CN103034856B - The method of character area and device in positioning image - Google Patents

Abstract

The present invention relates to the method for character area in a kind of positioning image, comprising: obtain original image; By Roberts operator, conversion is carried out to described original image and generate gray level image; Described Binary Sketch of Grey Scale Image is obtained edge image; Described edge image is expanded, extracts the positional information of the connected domain of described edge image.In addition, the device of character area in a kind of positioning image is also comprised.In above-mentioned positioning image, the method for character area and device can improve the accuracy of location.

Description

The method of character area and device in positioning image

Technical field

The present invention relates to image processing field, particularly relate to method and the device of character area in a kind of positioning image.

Background technology

In commercial activity, user adopts business card as the instrument of indicate identification usually.But the paper business card in conventional art is inconvenient to carry and take care of, user usually need manually by the Data Enter on paper business card in intelligent terminal.

In order to the information on paper business card automatically can be identified, used visiting-card management software needs the character area in the business card image that first locating takes pictures obtains usually, and then by OCR(OpticalCharacterRecognition, optical character identification) character area changes into text message by system.

But in conventional art, the method for character area is inaccurate in positioning image, can miss key message, make the accuracy of location not high.

Summary of the invention

Based on this, be necessary to provide a kind of method that can improve character area in the positioning image of accuracy.

A method for character area in positioning image, comprising:

Obtain original image;

By Roberts operator, conversion is carried out to described original image and generate gray level image;

Described Binary Sketch of Grey Scale Image is obtained edge image;

Described edge image is expanded, extracts the positional information of the connected domain of described edge image.

Wherein in an embodiment, the described step described original image being converted to gray level image by Roberts operator is:

According to formula:

$A_1=\sqrt{{(I(i+1,j+1,R)-I(i,j,R))}^2+{(I(i+1,j+1,G)-I(i,j,G))}^2+{(I(i+1,j+1,B)-I(i,j,B))}^2};$

$A_2=\sqrt{{(I(i,j+1,R)-I(i+1,j+1,R))}^2+{(I(i,j+1,G)-I(i+1,j+1,G))}^2+{(I(i,j+1,B)-I(i+1,j+1,B))}^2};$

$I_g(i,j)=\sqrt{{A_1}^2+{A_2}^2}$

Generate gray level image; Wherein (i, j) is the pixel in described original image, and I (i, j, R), I (i, j, G) and I (i, j, B) are respectively the RGB color component of pixel (i, j), A ₁for pixel (i, j) and colored Euler's distance of contiguous pixel (i+1, j+1), A ₂for pixel (i, j+1) and colored Euler's distance of contiguous pixel (i+1, j+1), I _gthe gray-scale value of gray level image at pixel (i, j) place that (i, j) is described generation.

Wherein in an embodiment, the described step described Binary Sketch of Grey Scale Image being obtained edge image is:

By maximum between-cluster variance algorithm, binaryzation is carried out to described gray level image and obtain edge image.

Wherein in an embodiment, the step of the positional information of the connected domain of the described edge image of described extraction is:

The positional information of the connected domain of described edge image is extracted according to zone marker algorithm.

Wherein in an embodiment, also comprise after the step of the positional information of the connected domain of the described edge image of described extraction:

Obtain textural characteristics and/or the histogram feature of described connected domain;

Obtain the sorter of the support vector machine preset;

Described sorter is utilized to screen described positional information according to described textural characteristics and/or histogram feature.

In addition, there is a need to provide a kind of device that can improve character area in the positioning image of accuracy.

A device for character area in positioning image, comprising:

Image collection module, for obtaining original image;

Gray level image generation module, generates gray level image for carrying out conversion by Roberts operator to described original image;

Image binaryzation module, for obtaining edge image by described Binary Sketch of Grey Scale Image;

Zone location module, for being expanded by described edge image, extracts the positional information of the connected domain of described edge image.

Wherein in an embodiment, described gray level image generation module is also for according to formula:

$A_1=\sqrt{{(I(i+1,j+1,R)-I(i,j,R))}^2+{(I(i+1,j+1,G)-I(i,j,G))}^2+{(I(i+1,j+1,B)-I(i,j,B))}^2};$

$A_2=\sqrt{{(I(i,j+1,R)-I(i+1,j+1,R))}^2+{(I(i,j+1,G)-I(i+1,j+1,G))}^2+{(I(i,j+1,B)-I(i+1,j+1,B))}^2};$

$I_g(i,j)=\sqrt{{A_1}^2+{A_2}^2}$

Generate gray level image; Wherein (i, j) is the pixel in described original image, and I (i, j, R), I (i, j, G) and I (i, j, B) are respectively the RGB color component of pixel (i, j), A ₁for pixel (i, j) and colored Euler's distance of contiguous pixel (i+1, j+1), A ₂for pixel (i, j+1) and colored Euler's distance of contiguous pixel (i+1, j+1), I _gthe gray-scale value of gray level image at pixel (i, j) place that (i, j) is described generation.

Wherein in an embodiment, described image binaryzation module also obtains edge image for carrying out binaryzation by maximum between-cluster variance algorithm to described gray level image.

Wherein in an embodiment, described zone location module is also for extracting the positional information of the connected domain of described edge image according to zone marker algorithm.

Wherein in an embodiment, described device also comprises region screening module, for obtaining textural characteristics and/or the histogram feature of described connected domain, obtain the sorter of the support vector machine preset, utilize described sorter to screen described positional information according to described textural characteristics and/or histogram feature.

The method of character area and device in above-mentioned positioning image, first by Roberts operator, original image is changed, obtain the gray level image that gray-scale value contains marginal information, then by Binary Sketch of Grey Scale Image is obtained edge image, thus the marginal information extracted in gray level image, then namely being obtained the positional information of connected domain by expanding, the position that namely in image, character area is residing in original image, making location more accurate.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the method for character area in positioning image in an embodiment;

Fig. 2 is the design sketch of original image in an embodiment;

Fig. 3 is the design sketch of edge image in an embodiment;

Fig. 4 is the schematic diagram of the positional information of the connected domain got in an embodiment;

Fig. 5 is the schematic diagram of positional information of the connected domain in an embodiment after screening;

Fig. 6 is the structural representation of the device of character area in positioning image in an embodiment;

Fig. 7 is the structural representation of the device of character area in positioning image in another embodiment.

Embodiment

In one embodiment, as shown in Figure 1, a kind of method of character area in positioning image, comprising:

Step S102, obtains original image.

Original image obtains by taking pictures.Such as, in one scenario, by mobile terminal, business card is taken pictures, obtain the photomap picture obtained of taking pictures.

Step S104, carries out conversion by Roberts operator to original image and generates gray level image.

Roberts operator is a kind of operator being found edge by local difference.In the present embodiment, can according to formula:

$A_1=\sqrt{{(I(i+1,j+1,R)-I(i,j,R))}^2+{(I(i+1,j+1,G)-I(i,j,G))}^2+{(I(i+1,j+1,B)-I(i,j,B))}^2};$

$A_2=\sqrt{{(I(i,j+1,R)-I(i+1,j+1,R))}^2+{(I(i,j+1,G)-I(i+1,j+1,G))}^2+{(I(i,j+1,B)-I(i+1,j+1,B))}^2};$

$I_g(i,j)=\sqrt{{A_1}^2+{A_2}^2}$

Generate gray level image; Wherein (i, j) is the pixel in original image, and I (i, j, R), I (i, j, G) and I (i, j, B) are respectively the RGB color component of pixel (i, j), A ₁for pixel (i, j) and colored Euler's distance of contiguous pixel (i+1, j+1), A ₂for pixel (i, j+1) and colored Euler's distance of contiguous pixel (i+1, j+1), I _g(i, j) gray level image gray-scale value at pixel (i, j) place for generating.

As can be seen from above-mentioned formula, if the pixel (i, j) that is adjacent of the pixel (i+1, j+1) in original image and (i, j+1) gap in RGB color deal is comparatively large, then the gray level image I generated _gin, the gray-scale value of pixel (i, j) is larger.That is, gray level image I _gthe part that middle gray-scale value is higher is the fringe region in original image.

Step S106, obtains edge image by Binary Sketch of Grey Scale Image.

In the present embodiment, step S106 generation gray level image can be carried out binaryzation by maximum between-cluster variance algorithm to region unit and obtain edge image.

Maximum between-cluster variance algorithm is again Otsu algorithm (OTSU algorithm).This algorithm obtains threshold value T by the pixel in traversal image-region, and threshold value T makes the value of the maximum between-cluster variance of gray level image be maximal value, then according to threshold value T by image-region binaryzation.The gray-scale value of pixel corresponding in the edge image that the pixel that namely in gray level image, gray-scale value is less than threshold value T obtains after binarization is 0; The gray-scale value of pixel corresponding in the edge image that the pixel that in gray level image, gray-scale value is greater than threshold value T obtains after binarization is 255.

In the present embodiment, by traversal image-region in pixel obtain the step of threshold value T before also can according to preset threshold interval pixel is filtered.Preferably, threshold interval is:

(Min+w ₁×Len,Max+w ₂×Len)；

And Len=Max-Min+1

Wherein, Min is the minimum value of gray-scale value in gray level image, and Max is the maximal value of gray-scale value in gray level image, and Len is intermediate variable, w ₁and w ₂for weight coefficient, preferred w ₁and w ₂equal value is between 0.1 to 0.4.

That is, can filter out according to above-mentioned threshold interval the pixel that gray-scale value is too low or gray-scale value is too high in advance, then obtain threshold value T according to the maximum between-cluster variance algorithm traversal pixel belonged in above-mentioned threshold interval.Filter out the part that gray-scale value is too low or too high, low gray-scale pixels point and high gray-scale pixels point can be removed on the impact of threshold value T, thus make location more accurate.

Step S108, expands edge image, extracts the positional information of the connected domain of edge image.

The expansion of bianry image is namely by adopting the pixel in structural element traversing graph picture, if there is the pixel that gray-scale value is 0 in the pixel that structural element covers, in edge image after then expanding, the gray-scale value of the pixel in the region that this structural element covers is 0.Structural element and pixel size are the template of M × N, and in the present embodiment, M and N is 3, and namely structural element is the template of 3 × 3 sizes.

In the present embodiment, the step extracting the positional information of the connected domain of edge image can be specially: the positional information extracting the connected domain of edge image according to zone marker algorithm.

Identical for gray-scale value in edge image (gray level image after binaryzation) after expansion and that continuous print pixel is formed connected domain (continuous print image-region) can be marked by zone marker algorithm, and obtains its positional information.

In the present embodiment, further, textural characteristics and/or the histogram feature of connected domain also can be obtained after extracting the step of the positional information of the connected domain of edge image, obtain the sorter of the support vector machine preset, utilize sorter according to textural characteristics and/or histogram feature screening positional information.

Textural characteristics and/or the histogram feature of the training image with typical writings feature can be extracted in advance, then to be inputted in support vector machine thus to generate the kernel function of sorter.After the positional information of connected domain getting edge image, the textural characteristics of the image-region that the positional information obtaining connected domain is again demarcated and/or histogram feature, and be input in support vector machine, by the kernel function of sorter, connected domain is screened, thus filter out the image-region meeting typical writings feature.

Due in the image-region that obtained by above-mentioned steps S102, step S102, step S106 step S108, the image-region of non-legible type may be comprised, such as, the logo of character type, polar plot etc., by the sorter of support vector machine, it is screened, the image-region with typical writings feature can be obtained more exactly, thus make location more accurate.

In one embodiment, please also refer to Fig. 2, Fig. 3, Fig. 4 and Fig. 5, wherein, Fig. 2 is the original image of business card (virtual portrait) gathered of taking pictures, Fig. 3 obtains edge image after binaryzation, and Fig. 4 is the design sketch of the positional information of the connected domain adopting zone marker algorithm to obtain, and this positional information is shown with the form of rectangle frame, Fig. 5 is the positional information of the connected domain obtained after sorter screening, and this positional information is shown with the form of rectangle frame.As can be seen from above-mentioned exemplary plot, the positional information of the character area in this business card image is accurately extracted at the receiving end out.

In one embodiment, as shown in Figure 6, the device of character area in a kind of positioning image, comprising: image collection module 102, image blurring module 104, mixing constant acquisition module 106 and image blend module 108.Wherein:

Image collection module 102, for obtaining original image.

Original image obtains by taking pictures.Such as, in one scenario, by mobile terminal, business card is taken pictures, obtain the photomap picture obtained of taking pictures.

Gray level image generation module 104, generates gray level image for carrying out conversion by Roberts operator to original image.

Roberts operator is a kind of operator being found edge by local difference.In the present embodiment, gray level image generation module 104 can be used for according to formula:

$A_1=\sqrt{{(I(i+1,j+1,R)-I(i,j,R))}^2+{(I(i+1,j+1,G)-I(i,j,G))}^2+{(I(i+1,j+1,B)-I(i,j,B))}^2};$

$A_2=\sqrt{{(I(i,j+1,R)-I(i+1,j+1,R))}^2+{(I(i,j+1,G)-I(i+1,j+1,G))}^2+{(I(i,j+1,B)-I(i+1,j+1,B))}^2};$

$I_g(i,j)=\sqrt{{A_1}^2+{A_2}^2}$

Generate gray level image; Wherein (i, j) is the pixel in original image, and I (i, j, R), I (i, j, G) and I (i, j, B) are respectively the RGB color component of pixel (i, j), A ₁for pixel (i, j) and colored Euler's distance of contiguous pixel (i+1, j+1), A ₂for pixel (i, j+1) and colored Euler's distance of contiguous pixel (i+1, j+1), I _g(i, j) gray level image gray-scale value at pixel (i, j) place for generating.

As can be seen from above-mentioned formula, if the pixel (i, j) that is adjacent of the pixel (i+1, j+1) in original image and (i, j+1) gap in RGB color deal is comparatively large, then the gray level image I generated _gin, the gray-scale value of pixel (i, j) is larger.That is, gray level image I _gthe part that middle gray-scale value is higher is the fringe region in original image.

Image binaryzation module 106, for obtaining edge image by Binary Sketch of Grey Scale Image.

In the present embodiment, image binaryzation module 106 can be used for gray level image generation module 104 to generate gray level image and carries out binaryzation by maximum between-cluster variance algorithm to region unit and obtain edge image.

Maximum between-cluster variance algorithm is again Otsu algorithm (OTSU algorithm).This algorithm obtains threshold value T by the pixel in traversal image-region, and threshold value T makes the value of the maximum between-cluster variance of gray level image be maximal value, then according to threshold value T by image-region binaryzation.The gray-scale value of pixel corresponding in the edge image that the pixel that namely in gray level image, gray-scale value is less than threshold value T obtains after binarization is 0; The gray-scale value of pixel corresponding in the edge image that the pixel that in gray level image, gray-scale value is greater than threshold value T obtains after binarization is 255.

In the present embodiment, the threshold interval that image binaryzation module 106 also can be used for according to presetting filters pixel.Preferably, threshold interval is:

(Min+w ₁×Len,Max+w ₂×Len)；

And Len=Max-Min+1

Wherein, Min is the minimum value of gray-scale value in gray level image, and Max is the maximal value of gray-scale value in gray level image, and Len is intermediate variable, w ₁and w ₂for weight coefficient, preferred w ₁and w ₂equal value is between 0.1 to 0.4.

That is, can filter out according to above-mentioned threshold interval the pixel that gray-scale value is too low or gray-scale value is too high in advance, then obtain threshold value T according to the maximum between-cluster variance algorithm traversal pixel belonged in above-mentioned threshold interval.Filter out the part that gray-scale value is too low or too high, low gray-scale pixels point and high gray-scale pixels point can be removed on the impact of threshold value T, thus make location more accurate.

Zone location module 108, for being expanded by edge image, extracts the positional information of the connected domain of edge image.

The expansion of bianry image is namely by adopting the pixel in structural element traversing graph picture, if there is the pixel that gray-scale value is 0 in the pixel that structural element covers, in edge image after then expanding, the gray-scale value of the pixel in the region that this structural element covers is 0.Structural element and pixel size are the template of M × N, and in the present embodiment, M and N is 3, and namely structural element is the template of 3 × 3 sizes.

In the present embodiment, zone location module 108 can be used for the positional information of the connected domain extracting edge image according to zone marker algorithm.

Identical for gray-scale value in edge image (gray level image after binaryzation) after expansion and that continuous print pixel is formed connected domain (continuous print image-region) can be marked by zone marker algorithm, and obtains its positional information.

In the present embodiment, as shown in Figure 7, further, in positioning image, the device of character area also comprises region screening module 110, for obtaining textural characteristics and/or the histogram feature of connected domain, obtain the sorter of the support vector machine preset, utilize sorter according to textural characteristics and/or histogram feature screening positional information.

Textural characteristics and/or the histogram feature of the training image with typical writings feature can be extracted in advance, then to be inputted in support vector machine thus to generate the kernel function of sorter.After the positional information of connected domain getting edge image, the textural characteristics of the image-region that the positional information that region screening module 110 can be used for obtaining connected domain is demarcated and/or histogram feature, and be input in support vector machine, by the kernel function of sorter, connected domain is screened, thus filter out the image-region meeting typical writings feature.

Due in the image-region that obtained by aforementioned modules, the image-region of non-legible type may be comprised, such as, the logo of word shape, polar plot etc., by the sorter of support vector machine, it is screened, the image-region with typical writings feature can be obtained more exactly, thus make location more accurate.

In one embodiment, please also refer to Fig. 2, Fig. 3, Fig. 4 and Fig. 5, wherein, Fig. 2 is the original image of business card (virtual portrait) gathered of taking pictures, Fig. 3 obtains edge image after binaryzation, and Fig. 4 is the design sketch of the positional information of the connected domain adopting zone marker algorithm to obtain, and this positional information is shown with the form of rectangle frame, Fig. 5 is the positional information of the connected domain obtained after sorter screening, and this positional information is shown with the form of rectangle frame.As can be seen from above-mentioned exemplary plot, the positional information of the character area in this business card image is accurately extracted at the receiving end out.

The method of character area and device in above-mentioned positioning image, first by Roberts operator, original image is changed, obtain the gray level image that gray-scale value contains marginal information, then by Binary Sketch of Grey Scale Image is obtained edge image, thus the marginal information extracted in gray level image, then namely being obtained the positional information of connected domain by expanding, the position that namely in image, character area is residing in original image, making location more accurate.

The above embodiment only have expressed several embodiment of the present invention, and it describes comparatively concrete and detailed, but therefore can not be interpreted as the restriction to the scope of the claims of the present invention.It should be pointed out that for the person of ordinary skill of the art, without departing from the inventive concept of the premise, can also make some distortion and improvement, these all belong to protection scope of the present invention.Therefore, the protection domain of patent of the present invention should be as the criterion with claims.

Claims (8)

1. the method for character area in positioning image, comprising:

Obtain original image;

By Roberts operator, conversion is carried out to described original image and generate gray level image;

Described Binary Sketch of Grey Scale Image is obtained edge image, be specially according to preset threshold interval the pixel in described gray level image is filtered after, carry out binaryzation by maximum between-cluster variance algorithm to described gray level image and obtain edge image, described default threshold interval is (Min+w ₁× Len, Max+w ₂× Len), wherein, Len=Max-Min+1, Min are the minimum value of gray-scale value in described gray level image, and Max is the maximal value of gray-scale value in described gray level image, and Len is intermediate variable, w ₁and w ₂for weight coefficient;

Described edge image is expanded, extracts the positional information of the connected domain of described edge image.

2. the method for character area in positioning image according to claim 1, is characterized in that, the described step described original image being converted to gray level image by Roberts operator is:

According to formula:

$A_1=\sqrt{{\left(I\right(i+1,j+1,R)-I(i,j,R\left)\right)}^2+{\left(I\right(i+1,j+1,G)-I(i,j,G\left)\right)}^2+{\left(I\right(i+1,j+1,B)-I(i,j,B\left)\right)}^2};$

$A_2=\sqrt{{\left(I\right(i,j+1,R)-I(i+1,j+1,R\left)\right)}^2+{\left(I\right(i,j+1,G)-I(i+1,j+1,G\left)\right)}^2+{\left(I\right(i,j+1,B)-I(i+1,j+1,B\left)\right)}^2};$

$I_g(i,j)=\sqrt{{A_1}^2+{A_2}^2}$

Generate gray level image; Wherein (i, j) is the pixel in described original image, and I (i, j, R), I (i, j, G) and I (i, j, B) are respectively the RGB color component of pixel (i, j), A ₁for pixel (i, j) and colored Euler's distance of contiguous pixel (i+1, j+1), A ₂for pixel (i, j+1) and colored Euler's distance of contiguous pixel (i+1, j+1), I _gthe gray-scale value of gray level image at pixel (i, j) place that (i, j) is described generation.

3. the method for character area in positioning image according to claim 1, it is characterized in that, the step of the positional information of the connected domain of the described edge image of described extraction is:

The positional information of the connected domain of described edge image is extracted according to zone marker algorithm.

4. the method for character area in positioning image according to claim 1, is characterized in that, also comprise after the step of the positional information of the connected domain of the described edge image of described extraction:

Obtain textural characteristics and/or the histogram feature of described connected domain;

Obtain the sorter of the support vector machine preset;

Described sorter is utilized to screen described positional information according to described textural characteristics and/or histogram feature.

5. the device of character area in positioning image, is characterized in that, comprising:

Image collection module, for obtaining original image;

Gray level image generation module, generates gray level image for carrying out conversion by Roberts operator to described original image;

Image binaryzation module, for described Binary Sketch of Grey Scale Image is obtained edge image, after the pixel of threshold interval to described gray level image also preset for basis filters, carry out binaryzation by maximum between-cluster variance algorithm to described gray level image and obtain edge image, described default threshold interval is (Min+w ₁× Len, Max+w ₂× Len), wherein, Len=Max-Min+1, Min are the minimum value of gray-scale value in described gray level image, and Max is the maximal value of gray-scale value in described gray level image, and Len is intermediate variable, w ₁and w ₂for weight coefficient;

Zone location module, for being expanded by described edge image, extracts the positional information of the connected domain of described edge image.

6. the device of character area in positioning image according to claim 5, is characterized in that, described gray level image generation module is also for according to formula:

$A_1=\sqrt{{\left(I\right(i+1,j+1,R)-I(i,j,R\left)\right)}^2+{\left(I\right(i+1,j+1,G)-I(i,j,G\left)\right)}^2+{\left(I\right(i+1,j+1,B)-I(i,j,B\left)\right)}^2};$

$A_2=\sqrt{{\left(I\right(i,j+1,R)-I(i+1,j+1,R\left)\right)}^2+{\left(I\right(i,j+1,G)-I(i+1,j+1,G\left)\right)}^2+{\left(I\right(i,j+1,B)-I(i+1,j+1,B\left)\right)}^2};$

$I_g(i,j)=\sqrt{{A_1}^2+{A_2}^2}$

Generate gray level image; Wherein (i, j) is the pixel in described original image, and I (i, j, R), I (i, j, G) and I (i, j, B) are respectively the RGB color component of pixel (i, j), A ₁for pixel (i, j) and colored Euler's distance of contiguous pixel (i+1, j+1), A ₂for pixel (i, j+1) and colored Euler's distance of contiguous pixel (i+1, j+1), I _gthe gray-scale value of gray level image at pixel (i, j) place that (i, j) is described generation.

7. the device of character area in positioning image according to claim 5, it is characterized in that, described zone location module is also for extracting the positional information of the connected domain of described edge image according to zone marker algorithm.

8. the device of character area in positioning image according to claim 5, it is characterized in that, described device also comprises region screening module, for obtaining textural characteristics and/or the histogram feature of described connected domain, obtain the sorter of the support vector machine preset, utilize described sorter to screen described positional information according to described textural characteristics and/or histogram feature.