CN100485714C - Method and device for recognizing test paper score - Google Patents
Method and device for recognizing test paper score Download PDFInfo
- Publication number
- CN100485714C CN100485714C CNB200710039969XA CN200710039969A CN100485714C CN 100485714 C CN100485714 C CN 100485714C CN B200710039969X A CNB200710039969X A CN B200710039969XA CN 200710039969 A CN200710039969 A CN 200710039969A CN 100485714 C CN100485714 C CN 100485714C
- Authority
- CN
- China
- Prior art keywords
- layer
- output
- error
- input
- score
- 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.)
- Expired - Fee Related
Links
Images
Abstract
The invention is related to an examination paper score identification method and apparatus, the method is realized by identifying the script handwriting of the score through a multi-layer perception network and which involves the study training process and identification process, the apparatus includes a desktop for placing the test paper, an upper panel being parallel to the desktop supported by 3-4 adjustable columns, wherein a pick-up head aiming at the test paper score, a laser lamp radiating the score region and 1-4 auxiliary lamps illuminating the test paper are provided on the upper panel, and the pick-up head is connected with a computer. The apparatus is easy to construct and convenient in operation, it can identify the handwritten scores of the test papers accurately and efficiently, thereby capable of meeting with the demand of realistic application.
Description
Technical field
The present invention relates to a kind of Handwritten Numeral Recognition Method and device, particularly a kind of test paper score recognition methods and device.
Background technology
The method of in the teaching process test paper score being handled can be divided into two classes now, and a class is the artificial treatment method, and a class is to utilize light reflection principle answer to be sticked into the method for line scanning.The former utilizes artificial method to processings of imparting knowledge to students of the mark of paper volume face, consumption great amount of manpower and material resources, and efficient is low, makes mistakes easily.The The latter light reflection principle requires to scribble with specific pen, though be widely used at present to answer sheet scanning identification, the discrimination height, but answer sheet pattern and Painting and writing tool are required to fix the cost height, practical application is difficult for realizing that its paper form is single, only is suitable for standardized examination paper.
Summary of the invention
The objective of the invention is to defective, the test paper score recognition methods and the device that provide a kind of discrimination height, are easy to use, the needs of full teaching at the prior art existence.
For reaching above-mentioned goal of the invention, design of the present invention is: test paper score is a handwriting digital, what the present invention adopted the identification of handwriting digital is the BP neural network, be also referred to as the multilayer layer sensing network, has three-decker, that is: input layer, hidden layer (also claiming the middle layer) and output layer, as shown in Figure 1.
Circle among the figure is represented neuron, and each neuron between the adjacent layer is realized full the connection, promptly descends each neuron of one deck all to realize being connected entirely with each neuron of last layer, does not connect but have between each neuron in every layer.
Utilize the Multilayer Perception network to carry out pattern-recognition a process must be arranged.This process is carried out in the mode of a kind of " study ".At first each input pattern (being handwriting picture) is set a desired output.Propagate (be called " pattern saequential transmission broadcast ") through the middle layer to output layer then to network input memory pattern, and by input layer.Actual output promptly is error with the difference of desired output.According to minimum this rule of square-error, successively revise the connection weights by output layer toward the middle layer, this process is called " error Back-Propagation ".Along with the alternate repetition of " pattern is saequential transmission broadcast " and " error Back-Propagation " process carries out.The actual output of network approaches to pairing desired output separately gradually, and network also constantly rises to the accuracy of the response of input pattern.By this learning process, determine with preserve each interlayer be connected weights after just can discern the image of input.
The BP Learning Algorithms is the mode of learning of Minimum Mean Square Error.Hypothesis BP network has N processing unit for every layer earlier, and each processing unit is non-linear input/output relation, and the output function of employing is:
f(x)＝1/[1+exp(-x)](1)
Training set comprises M sample mode to (X
k, Y
k) (k=1,2 ..., m), the input summation of P sample unit j is designated as net
Pj, the output note of unit j is made O
Pj, then:
net
pj＝∑W
jiO
pi (2)
O
pj＝1/[1+exp(-net
pj)] (3)
W wherein
JiBe neuron i, the weights of getting in touch between j.
If arbitrarily the network initial weight is set,, between the actual input of network and the desired output certain error is arranged, the define grid error so to each input pattern P:
E
p＝1/2∑(d
pj-O
pi)
2 (4)
E＝∑E
P (5)
D in the formula
PjRepresent P input pattern, the desired output of output unit j.
This basic BP network has very strong information processing capability.
According to above-mentioned inventive concept, the present invention adopts following technical proposals:
The recognition methods of a kind of paper volume face is characterized in that adopting the Multilayer Perception network that digital handwritten form is discerned, and its identification realizes by learning training process and identifying:
(1) learning training process steps:
1. import sample: at first will be imprinted on score chart picture on the paper and become digital signal input computing machine by the analog to digital conversion of image pick-up card;
2. pre-service: the digital picture that shooting is obtained is carried out pre-service such as denoising, tilt adjustments, wide high normalization then;
3. feature extraction: the dot matrix constitutive characteristic value sample that digital picture is taken out it is to (X
k, Y
k), as the input value of BP network.Promptly the value of a pixel on the digital picture all is kept in the array as an eigenwert;
4. BP network training: BP network training process is divided into " pattern is saequential transmission broadcast " " error Back-Propagation ":
(a) forward direction of " pattern is saequential transmission broadcast " tame and docile to be practiced when starting working, weighting parameters such as first initialization " input layer---hidden layer " and " hidden layer---output layer ";
Sample is scanned one by one, the individual digit image in the sample is extracted proper vector, they are transported to input layer, according to the weights W that connects between neuron
JiCalculate net
Pj, O
Pj, obtain the ideal output of this layer; These data obtain the ideal output of hidden layer equally as the input of hidden layer; Pass to output layer from hidden layer again, obtain the result;
(b) " error Back-Propagation " is that the result and the desirable output of output layer are compared, and calculates the weights error on each node of output layer, according to the error on each node of Error Calculation hidden layer on the output layer node; Calculate the error of hidden layer, input layer more respectively, to each interneuronal weights correction;
To add up error in the error Back-Propagation, calculate square error, if the square error value of meeting the expectation, and be no more than maximum cycle and then jump out circulation.If do not reach the error amount of expection, perhaps surpassed maximum cycle index, need to change training parameter.Finish up to training.
5. after training finishes, can under the catalogue at samples pictures place, generate 3 groups of data, store with the form of computer documents respectively.I.e. " win.dat ", " whi.dat ", " num.dat ".The number information of the weights between the weights between in store input layer and the hidden layer, hidden layer and the output layer and each layer node is used for following identification.
(2) identifying step:
1. import digital picture to be identified: gather the picture that comprises handwriting digital to be identified;
2. pre-service: the preprocess method in the same step (1) is the same;
3. feature extraction: in the same step (1) in the study training process feature extracting method the same, the dot matrix constitutive characteristic value sample that digital picture is taken out it is to (X
K, Y
K);
4. BP Network Recognition: the proper vector of extracting is transported to input layer, and the forward direction input treatment channel according to weights information activation pattern is saequential transmission broadcast obtains differentiating the output result on the neuron output node, promptly finish the identification of this handwriting picture.
A kind of test paper score recognition device, adopt above-mentioned test paper score recognition methods to operate utilization, comprise a table top of placing paper, it is characterized in that being provided with on the described table top 3～4 adjustable upright supports and a top panel with the table top arranged parallel; Be equipped with on the described top panel: the test paper score zone of the test paper score of a camera aligning paper, a laser lamp irradiation paper and 1～4 auxiliary light illuminate paper volume face; Described camera is through computer of data acquisition card connection.Camera obtains the image in paper volume face mid-score zone, sends into computer through data collecting card, and computer is handled and mark identification image, and generates form.
The present invention compared with prior art, have following conspicuous outstanding substantive distinguishing features and remarkable advantage: test paper score recognition methods of the present invention is to adopt the Multilayer Perception network that digital handwritten form is discerned, to Flame Image Process and identification, can precise and high efficiency mark on the identification paper.Test paper score recognition device of the present invention, the height of scalable Zhi Lizhu, the focal length of adjusting camera reaches the paper of discerning different big or small patterns, and is simple in structure, is easy to make up, and cost is low, is convenient to operation, can satisfy the practice requirement.
Description of drawings
Fig. 1 is three layers of sensing network illustraton of model.
Fig. 2 is seven sections perspective views of handwritten numeral image.
Fig. 3 is that the handwritten numeral image division becomes to think 3 * 4 sub-piece diagrammatic sketch.
Fig. 4 is the synoptic diagram that the stroke density feature extracts.
Fig. 5 is digit recognition result's a student number part diagrammatic sketch.
Fig. 6 is digit recognition result's a fractional part diagrammatic sketch.
Fig. 7 is an identification student number diagrammatic sketch.
Fig. 8 is an identification mark diagrammatic sketch.
Fig. 9 is the structural representation of a test paper score recognition device of the present invention.
Embodiment
Embodiment 1:
This test paper score recognition methods operation steps is as follows:
1, the pre-treatment step before the sample training:
At first will be imprinted on interesting areas on the paper (ROI) and produce simulating signal, become the digital signal input computing machine of 256 looks again by analog to digital conversion through video frequency collection card.The thin and thick of paper paper, whiteness, smooth finish, writing physical strength, stroke quality and the light and shade of light and even the variation that angle all will cause font when taking produce interference such as stain, the style of calligraphy characterized by hollow strokes, disconnected pen, commissure.Therefore, the character that quantizes that is generally obtained by shooting also needs multiple further processing.
Processing procedure comprise 256 looks change gray-scale map, with gray-scale map binaryzation, edge sharpening and denoising, treat the discriminating digit part the integral inclination adjustment, be partitioned into individual digit, the wide high normalization of standard and tighten the normalized process of resetting.
2, feature extracting method
Good feature extraction scheme is key one ring in the whole recognition system, very strong antijamming capability be arranged to font distortion and change in displacement.Be the 16*32 dimensional feature value of each numeral being got it in the present invention, and as the input value of BP network.When definite eigenwert is followed the example of, the following three phases of process:
Phase one: the extraction of macrofeature.The use of macrofeature can reflect the feature of handwritten numeral image each side comprehensively, improves the performance of Handwritten Digit Recognition System.
The extraction of macrofeature must be followed following principle:
(1) is easy to extract;
(2) have stronger classification capacity, promptly this feature should show bigger difference to different numerals, and identical numeral is then shown as far as possible little difference:
(3) have advantages of higher stability, reduce the influence of stroke fracture and adhesion as far as possible.
After picture specificationization and binaryzation, choose following several macrofeature according to above principle:
(1) 7 segment frames projection value
7 segment frames projection of shape as shown in Figure 2, projecting method be with any point to nearest edge projection, add up the projection of each edge at last and count, so just formed 7 numbers, after the normalization as the proper vector of a numeral.The essence that extracts 7 sections projection properties is Information Compression, and the Information Compression that is about to the m*n dimension becomes the information of 7 degree of freedom.
(2) thick meshed feature
After the handwritten form digital picture is cut out, the handwritten numeral image division is become equal-sized 3*4 sub-piece (as shown in Figure 3), obtain every in black pixel proportion P, constitute a vector x=P1, P2 ..., P12}.So just compression of images has been become the information of 12 dimensions.
Subordinate phase is to the extraction of stroke density feature.The stroke density feature has better anti-jamming capability to font distortion and change in displacement, because the writing style difference, the font of handwritten numeral differs greatly, and adopts the stroke density feature can obtain high recognition as the recognition feature of handwritten numeral.To the standard pane of 1 * W, from level, vertical, 45 °, 135 ° 4 scanning direction numerals, each direction is got n feature, is formed with the proper vector G of 4n component:
G＝(G
11，G
12，…，g
1ng
11，g
12，…g
2n，g
21，g
22，，g
3n，g
41，g
42，…，g
4n)
In order to reduce the number of component in the proper vector,, can be a scan line with several pixel column merger to reduce the node number of BP network input layer.This dimension-reduction treatment can effectively reduce the scale of BP network, improves the real-time of identification.Through experiment, on each direction, get 16 scan lines, form the proper vector G of 16 * 4=64 component.Only need the number of digital black pixel that scan line is scanned to do accumulation calculating for bianry image and can obtain G, for this reason, need all be quantified as 1～16 to the coordinate of 4 directions of pane, function g is used for calculating the number of the black pixel of a pixel column.If P is the pixel sum along certain direction, then the line number of pixel column is Pn=p/16 in scan line; If the row of the initial pixel column of any one scan line number is Pn, then the black pixel number in scan line is, s=1,2,3,4 wherein, t=1,2 ..., 16.
The synoptic diagram (horizontal direction) that Fig. 4 extracts for the stroke density feature.
Phase III, the feature extraction method of full scan method.What the present invention used is, contains digital picture to one by rank scanning, and the value on each pixel is all preserved data[as an eigenwert] in this array, specifically the VC code is as follows:
double**?code(BYTE*?lpDIBBits，int?num，LONG?lLineByte，LONG?lSwidth，LONG
lSheight)
{
// loop variable
int?i，j，k；
BYTE*lpSrc；
The two-dimensional array of proper vector is preserved in // foundation
double**data；
// be this array application two-dimensional storage space
data＝alloc_2d_dbl(num，lSwidth*lSheight)；
// each pixel of normalized sample is come out as a feature point extraction
// data scanning one by one
for(k＝0；k<num；k++)
{
// each data is lined by line scan
for(i＝0；i<lSheight；i++)
{
// each data is pursued column scan
for(j＝k*lSwidth；j<(k+1)*lSwidth；j++)
{
The pointer of the capable j row an of // sensing image i pixel
lpSrc＝lpDIBBits+i*lLineByte+j；
If // this pixel is a black
if(*(lpSrc)＝＝0)
// relevant position of proper vector is filled out 1
data[k][i*lSwidth+j-k*lSwidth]＝1；
If // this pixel is other
if(*(lpSrc)！＝0)
// relevant position of proper vector is filled out 0
data[k][i*lSwidth+j-k*lSwidth]＝0；
}
}
}
return(data)；
}
3, utilize BP algorithm training feedforward network, make network finish that function is forced and pattern-recognition.
Weights and some parameters of first initialization input layer-hidden layer and hidden layer-output layer before the training:
bpnn_randomize_weights(input_weights，n_in，n_hidden)；
bpnn_randomize_weights(hidden_weights，n_hidden，n_out)；
Corresponding respectively is input layer-hidden layer weights, hidden layer-output layer weights at random in interval (0.1,0.1) interior assignment.
double?momentum＝BpPa.m_a；
double?min_ex＝BpPa.m_ex；
int?n_hidden＝BpPa.m_hn；
double?eta＝BpPa.m_eta；
These four variablees are followed successively by related coefficient, least mean-square error, hidden layer node number and learning efficiency and compose initial value.Be such value in the present invention:
BpPa.m_a＝0；
BpPa.m_eta＝0.015；
BpPa.m_ex＝0.001；
BpPa.m_hn＝10；
Numeral with one group of 0-9 is that example begins to train below.
Sample is scanned one by one, the individual digit in the sample is extracted proper vector, it is transported to input layer, predetermined ideal is transported to the desirable output unit of BP network.
The desirable output matrix here is as follows:
double?out[][4]＝{0.1，0.1，0.1，0.1，
0.1，0.1，0.1，0.9，
0.1，0.1，0.9，0.1，
0.1，0.1，0.9，0.9，
0.1，0.9，0.1，0.1，
0.1，0.9，0.1，0.9，
0.1，0.9，0.9，0.1，
0.1，0.9，0.9，0.9，
0.9，0.1，0.1，0.1，
0.9，0.1，0.1，0.9}；
10 respectively corresponding 10 arabic numeral of row " 0 " arrive " 9 ".
Fl transmission is started working, data are passed to hidden layer from input layer, pass to output layer from hidden layer again, with the output and the desirable error of relatively calculating on each node of output layer of exporting of output layer, according to the error on each node of Error Calculation hidden layer on the output layer node.
Adjust weights then respectively, adjust weights between hidden layer and the output layer, adjust weights between hidden layer and the input layer according to the error on each node of hidden layer according to the error on each node of output layer.
Error is added up, calculate square error, if the square error value of meeting the expectation, and be no more than maximum cycle (being made as 15000 here) and then jump out circulation.
If do not reach the error amount of expection, perhaps surpassed maximum cycle index (15000), need to change training parameter.Finish up to training.
After once training finishes, can under the catalogue at samples pictures place, generate 3 files, be respectively " win.dat ", " whi.dat ", " num.dat ", the weights between weights, hidden layer and the output layer between the inside in store input layer of difference and the hidden layer and the number information of each layer node are used for the identification work of back.
4, the process of digit recognition and effect
Through the pre-service and the sample training work in early stage, found proper weighted value, just can carry out subsequent identification work.
At first, system can read a picture to be identified, indicates former figure, student number part, fractional part respectively with three handle CDIB, CDIB1, CDIB2 and CDIB3.
After student number part and the fractional part amplification respectively as Fig. 7, shown in Figure 8.
The weights file of preserving when at first reading training when beginning to discern, then numeral is scanned one by one, the proper vector of extracting is transported to input layer, import according to weights information activation forward direction, on each output node, differentiate the output result, just on this carry-out bit, put 1 greater than 0.5, have ten output nodes altogether, if the result who judges is less than " 9 ", think that identification rationally, if the result who judges then thinks identification error irrational result to be fixed as particular value 20 greater than " 9 ".
At last, the result of identification is saved in and discerns among " result.txt " under the same catalogue of picture,, be shown as " can't judge " to being decided to be the output of particular value 20.
Fig. 5 and Fig. 6 are carried out the identification of student number and mark, the result of identification such as Fig. 7 and Fig. 8 respectively.
Embodiment 2:
Referring to Fig. 9, this test paper score recognition device adopts above-mentioned test paper score recognition methods to manipulate, and comprises a table top 8 of placing paper 9,4 adjustable columns 5 is set on the table top 8 is supporting a top panel 1 with table top 8 arranged parallel; Be equipped with on the top panel 1: test paper score zone and 4 auxiliary lights of the test paper score of camera 2 aligning papers 9, laser lamp 3 irradiation papers 9 illuminate paper 9 volume faces; Camera 2 connects a computer 7 through data collecting card 6.Camera 2 obtains the image in paper volume face 9 mid-score zones, sends into computer 7 through data collecting card 6, and 7 pairs of images of computer are handled and mark identification, and generate form.
The operation steps of this device is as follows:
(1) paper to be identified 9 is placed on the table top 8, regulates adjustable column 5, change the distance between top panel 1 and the table top 8, make camera 2 obtain best focal length.
(2) take paper volume face image,, image is sent into computer 7 through data collecting card 6.
(3) utilize computer, the paper volume face image that obtains is carried out correction process.
(4) the handwritten form mark in the paper volume face image after proofreading and correct is discerned.
(5) numeral that identifies is sent to database automatically, after task to be identified is finished, does respective handling and generate form.
Claims (2)
1. test paper score recognition methods is characterized in that adopting the Multilayer Perception network that digital handwritten form is discerned, and its identification realizes by learning training process and identifying:
(1) learning training process steps:
1. import sample: at first will be imprinted on score chart picture on the paper and become digital signal input computing machine by the analog to digital conversion of image pick-up card;
Pre-service: the digital picture that shooting is obtained is carried out the adjustment and the wide high normalization pre-service of noise, inclination;
3. feature extraction: the dot matrix constitutive characteristic value sample that digital picture is taken out it to (YK), as the input value of BP network, promptly there is the value of a pixel on the digital picture in the array in XK as an eigenwert, k=1 wherein, 2 ..., m;
4. BP network training: BP network training process is divided into pattern and saequential transmission broadcasts and error Back-Propagation:
(a) pattern is saequential transmission broadcast: first initialization " input layer----hidden layer " and " hidden layer----output layer " weighting parameter; Then sample is scanned one by one, the single power word image in the sample is extracted proper vector, they are transported to input layer, according to the weights W that connects between neuron
JiCalculate net
Pj, O
Pj, net
PjBe the input summation to P sample unit j, O
PjBe the output of unit j, obtain the ideal output of this layer; These data obtain latent ideal output of reading layer equally as the input of hidden layer; Pass to output layer from the latent layer of reading again, obtain the result;
(b) error Back-Propagation: the result and the desirable output of output layer are compared, calculate the weights error on each node of output layer, according to the error on each node of Error Calculation hidden layer on the output layer node; Calculate the latent error of reading layer, input layer more respectively, to each interneuronal weights correction;
To add up error in the error Back-Propagation, calculate square error, if the square error value of meeting the expectation, and be no more than maximum cycle and then jump out circulation; If do not reach the error amount of expection, perhaps surpassed maximum cycle index, need to change training parameter, finish up to training;
5. after training finishes, the present generates 3 groups of data under the catalogue at samples pictures place, store with the form of computer documents respectively, i.e. " win.dat ", " whi.dat " and " num.dat ", the weights between in store input layer and latent weights, hidden layer and the output layer of reading between the layer and the number information of each layer node are used for following identification;
(2) identifying step:
1. import digital picture to be identified: gather the picture that comprises handwriting digital to be identified;
2. pre-service: the preprocess method in the same step (1) is the same;
3. feature extraction: in the same step (1) in the study training process feature extracting method the same, the dot matrix constitutive characteristic value sample that digital picture is taken out it is to (X
K, Y
K);
4. BP Network Recognition: the proper vector of extracting is transported to input layer, and the forward direction input treatment channel according to weights information activation pattern is saequential transmission broadcast obtains differentiating the output result on the neuron output node, promptly finish the identification of this handwriting picture.
2. test paper score recognition device, adopt test paper score recognition methods according to claim 1 to operate utilization, comprise a table top (8) of placing paper (9), it is characterized in that being provided with on the described table top (8) 3～4 adjustable columns (5) and supporting a top panel (1) with table top (8) arranged parallel; Described top panel is equipped with on (1): a camera (2) is aimed at the test paper score of paper (9), test paper score zone and 1～4 auxiliary light (4) of a laser lamp (3) irradiation paper (9) illuminates paper volume face; Described camera (2) connects a computer (7) through data collecting card (6); Camera (2) obtains the image in paper volume face (9) mid-score zone, sends into computer (7) through data collecting card (6), and computer (7) is handled and mark identification image, and generates form.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB200710039969XA CN100485714C (en) | 2007-04-25 | 2007-04-25 | Method and device for recognizing test paper score |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB200710039969XA CN100485714C (en) | 2007-04-25 | 2007-04-25 | Method and device for recognizing test paper score |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101038626A CN101038626A (en) | 2007-09-19 |
| CN100485714C true CN100485714C (en) | 2009-05-06 |
Family
ID=38889515
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB200710039969XA Expired - Fee Related CN100485714C (en) | 2007-04-25 | 2007-04-25 | Method and device for recognizing test paper score |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100485714C (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009193429A (en) * | 2008-02-15 | 2009-08-27 | Mitsubishi Electric Corp | Image reading device |
| CN102974089A (en) * | 2012-11-09 | 2013-03-20 | 沈豪杰 | Competition quickly correcting scoring system |
| CN105869085A (en) * | 2016-03-29 | 2016-08-17 | 河北师范大学 | Transcript inputting system and method for processing images |
| CN106778752A (en) * | 2016-11-16 | 2017-05-31 | 广西大学 | A kind of character recognition method |
| CN106503711A (en) * | 2016-11-16 | 2017-03-15 | 广西大学 | A kind of character recognition method |
| CN106503712A (en) * | 2016-11-16 | 2017-03-15 | 广西大学 | One kind is based on stroke density feature character recognition method |
| CN106874911A (en) * | 2017-03-03 | 2017-06-20 | 沈阳工程学院 | The area ratio that a kind of application partitioning is obtained is come the method that carries out printing digit recognizing |
| CN107038438A (en) * | 2017-03-16 | 2017-08-11 | 上海电机学院 | It is a kind of that method is read and appraised based on image recognition |
| CN107016417A (en) * | 2017-03-28 | 2017-08-04 | 青岛伟东云教育集团有限公司 | A kind of method and device of character recognition |
| CN107292308A (en) * | 2017-06-08 | 2017-10-24 | 东华大学 | A kind of intelligence system separating device |
| CN108959664A (en) * | 2018-09-26 | 2018-12-07 | 江苏曲速教育科技有限公司 | Distributed file system based on picture processor |
| CN111860317A (en) * | 2020-07-20 | 2020-10-30 | 青岛特利尔环保集团股份有限公司 | Boiler operation data acquisition method, system, equipment and computer medium |
| CN114881541B (en) * | 2022-07-06 | 2022-11-11 | 成都西交智汇大数据科技有限公司 | Evaluation method, device and equipment for examinee score and readable storage medium |
-
2007
- 2007-04-25 CN CNB200710039969XA patent/CN100485714C/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN101038626A (en) | 2007-09-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100485714C (en) | Method and device for recognizing test paper score | |
| CN108647681B (en) | A kind of English text detection method with text orientation correction | |
| Diem et al. | ICDAR 2013 competition on handwritten digit recognition (HDRC 2013) | |
| CN107609575B (en) | Calligraphy evaluation method, calligraphy evaluation device and electronic equipment | |
| CN112633382B (en) | Method and system for classifying few sample images based on mutual neighbor | |
| CN104463209A (en) | Method for recognizing digital code on PCB based on BP neural network | |
| CN103761531A (en) | Sparse-coding license plate character recognition method based on shape and contour features | |
| CN109241808A (en) | Two dimensional code auto-screening method in a kind of fashion images | |
| CN111340810B (en) | Intelligent evaluation method for Chinese character writing quality | |
| CN109522883A (en) | A kind of method for detecting human face, system, device and storage medium | |
| CN110276881A (en) | A kind of banknote serial number recognition methods based on convolution loop neural network | |
| CN110163117B (en) | Pedestrian re-identification method based on self-excitation discriminant feature learning | |
| CN106485651A (en) | The image matching method of fast robust Scale invariant | |
| CN109886082A (en) | A kind of Small object enhancing prediction module detection method based on SSD | |
| CN110163567A (en) | Classroom roll calling system based on multitask concatenated convolutional neural network | |
| CN110929746A (en) | Electronic file title positioning, extracting and classifying method based on deep neural network | |
| CN107085731A (en) | A kind of image classification method based on RGB D fusion features and sparse coding | |
| CN106778717A (en) | A kind of test and appraisal table recognition methods based on image recognition and k nearest neighbor | |
| CN109165698A (en) | A kind of image classification recognition methods and its storage medium towards wisdom traffic | |
| CN110458145A (en) | A kind of offline person's handwriting Individual Identification System and method based on two-dimentional behavioral characteristics | |
| CN110490153B (en) | Offline handwriting individual recognition system and method based on three-dimensional dynamic characteristics | |
| CN109284702B (en) | Answer sheet scoring and marking system based on image mode | |
| CN113220915B (en) | Remote sensing image retrieval method and device based on residual attention | |
| CN110490915A (en) | A kind of point cloud registration method being limited Boltzmann machine based on convolution | |
| CN110503090A (en) | Character machining network training method, character detection method and character machining device based on limited attention model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20090506 Termination date: 20120425 |