image IROIThe three channels R, G, B get the corresponding recognition features HR、HG、HBAnd identifying to obtain a result. R, G, B respectively represent the color representation mode of the image, and the recognition rate obtained by recognizing the recognition features of three channels is higher than that obtained by recognizing the recognition features of one channel.
As shown in fig. 7, a specific structural schematic diagram of an embodiment 7 of the circular image recognition feature extraction system provided by the present invention is shown:
the system comprises a normalization module 701, a partitioning module 702, a calculation module 703, a statistic module 704 and an assembly module 705, wherein:
the normalization module 701 is connected with the partitioning module 702, the partitioning module 702 is connected with the calculation module 703, the calculation module 703 is connected with the statistics module 704, and the statistics module 704 is connected with the assembly module 705.
The normalization module 701 performs size normalization on the first circular image to obtain a second circular image, wherein the first circular image is an acquired original circular image;
normalizing the first circular image size to a second circular image of M, i.e. IROIAnd the value of M is in direct proportion to the size of the first circular image. For example, the first circular image is divided into a 200 × 200 area, where 200 is 200 unit lengths.
The partitioning module 702 divides the second circular image into an annular space formed by a plurality of annular regions by using a mask template;
generating a plurality of ring mask templates with increasing radius and partially overlapped in the same circle center, namely generating n rings with equal area in an overlapped mode by taking a point (M/2 ) as a circle center O in a 200X 200 area, wherein the mask template is psimFor example, taking n-26, i.e. the radius of each ring is,
Figure GDA0002451173720000091
using a mask template to align the second circular image IROIMasking to obtain several annular regions Im，Im＝IROI*ψm
The calculation module 703 extracts a pattern pair of points of the annular region that are symmetric about the central point;
first, define the code value, as shown in FIG. 2, let O be the ring region ImP is ImAnd (3) extracting the coding value of the pixel point P from any one pixel point: and establishing a local coordinate system, wherein the direction along OP is called as a radial coordinate axis r, the direction vertical to OP is called as a tangential coordinate axis t, and P is a new local coordinate system origin. Respectively finding 4 symmetrical adjacent points in the radial direction r and the tangential direction t, and respectively recording the adjacent points as P in a counterclockwise direction1、P2、P3、P4In which P is1On the coordinate axis r, and P1On the opposite side of the coordinate axis t with respect to O, and the distances between the 4 neighborhood points and the point P are denoted as d, where d may be 1, where 1 is a unit length the same as 200, and is according to P1、P2、P3、P4The pixel value size is compared with the P point according to the following formula:
Figure GDA0002451173720000101
obtaining binary number code T ═ T (T)1T2T3T4) I (P) is the pixel value of point P, I (P)i) Is the taken point PiThe pixel value of (a) is obtained as a binary number code T ═ T (T)1T2T3T4)。
And then by the formula: F-8T1+4*T2+2*T3+T4And (4) converting the binary number code into a decimal number code, wherein F is the code value of the point P.
In order to increase the calculation speed and avoid interpolation calculation when calculating the code value of the P point, an approximation algorithm may be used when calculating the code value.
As shown in fig. 4, i.e., within the second circular image, the circular area is divided into 7 sectors in units of 45 degrees in the direction pointing from the center of the image to the point P.
As shown in fig. 5 and 6, the position diagrams of the three arbitrarily selected points A, B and C and the flow chart of calculating the code value are shown. Wherein A, B, C forms an angle theta with the x-axis in the horizontal directionA＜22.5°,22.5°＜θB＜67.5°,67.5°＜θC＜112.5°。
Fig. 9 is a schematic diagram of the annular space and the point symmetrical with respect to the center point. Taking points (i, j) and (i ', j') in the annular region, wherein the two points are symmetrical about the central point O, and the coordinate relationship between the two points is as follows:
Figure GDA0002451173720000111
calculating the code value in the neighborhood (including (i, j)) of k multiplied by k (k ranges from 3 to 11) of (i, j), counting the code value with the most occurrence times, taking the code value as the representative mode of the point (i, j), and marking the code value as S1. Similarly, the representative pattern S of (i ', j') can be obtained2If S is1>S2Then, the pattern pair of points symmetric about the center point is represented as (S)1，S2) Otherwise, it is represented as (S)2，S1)。
The statistics module 704 obtains a histogram of the pattern pairs of points of the ring-shaped region that are symmetric about the center point based on the pattern pairs;
counting each annular region ImAll the mode pairs in the annular region I can be obtained according to the frequency of the mode pairs in the annular region ImThe histogram description of all mode pairs in (1) is denoted as hmAccording to the combination relationship, the pattern pair has 16 × 16-256 combinations, so hmIs a 256-dimensional feature vector. Considering that (i, j) and (i ', j') are symmetric about the central point O, the actual calculation requires that the point (i, j) is sampled only in the upper half of the ring region, thereby avoiding unnecessary repeated statistics.
The assembly module 705 assembles the histograms into identified features.
All the annular spaces ImHistogram of (a)mAssembling is performed, for example, when n is 26, the identification feature H is (H)1,h2,......，h26). The assembling mode can be sequentially assembled from inside to outside.
In summary, the present embodiment provides a circular image recognition feature extraction system, after obtaining a circular image, performing normalization processing on the circular image, and dividing the circular image into an annular space including a plurality of annular regions by using a mask template; obtaining a pattern pair of points symmetrical about a central point of each annular region based on a method for calculating a code value; obtaining a histogram of a pattern pair of points symmetrical with respect to the central point for each of the ring-shaped regions based on the pattern pair; the histograms are combined to obtain identification features, the problem of angle rotation of the circular image can be solved by extracting the histograms of the mode pairs of the points which are symmetrical about the central point, the identification features of the circular image are extracted, and then the circular pattern is identified with high accuracy.
As shown in fig. 8, a specific structural schematic diagram of an embodiment 8 of the circular image recognition feature extraction system provided by the present invention is shown:
the system comprises a normalization module 801, a partitioning module 802, a calculation module 803, a statistic module 804 and an assembly module 805, wherein:
the normalization module 801 is connected to the partitioning module 802, the partitioning module 802 is connected to the calculation module 803, the calculation module 803 is connected to the statistics module 804, and the statistics module 804 is connected to the assembly module 805.
The normalization module 801 performs size normalization on the first circular image to obtain a second circular image, wherein the first circular image is an acquired original circular image;
normalizing the first circular image size to a second circular image of M, i.e. IROIAnd the value of M is in direct proportion to the size of the first circular image. For example, the first circular image is divided into a 200 × 200 area, where 200 is 200 unit lengths.
The partitioning module 802 divides the second circular image into an annular space formed by a plurality of annular regions by using a mask template;
generating a plurality of ring mask templates with increasing radius and partially overlapped in the same circle center, namely generating n rings with equal area in an overlapped mode by taking a point (M/2 ) as a circle center O in a 200X 200 area, wherein the mask template is psimFor example, taking n-26, i.e. the radius of each ring is,
Figure GDA0002451173720000121
using a mask template to align the second circular image IROIMasking to obtain several annular regions Im，Im＝IROI*ψm
The calculation module 803 extracts a pattern pair of points of the annular region that are symmetric about the center point;
first, define the code value, as shown in FIG. 2, let O be the ring region ImP is ImAnd (3) extracting the coding value of the pixel point P from any one pixel point: and establishing a local coordinate system, wherein the direction along OP is called as a radial coordinate axis r, the direction vertical to OP is called as a tangential coordinate axis t, and P is a new local coordinate system origin. Respectively finding 4 symmetrical adjacent points in the radial direction r and the tangential direction t, and respectively recording the adjacent points as P in a counterclockwise direction1、P2、P3、P4In which P is1On the coordinate axis r, and P1On the opposite side of the coordinate axis t with respect to OThe distances between 4 neighborhood points and a point P are recorded as d, wherein the value of d can be 1, and 1 is the unit length which is the same as the unit length of 200, according to P1、P2、P3、P4The pixel value size is compared with the P point according to the following formula:
Figure GDA0002451173720000122
obtaining binary number code T ═ T (T)1T2T3T4) I (P) is the pixel value of point P, I (P)i) Is the taken point PiThe pixel value of (a) is obtained as a binary number code T ═ T (T)1T2T3T4)。
And then by the formula: F-8T1+4*T2+2*T3+T4And (4) converting the binary number code into a decimal number code, wherein F is the code value of the point P.
In order to increase the calculation speed and avoid interpolation calculation when calculating the code value of the P point, an approximation algorithm may be used when calculating the code value.
As shown in fig. 4, i.e., within the second circular image, the circular area is divided into 8 sectors in units of 45 degrees in the direction pointing from the center of the image to point P.
As shown in fig. 5 and 6, the position diagrams of the three arbitrarily selected points A, B and C and the flow chart of calculating the code value are shown. Wherein A, B, C forms an angle theta with the x-axis in the horizontal directionA＜22.5°,22.5°＜θB＜67.5°,67.5°＜θC＜112.5°。
Fig. 9 is a schematic diagram of the annular space and the point symmetrical with respect to the center point. Taking points (i, j) and (i ', j') in the annular region, wherein the two points are symmetrical about the central point O, and the coordinate relationship between the two points is as follows:
Figure GDA0002451173720000131
obtaining the coding of k multiplied by k (k value range is 3-11) in the neighborhood (including (i, j)) of (i, j)Code value, the code value with the largest number of occurrences is counted, and the code value is taken as a representative pattern of the point (i, j) and is denoted as S1. Similarly, the representative pattern S of (i ', j') can be obtained2If S is1>S2Then, the pattern pair of points symmetric about the center point is represented as (S)1，S2) Otherwise, it is represented as (S)2，S1)。
The statistics module 804 obtains a histogram of pattern pairs of points of the ring-shaped region that are symmetric about the center point based on the pattern pairs;
counting each annular region ImAll the mode pairs in the annular region I can be obtained according to the frequency of the mode pairs in the annular region ImThe histogram description of all mode pairs in (1) is denoted as hmAccording to the combination relationship, the pattern pair has 16 × 16-256 combinations, so hmIs a 256-dimensional feature vector. Considering that (i, j) and (i ', j') are symmetric about the central point O, the actual calculation requires that the point (i, j) is sampled only in the upper half of the ring region, thereby avoiding unnecessary repeated statistics.
The assembly module 805 assembles the histograms into identified features.
All the annular spaces ImHistogram of (a)mAssembling is performed, for example, when n is 26, the identification feature H is (H)1,h2,......，h26). The assembling mode can be sequentially assembled from inside to outside.
In summary, the present embodiment provides a circular image recognition feature extraction system, after obtaining a circular image, performing normalization processing on the circular image, and dividing the circular image into an annular space including a plurality of annular regions by using a mask template; obtaining a pattern pair of points symmetrical about a central point of each annular region based on a method for calculating a code value; obtaining a histogram of a pattern pair of points symmetrical with respect to the central point for each of the ring-shaped regions based on the pattern pair; the histograms are combined to obtain identification features, the problem of angle rotation of the circular image can be solved by extracting the histograms of the mode pairs of the points which are symmetrical about the central point, the identification features of the circular image are extracted, and then the circular pattern is identified with high accuracy.
To further optimize the present solution, the system further includes a support vector machine classifier 806, wherein:
support vector machine classifier 806 is connected to assembly module 805.
The assembling module 805 sends the identification features to the support vector machine classifier 806, and the support vector machine classifier 806 is used for identifying the identification features to obtain an identification result.
The support vector machine method is based on VC (virtual c-dimensional) theory of statistical learning theory and the principle of minimum structural risk, and seeks the best compromise between the complexity of the model (i.e. learning precision of a specific training sample) and the learning ability (i.e. ability of identifying any sample without error) according to limited sample information so as to obtain the best popularization ability.
As used herein, the support vector machine classifier 806 is a support vector machine classifier 806 that is trained using a plurality of samples and can perform image recognition based on recognition features. The transfer module may identify the original image by sending the identified image to the support vector machine classifier 806.
To further optimize the present solution, support vector machine classifier 806 uses a linear kernel as its kernel.
In order to further optimize the present solution, the support vector machine classifier 806 identifies the identification features of the three channels to obtain an identification result. For example for the second circular image IROIThe three channels R, G, B get the corresponding recognition features HR、HG、HBAnd identifying to obtain a result. R, G, B respectively represent the color representation mode of the image, and the recognition rate obtained by recognizing the recognition features of three channels is higher than that obtained by recognizing the recognition features of one channel.
The embodiments are described in a progressive manner in the specification, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (8)

1. A circular image recognition feature extraction method is characterized by comprising the following steps:
carrying out size normalization on a first circular image to obtain a second circular image, wherein the first circular image is an obtained original circular image;
dividing the second circular image into an annular space consisting of a plurality of annular areas by using a mask template;
extracting a pattern pair of points of the annular region which are symmetrical about the central point, and taking points (i, j) and (i ', j') in the annular region, wherein the two points are symmetrical about the central point O, and the coordinate relationship between the two points is as follows:
Figure FDA0002318293700000011
and (5) solving the code value in the k multiplied by k neighborhood of the (i, j), wherein the k value range is 3-11, counting the code value with the most occurrence times, taking the code value as a representative mode of the point (i, j), and marking the code value as S1Similarly, the representative pattern S of (i ', j') can be obtained2If S is1>S2Then, the pattern pair of points symmetric about the center point is represented as (S)1，S2) Otherwise, it is represented as (S)2，S1)；
Obtaining a histogram of the pattern pair of the points of the ring-shaped region symmetric about the central point based on the pattern pair;
and assembling the histograms to obtain identification features.
2. The method of claim 1, further comprising:
and sending the identification features to a support vector machine classifier, and identifying the identification features by using the support vector machine classifier to obtain an identification result.
3. The method of claim 2, wherein the support vector machine classifier uses a linear kernel function as its kernel function.
4. The method of claim 3, wherein the identifying features by the SVM classifier to obtain an identifying result specifically comprises:
and the support vector machine classifier identifies the characteristics of the three channels to obtain an identification result.
5. A circular image recognition feature extraction system, comprising:
the normalization module is used for carrying out size normalization on the first circular image to obtain a second circular image, wherein the first circular image is an acquired original circular image;
the partitioning module is used for dividing the second circular image into an annular space formed by a plurality of annular areas by using a mask template;
a calculation module for extracting a pattern pair of points of the annular region that are symmetric about a center point; taking points (i, j) and (i ', j') in the annular region, wherein the two points are symmetrical about the central point O, and the coordinate relationship between the two points is as follows:
Figure FDA0002318293700000021
and (5) solving the code value in the k multiplied by k neighborhood of the (i, j), wherein the k value range is 3-11, counting the code value with the most occurrence times, taking the code value as a representative mode of the point (i, j), and marking the code value as S1Similarly, the representative pattern S of (i ', j') can be obtained2If S is1>S2Then, the pattern pair of points symmetric about the center point is represented as (S)1，S2) Otherwise, it is represented as (S)2，S1)；
A statistical module to derive a histogram of the pattern pair of the points of the ring-shaped region symmetric about a center point based on the pattern pair;
and the assembling module is used for assembling the histogram to obtain the identification features.
6. The system of claim 5, further comprising a support vector machine classifier, wherein:
the support vector machine classifier is connected with the assembly module;
and sending the identification features to a support vector machine classifier, and identifying the identification features by using the support vector machine classifier to obtain an identification result.
7. The system of claim 6, wherein the support vector machine classifier uses a linear kernel function as its kernel function.
8. The system of claim 7, wherein the identifying features by the support vector machine classifier obtains an identifying result specifically as:
and the support vector machine classifier identifies the characteristics of the three channels to obtain an identification result.
CN201710164058.3A 2017-03-17 2017-03-17 Circular image recognition feature extraction method and system Active CN106980862B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710164058.3A CN106980862B (en) 2017-03-17 2017-03-17 Circular image recognition feature extraction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710164058.3A CN106980862B (en) 2017-03-17 2017-03-17 Circular image recognition feature extraction method and system

Publications (2)

Publication Number Publication Date
CN106980862A CN106980862A (en) 2017-07-25
CN106980862B true CN106980862B (en) 2020-06-09

Family

ID=59338838

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710164058.3A Active CN106980862B (en) 2017-03-17 2017-03-17 Circular image recognition feature extraction method and system

Country Status (1)

Country Link
CN (1) CN106980862B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108898573B (en) * 2018-04-23 2021-11-02 西安电子科技大学 Infrared small target rapid extraction method based on multidirectional annular gradient method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0712520A (en) * 1990-12-27 1995-01-17 Internatl Business Mach Corp <Ibm> Method for discovering standard
CN102156870A (en) * 2011-04-12 2011-08-17 张小军 Device and extraction method for extracting invariant characteristics of local rotation of image
CN103646239A (en) * 2013-12-25 2014-03-19 武汉大学 Polar coordinate Fourier transform based rotation invariance image characteristic extraction method
CN104199931A (en) * 2014-09-04 2014-12-10 厦门大学 Trademark image consistent semantic extraction method and trademark retrieval method
CN105426896A (en) * 2015-11-16 2016-03-23 成都神州数码索贝科技有限公司 Car logo automatic identification method and system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0712520A (en) * 1990-12-27 1995-01-17 Internatl Business Mach Corp <Ibm> Method for discovering standard
CN102156870A (en) * 2011-04-12 2011-08-17 张小军 Device and extraction method for extracting invariant characteristics of local rotation of image
CN103646239A (en) * 2013-12-25 2014-03-19 武汉大学 Polar coordinate Fourier transform based rotation invariance image characteristic extraction method
CN104199931A (en) * 2014-09-04 2014-12-10 厦门大学 Trademark image consistent semantic extraction method and trademark retrieval method
CN105426896A (en) * 2015-11-16 2016-03-23 成都神州数码索贝科技有限公司 Car logo automatic identification method and system

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Extracting local texture features for image-based coin recognition;Shen, L 等;《IET Image Processing》;20110831;第5卷(第5期);394-401 *
HeikkiläM.等.Description of Interest Regions with Center-Symmetric Local Binary Patterns.《Computer Vision, Graphics and Image Processing》.2006,第4338卷58-69. *
一种抗旋转、尺度和平移处理的图像水印算法;姚俊 等;《计算机应用》;20050103;第24卷(第12期);19-21,27 *
基于局部二值模式的交通标志识别算法研究;贾月圆;《中国优秀硕士学位论文全文数据库 信息科技辑》;20131215(第S2期);1138-1376 *

Also Published As

Publication number Publication date
CN106980862A (en) 2017-07-25

Similar Documents

Publication Publication Date Title
US11423701B2 (en) Gesture recognition method and terminal device and computer readable storage medium using the same
EP3274921B1 (en) Multi-layer skin detection and fused hand pose matching
CN108596197B (en) Seal matching method and device
US5054094A (en) Rotationally impervious feature extraction for optical character recognition
CN108764004B (en) Annular coding mark point decoding and identifying method based on coding ring sampling
CN110472625B (en) Chinese chess piece visual identification method based on Fourier descriptor
CN110852311A (en) Three-dimensional human hand key point positioning method and device
Costa et al. A fully automatic method for recognizing hand configurations of Brazilian sign language
Fang et al. Real-time hand posture recognition using hand geometric features and fisher vector
CN108830283B (en) Image feature point matching method
CN106503694A (en) Digit recognition method based on eight neighborhood feature
US20130342444A1 (en) Method and Apparatus for Hand Gesture Trajectory Recognition
CN103198299A (en) Face recognition method based on combination of multi-direction dimensions and Gabor phase projection characteristics
US9117132B2 (en) System and method facilitating designing of classifier while recognizing characters in a video
US9058517B1 (en) Pattern recognition system and method using Gabor functions
CN111079684B (en) Three-dimensional face detection method based on rough-fine fitting
CN106980862B (en) Circular image recognition feature extraction method and system
KR101151435B1 (en) Apparatus and method of recognizing a face
CN110599478A (en) Image area copying and pasting tampering detection method
JP4541995B2 (en) Figure recognition method
Li et al. Script identification of camera-based images
CN112396638A (en) Image processing method, terminal and computer readable storage medium
Valiente et al. A process for text recognition of generic identification documents over cloud computing
JP2007004709A (en) Object pattern detecting method and device thereof
CN105139428A (en) Quaternion based speeded up robust features (SURF) description method and system for color image

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