CN106250901A - A kind of digit recognition method based on image feature information - Google Patents
A kind of digit recognition method based on image feature information Download PDFInfo
- Publication number
- CN106250901A CN106250901A CN201610143280.0A CN201610143280A CN106250901A CN 106250901 A CN106250901 A CN 106250901A CN 201610143280 A CN201610143280 A CN 201610143280A CN 106250901 A CN106250901 A CN 106250901A
- Authority
- CN
- China
- Prior art keywords
- image
- numeral
- method based
- feature information
- recognition method
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
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
- G06V10/20—Image preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/24—Character recognition characterised by the processing or recognition method
- G06V30/242—Division of the character sequences into groups prior to recognition; Selection of dictionaries
- G06V30/244—Division of the character sequences into groups prior to recognition; Selection of dictionaries using graphical properties, e.g. alphabet type or font
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/24—Character recognition characterised by the processing or recognition method
- G06V30/248—Character recognition characterised by the processing or recognition method involving plural approaches, e.g. verification by template match; Resolving confusion among similar patterns, e.g. "O" versus "Q"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
A kind of digit recognition method based on image feature information, comprises the steps: step one, the image of acquisition reference numbers;Step 2, in step one obtain image carry out standard deviation filtering;Step 3, image filtered in step 2 is carried out binary conversion treatment;Step 4, solve numeral connected region;Step 5, length, width, area and the eccentricity feature of extraction numeral;Step 6, the digital picture extracted in test image;Step 7, structure similarity function;Step 8, the feature of acquisition test picture numeral;Numeral and the similarity of reference numbers image in step 9, calculating test picture;If the similarity in step 10 step 9 is less than threshold value p, then explanatory diagram picture do not has numeral;If the similarity in step 9 is more than threshold value p, then identify the numeral in image.The impact that the identification of numeral is brought by the factors such as the present invention solves environmental change extraneous during traditional numeral identifies in method, the change in location of test.
Description
Technical field
The present invention relates to a kind of digit recognition method based on image feature information, belong to numeral identification field.
Background technology
Owing to image acquiring device faster and can obtain image by ratio easily, image processing techniques is in reality
The production application on border has the beating and double roasting field of more use, especially Nicotiana tabacum L., the monitoring of such as on-the-spot flow, raw material matter
The monitoring of amount, the inspection etc. of finished product, and often can also be more severe in production environment, it is desirable to the process section analyzed in real time possesses
Stronger advantage;Sometimes, not only the actual state produced is monitored, but also need the result monitoring and phase
Corresponding station or mark are associated, and therefore the numeral in image identifies it is important in image procossing and application
Ring;In the numeral of conventional image identifies, the similarity of the color histograms scattergram of with good grounds image is identified, such as week
The patent that number of patent application is 201510456753.8 of elegant sesame, patent name is: a kind of gradient image and similarity feature add
The digit recognition method of power;Based on image area and fundamental digital template modeling data have Model Identification, such as University Of Suzhou
The patent of the Patent No. 201310246733.9 of application, patent name is: handwriting digital based on image covariance feature
Recognition methods and device;During conventional image recognition, owing to actual production scene situation is complicated and changeable, above-mentioned side
Method typically can run into following problem in actual application process: the change of (1) ambient is to color of image distribution etc.
Impact, the change of external environment can affect the change of distribution of color, reduces the precision identified;(2) the not solid of digit position is identified
The qualitative impact on extracting feature;Identify that the difference of position can affect the calculating of its image similarity;Because the number at paving leaf platform
Word location information can translate;A kind of new applicable digit recognition method it has to be ensured that can be applicable to different outside
The change of boundary's environment, the change of the translation of diverse location and rotation etc., the identification of different digital size, local similarity is known
Not.
The most accurate and simple from removable, ambient light is in constantly change, in the environment of position translation and rotation
Identifying corresponding digital information in image accurately is the research contents of the present invention.
Summary of the invention
It is an object of the invention to provide a kind of digit recognition method based on image feature information, to solve above-mentioned asking
Topic.
Present invention employs following technical scheme:
A kind of digit recognition method based on image feature information, it is characterised in that comprise the steps:
Step one, the image of acquisition reference numbers;
Step 2, in step one obtain image carry out standard deviation filtering;
Step 3, image filtered in step 2 is carried out binary conversion treatment;
Step 4, solve numeral connected region;
Step 5, length, width, area and the eccentricity feature of extraction numeral;
Step 6, the digital picture extracted in test image;
Step 7, structure similarity function;
Step 8, the feature of acquisition test picture numeral;
Numeral and the similarity of reference numbers image in step 9, calculating test picture;
If the similarity in step 10 step 9 is less than threshold value p, then explanatory diagram picture do not has numeral;If in step 9
Similarity is more than threshold value p, then identify the numeral in image.
Further, the digit recognition method based on image feature information of the present invention, it is also possible to have a feature in that it
In, in step one, reference numbers is numeral 1~8.
Further, the digit recognition method based on image feature information of the present invention, it is also possible to have a feature in that it
In, in step 3, the process of binaryzation is as follows:
Calculating threshold value t that gradation of image processes, the computational methods of t are: by the gamma characteristic of image, divide the image into background
With target two parts, the inter-class variance between background and target is the biggest, illustrates that the two-part difference of pie graph picture is the biggest, works as portion
Subhead mislabels and is divided into background or part background mistake to be divided into target that two parts difference all can be caused to diminish, it is assumed that a pictures has n
Individual pixel, wherein gray value is n1 less than the pixel count of threshold value t, is n2 more than or equal to the pixel count of threshold value t, (n1+n2=n),
W1 and w2 represents the respective proportion of both pixels,
W1=n1/n (1)
W2=n2/n (2)
Again assuming that, all gray values are respectively μ 1 and σ 1, all gray values less than meansigma methods and the variance of the pixel of threshold value t
Being respectively μ 2 and σ 2 more than or equal to the meansigma methods of pixel and the variance of threshold value t, thus it is possible to obtain class inherited P, t value is for making
Obtain P and obtain the data of maximum;T is worth solving and can use Non-Linear Programming and genetic algorithm;
P=w1w2 (μ 1-μ 2)2 (3)
Using the t value that solves as threshold value, carry out image array J binary conversion treatment;Obtain binaryzation matrix Img.
Further, the digit recognition method based on image feature information of the present invention, it is also possible to have a feature in that it
In, in step 4, the finding method of connected region is: according to binaryzation matrix Img, differentiates around each pixel coordinate point and is
The no data having same grayscale are distributed, concrete for travel through labelling image exactly, by the pixel value index marker of labelling image
Set, it is each that the pixel finding the value representing current collection minimum in tag set to be assigned to original image current location obtains Nicotiana tabacum L.
The connected relation of grid area image, can find region according to four connections or eight connectivity is found;Draw each connected region
Sequence number num matrix;And the type of each connected region.
Further, the digit recognition method based on image feature information of the present invention, it is also possible to it is right to have a feature in that
The location of pixels at each connected region place calculates the pixel that this connected region is comprised, and total number of pixel is pixel faces
Long-pending.
Further, the digit recognition method based on image feature information of the present invention, it is also possible to have a feature in that it
In, to each connected region num=i, find and comprise each connected region num=i, the external square of minimum of i=1:max (num)
Shape;The long a of rectangle, wide b, be the length of each connected region, wide.
Further, the digit recognition method based on image feature information of the present invention, it is also possible to it is right to have a feature in that
Each connected region num, finds and comprises each connected region num=(i), the fitted ellipse curve of i=1:max (num);Oval
Major axis c, short axle is d, and oval focal length is f, and obtains eccentricity e of each connected region.
Further, the digit recognition method based on image feature information of the present invention, it is also possible to have a feature in that step
In rapid six, test picture is carried out color segmentation, is partitioned into the region at numeral place.
Further, the digit recognition method based on image feature information of the present invention, it is also possible to have a feature in that structure
Make similarity measurements flow function:
I=1~8;What corr represented is correlation coefficient, erro be one set the least numeral, take erro <
0.000000005;Sqrt is square root;Sum is summation, and abs is absolute value.
Further, the digit recognition method based on image feature information of the present invention, it is also possible to have a feature in that threshold
Value p is set to 0.5%.
The beneficial effect of the invention
The present invention solves environmental change extraneous during traditional numeral identifies, the position of test in method
The impact that the identification of numeral is brought by the factors such as change;
The present invention uses the invariant features of image digitization in application, solves stability and the precision of numeral identification,
For the identification of numeral in actual production, and the identification of numeral and technique, the association of raw materials quality, image procossing application is laid
Good basis.
Accompanying drawing explanation
Fig. 1 is the blank belt reference numbers figure of Nicotiana tabacum L.;
Fig. 2 is the standard deviation filtering figure of reference numbers image;
Fig. 3 is that quasi-deviation filters binary conversion treatment figure;
Fig. 4 is connected region digital distribution coordinate diagram;
Fig. 5 is connected region minimum enclosed rectangle figure;
Fig. 6 is test picture original graph;
Fig. 7 is test picture numeric area segmentation figure;
Fig. 8 is that test digit position there occurs skew original graph;
Fig. 9 is that test numeral ambient background there occurs change original graph;
Figure 10 is that test digit position rotates and the same hourly variation of external light source;
Figure 11 is a kind of digit recognition method flow chart based on image feature information.
Detailed description of the invention
The detailed description of the invention of the present invention is described below in conjunction with accompanying drawing.
Embodiment 1:
As shown in figure 11, the step of digit recognition method based on image feature information includes:
Step 101, first obtaining the single image of benchmark numeral 1-8 in static state, numeral is as shown in Figure 1.
Step 102, to obtain view data carry out standard deviation filtering, calculate image local color standard deviation time,
When input picture I performs standard deviation filtering, arranging parameter is that the s specifying neighborhood ties up unit matrix, and s is the nature of odd number
Number, the named J of filtered view datai, i=1...8;
Step 103, binary conversion treatment, specifically include following steps:
Step 103a: calculating threshold value t that gradation of image processes, the computational methods of t are: by the gamma characteristic of image, will figure
As being divided into background and target two parts.Inter-class variance between background and target is the biggest, and the two-part difference of pie graph picture is described
The biggest, when partial target mistake is divided into background or part background mistake to be divided into target that two parts difference all can be caused to diminish.Assuming that one
Pictures has n pixel, and wherein gray value is n1 less than the pixel count of threshold value t, is n2 more than or equal to the pixel count of threshold value t,
(n1+n2=n).W1 and w2 represents the respective proportion of both pixels.
W1=n1/n (1)
W2=n2/n (2)
Again assuming that, all gray values are respectively μ 1 and σ 1, all gray values less than meansigma methods and the variance of the pixel of threshold value t
It is respectively μ 2 and σ 2 more than or equal to the meansigma methods of pixel and the variance of threshold value t.Thus it is possible to obtain class inherited P, t value is for making
Obtain P and obtain the data of maximum;Solving of t value can use Non-Linear Programming and genetic algorithm;
P=w1w2 (μ 1-μ 2)2 (3)
Obtain the filtering matrix J of image;(as a example by numeral 8, lower same)
The view data obtained is carried out standard deviation filtering, when calculating the local color standard deviation of image, to input
When image I performs standard deviation filtering, arranging parameter is that the s specifying neighborhood ties up unit matrix, and s is the natural number of odd number, in this reality
Execute in mode, choose s=9;The named J of filtered view datai, i=1...8;
Obtain figure as shown in Figure 2.Trying to achieve threshold value t according to formula (3) is 0.2275;
Step 103b, binary conversion treatment that J is carried out:
According to the t value solved as threshold value, carry out image array J binary conversion treatment;Obtain binaryzation matrix Img,
The image arrived is as shown in Figure 3.
Step 104, solve the connected region of numeral, specifically include following steps:
Step 104a: view data is normalized, it is achieved the normalization operation of image array.So-called normalization
Be exactly the value of each element making matrix between zero and one.
Step 104b, the information trying to achieve connected region are as follows:
The number of connected region is 1;The coordinate diagram of connected region is shown in Fig. 4.The connected region of Img is marked, comprises
Marked the class label of each connected region in Img, the value of these labels is 1,2, num (num represent connected region
Number).The finding method of connected region is according to binaryzation matrix Img, differentiate whether have around each pixel coordinate point identical
Gray scale data distribution, concrete for travel through labelling image exactly, by the pixel value index marker set of labelling image, find
Tag set represents the minimum value of current collection be assigned to the pixel of original image current location and obtain each grid area of Nicotiana tabacum L.
The connected relation of image, can find region according to 4 connections or 8 connections are found;Draw the sequence number num square of each connected region
Battle array;And the type of each connected region.
Step 105 and step 106, the long width characteristics extracting numeral method as follows:
The minimum external matrix of acquisition pixel:
To each connected region num=i, find and comprise each connected region num=i, outside the minimum of i=1:max (num)
Connect rectangle;The long a of rectangle, wide b, be the length of each connected region, wide;The method finding minimum rectangle is as follows:
Find the two-dimensional pixel coordinate of all num=i, i=1:max (num);Minima a1 of the first dimension data, maximum
Value a2, and minima b1 of the second dimension data, maximum b2, x=a1;X=a2;Y=b1;Y=b2;Article four, straight line is determined
Region be minimum rectangle;The long a of rectangle, wide b calculation is as follows:
A=max (a2-a1, b2-b1) (4)
B=min (a2-a1, b2-b1) (5)
As it is shown in figure 5, calculate to long a=127;B=65.
Step 107: the location of pixels at each connected region place is calculated the pixel that this connected region is comprised, as
Total number of element is elemental area Area;The area obtaining image is 4627 pixels.
Step 108, the eccentricity feature of extraction numeral:
Trying to achieve the matched curve of ellipse, method is as follows:
To each connected region num, finding and comprise each connected region num=(i), the matching of i=1:max (num) is ellipse
Circular curve;Oval major axis c, short axle is d, and oval focal length is f, and oval eccentricity e uses formula 6 to calculate.
The method of fitted ellipse curve is as follows: first look for the coordinate points of each connected region;According to these coordinate point ranges
Go out ellipse algebraically primitive form:
Ax2+Bxy+Cy2+ Dx+Ey+F=0 (7)
Using including but not limited to genetic algorithm, ant group algorithm, least square etc. simulates curvilinear equation;
Calculating oval eccentricity is 0.8819.
Step 201 (does not shows in Figure 11): according to solving in i=1~8 single image numerals, long, wide, region area,
And eccentricity construction feature matrix K Q, form the profile identification of different pieces of information;
KQ=[a, b, Area, f]; (10)
The KQ of table 1 image digitization 1~8
Step 109, the view data of acquisition test picture;Test image in present embodiment is as shown in Figure 6.
Step 109a: the data of test picture are carried out color segmentation, is partitioned into the region at numeral place, the figure after segmentation
As shown in Figure 7.
Step 109b, ask for test image image digitization characteristic KQ1, the results are shown in Table 2;
Table 2 embodiment 1 is tested the KQ1 of image
Length in pixels | Pixel wide | Region area | Eccentricity |
131 | 77 | 3619 | 0.82462 |
Step 109c, signature identification matrix K Q of known data and test image feature data KQ1 are carried out normalizing
Changing, normalized computational methods are:
Normalization characteristic matrix K Q, is shown in Table 3, is corresponding in turn to the normalization characteristic matrix of numeral 1 to 8 in table 3 from top to bottom
KQ。
Table 3 normalization characteristic matrix K Q
0.666667 | 0 | 0 | 1 |
0.111111 | 1 | 0.738875 | 0.400742 |
0 | 0.732394 | 0.632586 | 0.528803 |
0.333333 | 0.971831 | 0.739978 | 0 |
1 | 0.788732 | 0.910997 | 0.464747 |
0 | 0.56338 | 0.644722 | 0.4981 |
0.666667 | 0.676056 | 0.358588 | 0.717125 |
0.611111 | 0.535211 | 1 | 0.658166 |
Normalization characteristic matrix K Q1, is shown in Table 4:
Table 4 normalization characteristic matrix K Q1
0.833333 | 0.704225 | 0.629275 | 0.465422 |
Step 111: obtain the similarity data of test picture, and judge to have in figure nil:
Structure similarity measurements flow function:
I=1:8;What corr represented is correlation coefficient, erro be one set the least numeral, take erro <
0.000000005;Sqrt is square root;Sum is summation, and abs is absolute value;
The similarity that numeral to be measured in table 5 present embodiment is digital with benchmark
Digital ID | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
Similarity | 0.299606 | 0.232027 | 0.685907 | 0.556213 | 2.629166 | 0.781231 | 0.013483 | 0.499124 |
Step 112, if maximum similarity max (sim) < P, then enters step 114, does not has numeral in test picture;
Step 113, if max (sim) >=P, then enters step 115, solves and test image digitization feature:
Now maximum with the Digital ID characteristic similarity data corresponding to feature are the numeral identification of test image
Result;Threshold value P be chosen as 0.5%.
In present embodiment, the data of the test picture trying to achieve identification are numeral 5.
Embodiment 2
In the case of test digit position there occurs skew, contrast existing two kinds of digit recognition methods, i.e. COLOR COMPOSITION THROUGH DISTRIBUTION
Recognition methods and digit area recognition methods and the inventive method effect to numeral identification.
As shown in Figure 8, what wherein file 23-29 characterized is to test image with data template image in situ, and camera is swept
The picture retouched;What 30-36 characterized is that test Pictures location occurs skew and creates the picture of rotary taking.
The existing two kinds of numeral identification sides of table 6 contrast with the method for the present invention
Accuracy contrasts, and result is as shown in table 7.
The various digit recognition method of table 7 is the contrast table of accuracy in the case of digit position there occurs skew
Method name | COLOR COMPOSITION THROUGH DISTRIBUTION identification | Digit area identification | This method identification |
Original position test accuracy | 85.714% | 42.857% | 100% |
Test accuracy after change in location | 28.571% | 42.857% | 85.714% |
Embodiment 3
In the case of ambient background there occurs change, contrast existing two kinds of digit recognition methods and the inventive method
Effect to numeral identification.
As it is shown in figure 9, file 37-44 sign is that the position testing image there occurs translation, the ambient light of test sample
Line there occurs and changes largely, is primarily referred to as ambient and becomes strong.
During the change of table 8 light, existing two kinds of numeral identification sides contrast with the method for the present invention
Accuracy comparing result is as shown in table 9:
The accuracy rate that the various digit recognition method of table 9 identifies in the case of background light changes.
Method name | COLOR COMPOSITION THROUGH DISTRIBUTION identification | Digit area identification | This method identification |
Background light changes test accuracy | 12.500% | 0% | 100% |
Embodiment 4
Test digit position rotates and external light source changes in the case of simultaneously, contrast existing two kinds of numerals and know
Other method and the inventive method effect to numeral identification.As shown in Figure 10, wherein file 45-56 characterize be test image
Position there occurs rotation, and the ambient of test sample there occurs and changes largely, is primarily referred to as ambient and becomes strong.Respectively
Method is as shown in table 10 to the testing result of each figure.
The contrast of each recognition methods when table 10 digit position rotates and external light source changes simultaneously
4) accuracy contrast, is shown in Table 11.
Table 11 accuracy contrasts
Claims (10)
1. a digit recognition method based on image feature information, it is characterised in that comprise the steps:
Step one, the image of acquisition reference numbers;
Step 2, in step one obtain image carry out standard deviation filtering;
Step 3, image filtered in step 2 is carried out binary conversion treatment;
Step 4, solve numeral connected region;
Step 5, length, width, area and the eccentricity feature of extraction numeral;
Step 6, the digital picture extracted in test image;
Step 7, structure similarity function;
Step 8, the feature of acquisition test picture numeral;
Numeral and the similarity of reference numbers image in step 9, calculating test picture;
If the similarity in step 10 step 9 is less than threshold value p, then explanatory diagram picture do not has numeral;If similar in step 9
Degree more than threshold value p, then identifies the numeral in image.
2. digit recognition method based on image feature information as claimed in claim 1, it is characterised in that:
Wherein, in step one, reference numbers is numeral 1~8.
3. digit recognition method based on image feature information as claimed in claim 1, it is characterised in that:
Wherein, in step 3, the process of binaryzation is as follows:
Calculating threshold value t that gradation of image processes, the computational methods of t are: by the gamma characteristic of image, divide the image into background and mesh
Mark two parts, the inter-class variance between background and target is the biggest, illustrates that the two-part difference of pie graph picture is the biggest, when part mesh
Mislabel and be divided into background or part background mistake to be divided into target that two parts difference all can be caused to diminish, it is assumed that a pictures has n picture
Element, wherein gray value is n1 less than the pixel count of threshold value t, is n2 more than or equal to the pixel count of threshold value t, (n1+n2=n), w1 and
W2 represents the respective proportion of both pixels,
W1=n1/n (1)
W2=n2/n (2)
Again assuming that, all gray values are respectively μ 1 and σ 1 less than meansigma methods and the variance of the pixel of threshold value t, and all gray values are more than
Being respectively μ 2 and σ 2 equal to the meansigma methods of pixel and the variance of threshold value t, thus it is possible to obtain class inherited P, t value takes for making P
Obtain the data of maximum;T is worth solving and can use Non-Linear Programming and genetic algorithm;
P=w1w2 (μ 1-μ 2)2 (3)
Using the t value that solves as threshold value, carry out image array J binary conversion treatment;Obtain binaryzation matrix Img.
4. digit recognition method based on image feature information as claimed in claim 1, it is characterised in that:
Wherein, in step 4, the finding method of connected region is: according to binaryzation matrix Img, differentiates each pixel coordinate point
Around whether have the data of same grayscale to be distributed, concrete for travel through labelling image exactly, by the pixel value rope of labelling image
Drawing tag set, the pixel finding the value representing current collection minimum in tag set to be assigned to original image current location obtains cigarette
The connected relation of leaf each grid area image, can find region according to four connections or eight connectivity is found;Draw each company
The sequence number num matrix in logical region;And the type of each connected region.
5. digit recognition method based on image feature information as claimed in claim 4, it is characterised in that:
In step 5, the extracting method of the area features of numeral is: the location of pixels at each connected region place is calculated this
The pixel that connected region is comprised, total number of pixel is elemental area.
6. digit recognition method based on image feature information as claimed in claim 1, it is characterised in that:
In step 5, the extracting method of the length and width of numeral is: to each connected region num=i, finds and comprises each company
The minimum enclosed rectangle of logical region num=i, i=1:max (num);The long a of rectangle, wide b, be the length of each connected region,
Wide.
7. digit recognition method based on image feature information as claimed in claim 1, it is characterised in that:
In step 5, to each connected region num, find and comprise each connected region num=(i), the plan of i=1:max (num)
Close elliptic curve;Oval major axis c, short axle is d, and oval focal length is f, and obtains eccentricity e of each connected region.
8. digit recognition method based on image feature information as claimed in claim 1, it is characterised in that:
In step 6, test picture is carried out color segmentation, is partitioned into the region at numeral place.
9. digit recognition method based on image feature information as claimed in claim 1, it is characterised in that:
In step 7, structure similarity measurements flow function is as follows:
I=1~8;What corr represented is correlation coefficient, erro be one set the least numeral, take erro <
0.000000005;Sqrt is square root;Sum is summation, and abs is absolute value.
10. digit recognition method based on image feature information as claimed in claim 1, it is characterised in that:
In step 10, threshold value p is set to 0.5%.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610143280.0A CN106250901A (en) | 2016-03-14 | 2016-03-14 | A kind of digit recognition method based on image feature information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610143280.0A CN106250901A (en) | 2016-03-14 | 2016-03-14 | A kind of digit recognition method based on image feature information |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106250901A true CN106250901A (en) | 2016-12-21 |
Family
ID=57626447
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610143280.0A Pending CN106250901A (en) | 2016-03-14 | 2016-03-14 | A kind of digit recognition method based on image feature information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106250901A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108268895A (en) * | 2018-01-12 | 2018-07-10 | 上海烟草集团有限责任公司 | The recognition methods of tobacco leaf position, electronic equipment and storage medium based on machine vision |
CN108563967A (en) * | 2018-03-15 | 2018-09-21 | 青岛海信移动通信技术股份有限公司 | A kind of screenshot method and device |
CN107330465B (en) * | 2017-06-30 | 2019-07-30 | 清华大学深圳研究生院 | A kind of images steganalysis method and device |
CN111325214A (en) * | 2020-02-27 | 2020-06-23 | 珠海格力智能装备有限公司 | Jet printing character extraction processing method and device, storage medium and electronic equipment |
CN112488114A (en) * | 2020-11-13 | 2021-03-12 | 宁波多牛大数据网络技术有限公司 | Picture synthesis method and device and character recognition system |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104463195A (en) * | 2014-11-08 | 2015-03-25 | 沈阳工业大学 | Printing style digital recognition method based on template matching |
CN104700092A (en) * | 2015-03-26 | 2015-06-10 | 南京理工大学 | Small-character number identification method based on template and feature matching |
-
2016
- 2016-03-14 CN CN201610143280.0A patent/CN106250901A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104463195A (en) * | 2014-11-08 | 2015-03-25 | 沈阳工业大学 | Printing style digital recognition method based on template matching |
CN104700092A (en) * | 2015-03-26 | 2015-06-10 | 南京理工大学 | Small-character number identification method based on template and feature matching |
Non-Patent Citations (3)
Title |
---|
DAN CLAUDIU CIRESAN ETC.: ""Deep Big Simple Neural Nets Excel on Hand-written Digit Recognition"", 《ARXIV:1003.0358V1 CS.NE》 * |
李刚等: ""基于图像直方图的车牌图像二值化方法研究"", 《交通运输系统工程与信息》 * |
王璇等: ""基于BP神经网络的手写数字识别的算法"", 《控制理论与应用》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107330465B (en) * | 2017-06-30 | 2019-07-30 | 清华大学深圳研究生院 | A kind of images steganalysis method and device |
CN108268895A (en) * | 2018-01-12 | 2018-07-10 | 上海烟草集团有限责任公司 | The recognition methods of tobacco leaf position, electronic equipment and storage medium based on machine vision |
CN108563967A (en) * | 2018-03-15 | 2018-09-21 | 青岛海信移动通信技术股份有限公司 | A kind of screenshot method and device |
CN111325214A (en) * | 2020-02-27 | 2020-06-23 | 珠海格力智能装备有限公司 | Jet printing character extraction processing method and device, storage medium and electronic equipment |
CN111325214B (en) * | 2020-02-27 | 2023-02-14 | 珠海格力智能装备有限公司 | Jet printing character extraction processing method and device, storage medium and electronic equipment |
CN112488114A (en) * | 2020-11-13 | 2021-03-12 | 宁波多牛大数据网络技术有限公司 | Picture synthesis method and device and character recognition system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106250901A (en) | A kind of digit recognition method based on image feature information | |
Miao et al. | A semi-automatic method for road centerline extraction from VHR images | |
CN102324032B (en) | Texture feature extraction method for gray level co-occurrence matrix in polar coordinate system | |
CN103400151A (en) | Optical remote-sensing image, GIS automatic registration and water body extraction integrated method | |
Wang et al. | Cracking classification using minimum rectangular cover–based support vector machine | |
CN108074243A (en) | A kind of cellular localization method and cell segmentation method | |
Yuan et al. | Learning to count buildings in diverse aerial scenes | |
CN108052886A (en) | A kind of puccinia striiformis uredospore programming count method of counting | |
CN104680185B (en) | Hyperspectral image classification method based on boundary point reclassification | |
CN104751475A (en) | Feature point optimization matching method for static image object recognition | |
Han et al. | Vision-based crack detection of asphalt pavement using deep convolutional neural network | |
CN105243387A (en) | Open-pit mine typical ground object classification method based on UAV image | |
CN115599844A (en) | Visual detection method for misloading and neglected loading of airplane airfoil connecting piece | |
CN112489026A (en) | Asphalt pavement disease detection method based on multi-branch parallel convolution neural network | |
Chopin et al. | A hybrid approach for improving image segmentation: application to phenotyping of wheat leaves | |
CN107316296A (en) | A kind of method for detecting change of remote sensing image and device based on logarithmic transformation | |
CN117853722A (en) | Steel metallographic structure segmentation method integrating superpixel information | |
CN114332534A (en) | Hyperspectral image small sample classification method | |
CN111882573B (en) | Cultivated land block extraction method and system based on high-resolution image data | |
CN107657262B (en) | A kind of computer automatic sorting Accuracy Assessment | |
Gajanan et al. | Detection of leaf disease using feature extraction for Android based system | |
CN103093239B (en) | A kind of merged point to neighborhood information build drawing method | |
CN112381730A (en) | Remote sensing image data amplification method | |
CN109376619B (en) | Cell detection method | |
Gajanan et al. | Android based plant disease identification system using feature extraction technique |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20161221 |
|
WD01 | Invention patent application deemed withdrawn after publication |