CN105447489A - Character and background adhesion noise elimination method for image OCR system - Google Patents

Abstract

The invention discloses a character and background adhesion noise elimination method for an image OCR system, comprising the following steps: (1) setting the effective range of the stroke length of a character to be identified; (2) selecting an image containing the character to be identified in a natural environment, and calculating the difference boundary diagram of the image by means of mean difference template; (3) thresholding the obtained difference boundary diagram of the image to be identified to form a ternary-valued boundary diagram of the image to be identified; and (4) carrying out searching in the ternary-valued boundary diagram according to the angle-of-inclination range and the linear boundary. By using the character and background adhesion noise elimination method for the image OCR system of the invention, the problem that a character with adhesion noise is hard to position for the existing OCR application system is solved in a complex and changeable natural environment, the character position can be extracted accurately, and the real-time requirements of engineering application can be satisfied.

Description

A kind of character of picture OCR recognition system and background adhesion noise cancellation method

Technical field

The present invention relates to natural scene picture OCR recognition system, particularly relate to a kind of character and background adhesion noise cancellation method of picture OCR recognition system.

Background technology

At artificial intelligence field, optical character identification OCR is a very important technology, due to the classification of mass picture and the needs of search in the universal of smart mobile phone and cloud storage, make natural scene picture OCR identify the focus becoming Recent study, OCR technology generally comprises character locating, Character segmentation, several process such as character recognition, the wherein accuracy and runtime of character locating, directly affects the quality of OCR recognition technology, is the key of whole OCR system.

OCR character locating utilizes the structural information of character string usually, by the mode of global search, character string region is positioned, and the extraction of character string structural information, the most frequently used is also be the edge being extracted character string picture by binaryzation technology compared with effective method, conventional binarization method has fixed threshold method, Adaptive Thresholding, Global thresholding and local thresholding method etc., no matter any algorithm, when in the face of environment complicated and changeable, as Various Seasonal, the complex situations such as different weather environment, the various noise of capital introducing more or less, these noises will be easy to cause locating unsuccessfully, or produce a large amount of false character information, thus the calculated amount of subsequent treatment is increased greatly.

To the noise that binaryzation is introduced, existing way normally adopts some filtering methods to remove filtering noise, such as medium filtering, mathematical morphology, two dimensional wavelet analysis etc., these diverse ways have different effects to different images, although conventional linear low-pass filters and the method for neighborhood averaging can remove partial noise, but they have image blurring negative interaction, the method of medium filtering can eliminate isolated noise spot, and the Fuzzy comparisons produced is few, but it removes the effect of noise to bianry image and bad, part black patch can erode by mathematical morphology to a certain extent, but so often cause original image to be out of shape aggravation, be unfavorable for follow-up calculating.

In addition, in natural scene picture OCR identifies, due to illumination, weather, the impact of the factors such as foreign material, make binaryzation noise more, in practical engineering application, independently, discrete noise is often distinguished than being easier to, and be often difficult to process with the noise that character or object to be identified there occurs adhesion, the result that adhesion noise brings is exactly that corresponding character position scope receives interference, the accurate of character is caused to locate unsuccessfully, thus have impact on the overall discrimination of system, a key issue during the Characters Stuck noise how eliminated in natural scene has become OCR to identify, therefore, a kind of practical application needs that can meet are proposed, the real-time Characters Stuck noise cancellation technique realized under various complex environment seems very important, by eliminating adhesion noise, the interference in non-character region can be reduced, thus the accuracy of OCR character locating is provided.

Summary of the invention

The object of the present invention is to provide a kind of character and background adhesion noise cancellation method of picture OCR recognition system, for the OCR technology of existing natural scene picture, the deficiency existed in noise processed, propose a kind of new adhesion noise cancellation method, in physical environment complicated and changeable, this method solve existing OCR application system to the problem with adhesion noise character location difficulty, realize the accurate extraction of character position, and the requirement of real-time of engineer applied can be met.

In order to achieve the above object, technical scheme provided by the invention is: character and the background adhesion noise cancellation method of described natural scene picture OCR recognition system comprise the following step performed in order:

The character of picture OCR recognition system and a background adhesion noise cancellation method, character and the background adhesion noise cancellation method of described picture OCR recognition system comprise as follows:

1) priori of the engineering-environment applied according to OCR system, arranges the effective range of the stroke length StrokeLen of character to be identified;

2) choose the image containing character to be identified under physical environment, and with the mode computed image difference boundary graph of mean difference template, described difference boundary graph comprises character zone and background area;

3) the difference boundary graph of image to be identified obtained above is carried out thresholding, form the three-valued boundary graph of image to be identified, described three-valued boundary graph comprises region to be identified and background area;

4) in the three-valued boundary graph of image to be identified obtained above, according in range of tilt angles, linear boundary searches for, check this border whether beyond the effective range of stroke length, when not exceeding the effective range of stroke length, then think effective frontier point, and retain, when exceeding the effective range of stroke length, then this edge determination is noise, it is removed from three value boundary images.

Described step 3) in, by as follows for the formula that the difference boundary graph of described image to be identified carries out thresholding: $Threshold\left(P\right(x,y\left)\right)=\left\{\begin{array}{cc}c1,& ifP(x,y)<ParamA\\ c2,& ifP(x,y)>ParamB\\ c3,& Others\end{array}\right.$

Wherein, P (x, y) is the difference value of pixel (x, y) in difference diagram, ParamA and ParamB is the threshold value pre-set, and c1, c2 and c3 are signless integers, and span is [0,255]; Wherein c1 represents border brighter around the brightness ratio of current pixel point, and c2 represents the border that the brightness ratio of current pixel point is darker comparatively around, and c3 represents the value on non-border.

Described c1 color is black, and value is c1=0; Described c2 color is white, value is c2=255, described c3 color is grey, value is c3=128, the value of difference boundary pixel point is respectively 0,255, but not the value of boundary pixel point is 128, the pixel of namely described black and white is boundary pixel point, and the pixel of described grey is non-boundary pixel point.

Described step 4) in, by the three-valued image of described image to be identified according in range of tilt angles, the linear boundary algorithm steps that carries out searching for is as follows:

4.1) calculate the straight line declining displacement of angle of inclination in [α, β] scope and detect masterplate;

4.2) to each frontier point in three-valued image, the straight-line detection masterplate obtained in utilizing step a), checks from this frontier point, boundary value identical, and with the continuum boundary line of stencil matching;

4.3) to the continuum boundary line of coupling, checking whether its length belongs to effective range, when belonging to effective range, then retaining, otherwise this boundary line is noise margin, and the border pixel values that frontier points all on boundary line comprise is set to c3.

Described step 4-1) in, the algorithm steps of the straight-line detection masterplate of described calculating angle of inclination in [α, β] scope is as follows:

4.1.1) according to step 1) set by the longest stroke length StrokeLen, according to length Len and this masterplate maximum deviation value MaxOffY in y-direction of engineering practice calculated line masterplate quantity Num, straight line masterplate lines, computing formula is as follows respectively;

$\left\{\begin{array}{c}Num=4StrokeLen\\ Len=2StrokeLen\\ MaxOffY=Len\end{array}\right.$

4.1.2) to each straight line declining displacement masterplate, the off-set value in y direction is calculated as follows.

$y=\frac{x*OffY}{Len},x\&Element;\&lsqb;0,Len\&rsqb;,OffY\&Element;\&lsqb;-MaxoffY,MaxOffY\&rsqb;$

Described step 4.2) in the three-valued image of image to be identified carried out the searching algorithm of linear boundary, judge from certain frontier point, boundary value identical, and as follows with the method for the continuum boundary line of straight line declining displacement stencil matching:

4.2.1) when the live width of described straight line is the generic linear of 1, the method, from starting point, from left to right checks whether the equivalent boundary of direct neighbor, works as existence, then on this straight line, all frontier points all belong to match point, otherwise just belong to not match point;

4.2.2) when described straight line live width is N, based on the general line of eight connectivity, the method is from starting point, detect the equivalent boundary under eight connectivity meaning, when there is such general line, then on this general line, all frontier points all belong to match point, otherwise just belong to not match point.

The invention has the beneficial effects as follows: the character of a kind of picture OCR recognition system of the present invention and background adhesion noise cancellation method, for the OCR technology of existing natural scene picture, the deficiency existed in noise processed, propose a kind of new adhesion noise cancellation method, in physical environment complicated and changeable, this method solve existing OCR application system to the problem with adhesion noise character location difficulty, realize the accurate extraction of character position, and the requirement of real-time of engineer applied can be met.

Accompanying drawing explanation

Fig. 1 is a kind of character of picture OCR recognition system and the process flow diagram of background adhesion noise cancellation method of the embodiment of the present invention;

Fig. 2 is the mean difference template of the embodiment of the present invention;

Fig. 3 is a kind of schematic diagram scanning difference border general line of the embodiment of the present invention, and the rectangular block of figure empty represents non-border pixel point, and in figure, non-blank-white rectangular block represents boundary pixel point;

Fig. 4 is the process flow diagram that the embodiment of the present invention carries out described three-valued image searching on linear pattern general line border;

Fig. 5 is that the present invention calculates the straight-line detection masterplate process flow diagram of angle of inclination in [α, β] scope;

Fig. 6 is the generic linear schematic diagram be made up of difference boundary pixel in the embodiment of the present invention;

Fig. 7 to be the present invention be use during general line judges, based on the neighbor schematic diagram under eight connectivity meaning;

Fig. 8 to be width of the present invention be 2 general line schematic diagram;

Fig. 9 to be width of the present invention be 3 general line schematic diagram;

Figure 10 is that the embodiment of the present invention is applied to the 1st and has the car plate of adhesion picture before treatment;

Figure 11 is the design sketch after the embodiment of the present invention is applied to the 1st car plate process;

Figure 12 is that the embodiment of the present invention is applied to the 2nd and has the car plate of adhesion picture before treatment;

Figure 13 is the design sketch after the embodiment of the present invention is applied to the 2nd car plate process;

Figure 14 is that the embodiment of the present invention is applied to the 3rd and has the car plate of adhesion picture before treatment;

Figure 15 is the design sketch after the embodiment of the present invention is applied to the 3rd car plate process;

Figure 16 is that the embodiment of the present invention is applied to the 4th and has the car plate of adhesion picture before treatment;

Figure 17 is the design sketch after the embodiment of the present invention is applied to the 4th car plate process;

Figure 18 is that the embodiment of the present invention is applied to the 5th and has the car plate of adhesion picture before treatment;

Figure 19 is the design sketch after the embodiment of the present invention is applied to the 5th car plate process.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described in further detail.

As shown in Figure 1, be the character of a kind of picture OCR of embodiment of the present invention recognition system and the process flow diagram of background adhesion noise cancellation method, described method comprises:

1) priori of the engineering-environment applied according to OCR system, arranges the effective range of the stroke length StrokeLen of character to be identified.This parameter has close relationship with actual engineering, will be set to configurable parameter in actual application.When the OCR that the method is applied to bayonet socket license plate image identifies, this Selecting parameter is 35 pixels.In the OCR identifying of traffic electronic police picture, this Selecting parameter is 30 pixels.

2) choose the image containing character to be identified under physical environment, and with the mode computed image difference boundary graph of mean difference template as shown in Figure 2, described difference boundary graph comprises character zone and background area.

3) the difference boundary graph of described image to be identified is carried out thresholding, form the three-valued boundary graph of image to be identified, described image to be identified comprises region to be identified and background area, carries out three-valued by the difference diagram of described image to be identified according to following formula:

$Threshold\left(P\right(x\&CenterDot;y\left)\right)=\left\{\begin{array}{cc}c1,& ifP(x\&CenterDot;y)<ParamA\\ c2,& ifP(x,y)>ParamB\\ c3,& Others\end{array}\right.$

Wherein, P (x, y) be the difference value of pixel (x, y) in difference diagram, ParamA and ParamB is the threshold value pre-set, c1, the span of c2 and c3 is [0,255], and wherein c1 represents border brighter around the brightness ratio of current pixel point, c2 represents the border that the brightness ratio of current pixel point is darker comparatively around, and c3 represents the value on non-border.In engineering, conveniently debug, the value of three constants should have larger otherness as far as possible, and described c1 color is black, and value is c1=0; Described c2 color is white, value is c2=255, described c3 color is grey, value is c3=128, the value of difference boundary pixel point is respectively 0,255, but not the value of boundary pixel point is 128, the pixel of namely described black and white is boundary pixel point, and the pixel of described grey is non-boundary pixel point.

4) in the three-valued boundary graph of image to be identified obtained above, according in range of tilt angles, linear boundary is searched for, check this border whether beyond the effective range of stroke length, if do not exceed the effective range of stroke length, then think effective frontier point and retain, if exceed the effective range of stroke length, then this edge determination is noise margin, it is removed from three value boundary images, as shown in Figure 3, wherein indicate three testing processes, from difference boundary pixel point A respectively, C and E sets out, have detected the linear boundary in Num direction, in figure 3, there are two linear pattern difference borders, AB and CD respectively, E is then an isolated frontier point, if AB and CD exceeds the effective range of stroke length, just its color value is set to non-boundary value c2, so just remove this linear boundary.

Difference diagram after above-mentioned steps process, is just a cancellation the image of adhesion noise, now just can apply it in the follow-up flow process of OCR character recognition.

In above process, the search of linear boundary is even more important, and as shown in Figure 4, for the present invention to carry out the process flow diagram of linear boundary search to three-valued image, described flow process is as follows:

Step 4.1) calculate angle of inclination at [α, β] straight line declining displacement in scope detects masterplate, here in fact two involved parameters illustrate the range of tilt angles of OCR character, minimum cant is α, maximum tilt angle is β, their selection has important impact to the elimination effect of noise and performance, configurable parameter to be set in actual application, when the OCR that the method is applied to bayonet socket license plate image identifies, Selecting parameter can be α=-15 °, β=15 °, when the OCR being applied to traffic electronic police picture identifies, proposed parameter is chosen as α=-25 °, β=25 °, when being applied to traffic and disobeying the OCR identification stopping evidence obtaining picture, because separated stop board angle of inclination is often larger, parameter recommendation is chosen as α=-45 °, β=45 °,

Step 4.2) to each frontier point in three-valued image, utilize step 4.1) described in straight-line detection masterplate, check from this frontier point, boundary value identical, and with the continuum boundary line of stencil matching;

Step 4.3) to the continuum boundary line mated, check whether its length belongs to effective range.If belong to effective range, then retain, otherwise this boundary line is noise margin, and the border pixel values that frontier points all on boundary line comprise is set to c3, the length effective range of continuum boundary line described here, refers to the effective range of stroke length.

Masterplate is detected to aforesaid rectilinear, its objective is to accelerate the speed of searching for, whole algorithm is made to meet requirement to real-time in practical application, straight-line detection masterplate has several, each masterplate is for answering some specific pitch angle, wherein preservation information be side-play amount in angled straight lines between each y coordinate and horizontal line, Fig. 5 is the process flow diagram of line detection algorithm, and concrete operations are as follows:

Step 4.1.1) according to described step 1) in set the longest stroke length StrokeLen, according to engineering practice calculated line masterplate quantity Num, the length Len of straight line masterplate lines and the maximum deviation value MaxOffY of y.Computing formula is as follows:

$\left\{\begin{array}{c}Num=4StrokeLen\\ Len=2StrokeLen\\ MaxOffY=Len\end{array}\right.;$

Step 4.1.2) to each straight line declining displacement masterplate, be calculated as follows the off-set value in y direction.

$y=\frac{x*OffY}{Len},x\&Element;\&lsqb;0,Len\&rsqb;,OffY\&Element;\&lsqb;-MaxOffY,MaxOffY\&rsqb;$

The search on three-valued image cathetus type border is the key of this method, to each difference frontier point, from this point, coupling is gone with straight-line detection masterplate, check whether the general line type border that there is continuous difference boundary pixel point and form, general line is detected, as follows:

4.2.1) when the live width of described straight line is the generic linear of 1, the method is from starting point, from left to right check whether the equivalent boundary of direct neighbor, if existed, then this frontier point belongs to match point, otherwise just belongs to not match point, as shown in Figure 6, under the definition of generic linear, having two boundary straight line in this figure, is AB and CD respectively.

4.2.2) when the live width of described straight line be N, general line (as shown in Figure 7) based on eight connectivity, the method is from starting point, detect the equivalent difference frontier point within the scope of eight connectivity, if there is such frontier point, then this frontier point belongs to match point, otherwise just belongs to not match point.The implication of eight connectivity described here as shown in Figure 7, if two difference frontier points meet the position relationship shown in it, these two difference frontier points are then become to be that eight connectivity is adjacent, according to this definition, as Fig. 8, shown in 9, wherein just give the example of two general lines, in fig. 8, general line AB has extended to B from starting point A, width is 2, horizontal length is 13, in fig .9, general line CD has extended to D from starting point C, width is 3, horizontal length is 16, the general line defining out in this manner has very high matching degree to the adhesion noise in engineering, such general line is disposed from difference boundary graph, greatly can reduce noise, the particularly impact of adhesion noise, improve the precision of OCR character locating, in the application of Practical Project, Parameter N is less than 4 usually.

The embodiment of the present invention proposes a kind of character and background adhesion noise cancellation method of picture OCR recognition system, according to engineering practice, after setting corresponding parameter, then utilize mean difference masterplate to calculate and obtain the three-valued difference diagram of image, finally, in all difference borders, search meets the general line of correlated condition, this straight line is exactly interference noise, by eliminating the relevant difference frontier point of this general line, noise information in picture to be identified is suppressed greatly, and finally make candidate characters region be located accurately, thus improve the final accuracy of OCR character recognition.

Shown in Figure 10-19, it is the design sketch that the present invention is applied in Car license recognition, there is shown the license plate image that 5 have adhesion, as shown in Figure 10, the reason of four character application adhesion noises in the middle of car plate is joined together, if when traditionally connected component carries out License Plate, the connected component that these four characters are formed will be considered to an entirety, and after applying the present invention, as shown in figure 11, these four characters have been fully segmented out, thus be conducive to the location of car plate, as Figure 12, 14, 16, described in 18, these car plates are also similar effects, namely after process of the present invention, as Figure 13, 15, 17, shown in 19, effectively adhesion noise can both be eliminated, make the dividing processing that character wherein can be clear and definite, thus improve the discrimination of system.

The character of a kind of picture OCR recognition system of the present embodiment and background adhesion noise cancellation method, for the OCR technology of existing natural scene picture, the deficiency existed in noise processed, propose a kind of new adhesion noise cancellation method, in physical environment complicated and changeable, this method solve existing OCR application system to the problem with adhesion noise character location difficulty, realize the accurate extraction of character position, and the requirement of real-time of engineer applied can be met.

One of ordinary skill in the art of the present invention are appreciated that; the above embodiment of the present invention is only one of the preferred embodiments of the present invention; for length restriction; here can not all embodiments of particularize; any enforcement that can embody the claims in the present invention technical scheme, all in protection scope of the present invention.

It should be noted that; above content is in conjunction with concrete embodiment further description made for the present invention; can not assert that the specific embodiment of the present invention is only limitted to this; under above-mentioned guidance of the present invention; those skilled in the art can carry out various improvement and distortion on the basis of above-described embodiment, and these improve or distortion drops in protection scope of the present invention.

Claims (6)

1. the character of picture OCR recognition system and a background adhesion noise cancellation method, is characterized in that: character and the background adhesion noise cancellation method of described picture OCR recognition system comprise as follows:

1) priori of the engineering-environment applied according to OCR system, arranges the effective range of the stroke length StrokeLen of character to be identified;

2) choose the image containing character to be identified under physical environment, and with the mode computed image difference boundary graph of mean difference template, described difference boundary graph comprises character zone and background area;

3) the difference boundary graph of image to be identified obtained above is carried out thresholding, form the three-valued boundary graph of image to be identified, described three-valued boundary graph comprises region to be identified and background area;

4) in the three-valued boundary graph of image to be identified obtained above, according in range of tilt angles, linear boundary searches for, check this border whether beyond the effective range of stroke length, when not exceeding the effective range of stroke length, then think effective frontier point, and retain, when exceeding the effective range of stroke length, then this edge determination is noise, it is removed from three value boundary images.

2. the character of a kind of picture OCR recognition system according to claim 1 and background adhesion noise cancellation method, is characterized in that: described step 3) in, by as follows for the formula that the difference boundary graph of described image to be identified carries out thresholding:

$Threshold\left(P\right(x,y\left)\right)=\left\{\begin{array}{cc}c1,& ifP(x,y)<ParamA\\ c2,& ifP(x,y)>ParamB\\ c3,& Others\end{array}\right.$

Wherein, P (x, y) is the difference value of pixel (x, y) in difference diagram, ParamA and ParamB is the threshold value pre-set, c1, c2 and c3 are signless integers, and span is [0,255], wherein c1 represents border brighter around the brightness ratio of current pixel point, and c2 represents the border that the brightness ratio of current pixel point is darker comparatively around, and c3 represents the value on non-border.

3. the character of a kind of picture OCR recognition system according to claim 2 and background adhesion noise cancellation method, it is characterized in that: described c1 color is black, value is c1=0; Described c2 color is white, value is c2=255, described c3 color is grey, value is c3=128, the value of difference boundary pixel point is respectively 0,255, but not the value of boundary pixel point is 128, the pixel of namely described black and white is boundary pixel point, and the pixel of described grey is non-boundary pixel point.

4. the character of a kind of picture OCR recognition system according to claim 1 and background adhesion noise cancellation method, it is characterized in that: described step 4) in, by the three-valued image of described image to be identified according in range of tilt angles, the linear boundary algorithm steps that carries out searching for is as follows:

4.1) calculate the straight line declining displacement of angle of inclination in [α, β] scope and detect masterplate;

4.2) to each frontier point in three-valued image, the straight-line detection masterplate obtained in utilizing step a), checks from this frontier point, boundary value identical, and with the continuum boundary line of stencil matching;

4.3) to the continuum boundary line of coupling, checking whether its length belongs to effective range, when belonging to effective range, then retaining, otherwise this boundary line is noise margin, and the border pixel values that frontier points all on boundary line comprise is set to c3.

5. the character of a kind of picture OCR recognition system according to claim 3 and background adhesion noise cancellation method, it is characterized in that: described step 4-1) in, the algorithm steps of the straight-line detection masterplate of described calculating angle of inclination in [α, β] scope is as follows:

4.1.1) according to described step 1) in set the longest stroke length StrokeLen, according to engineering practice calculated line masterplate quantity Num, the length Len of straight line masterplate lines and the maximum deviation value MaxOffY of y.Computing formula is as follows:

$\left\{\begin{array}{c}Num=4StrokeLen\\ Len=2StrokeLen\\ MaxOffY=Len\end{array}\right.;$

4.1.2) to each straight line declining displacement masterplate, the off-set value in y direction is calculated as follows:

$y=\frac{x*offY}{Len},x\&lsqb;0,Len\&rsqb;,OffY\&Element;\&lsqb;-MaxOffY,MaxOffY\&rsqb;.$

6. the character of a kind of picture OCR recognition system according to claim 3 and background adhesion noise cancellation method, it is characterized in that: described step 4.2) in the three-valued image of image to be identified carried out the searching algorithm of linear boundary, judge from certain frontier point, boundary value identical, and as follows with the method for the continuum boundary line of straight line declining displacement stencil matching:

4.2.1) when the live width of described straight line is the generic linear of 1, the method, from starting point, from left to right checks whether the equivalent boundary of direct neighbor, works as existence, then on this straight line, all frontier points all belong to match point, otherwise just belong to not match point;

4.2.2) when described straight line live width is N, based on the general line of eight connectivity, the method is from starting point, detect the equivalent boundary under eight connectivity meaning, when there is such general line, then on this general line, all frontier points all belong to match point, otherwise just belong to not match point.