CN109726643A

CN109726643A - The recognition methods of form data, device, electronic equipment and storage medium in image

Info

Publication number: CN109726643A
Application number: CN201811528393.8A
Authority: CN
Inventors: 郑磊波; 王洪伟; 刘天悦
Original assignee: Chengdu Kingsoft Interactive Entertainment Co Ltd; Beijing Jinshan Digital Entertainment Technology Co Ltd
Current assignee: Beijing Kingsoft Digital Entertainment Co Ltd; Chengdu Kingsoft Interactive Entertainment Co Ltd; Beijing Jinshan Digital Entertainment Technology Co Ltd
Priority date: 2018-12-13
Filing date: 2018-12-13
Publication date: 2019-05-07
Anticipated expiration: 2038-12-13
Also published as: CN109726643B; CN112818813B; CN112818812B; CN112818813A; CN112818812A

Abstract

The embodiment of the invention provides recognition methods, device, electronic equipment and the storage mediums of form data in a kind of image, which comprises receives the target image with table；The form image comprising table is determined from target image；Line of text detection is carried out to form image, determines the position of line of text in form image；Form image is identified according to the position of line of text, obtains the form data of form image, wherein the form data includes text information and tableau format information.Due to identifying that obtained form data includes text information and tableau format information, rather than just the word content in table, therefore the diversity of the Table recognition result in image is improved, be further processed using subsequent progress table recovery etc..

Description

The recognition methods of form data, device, electronic equipment and storage medium in image

Technical field

The present invention relates to technical field of image processing, recognition methods, dress more particularly to form data in a kind of image It sets, electronic equipment and storage medium.

Background technique

Having a kind of image in field of image processing is the image for including table, in order to obtain the content of the table in image, It needs to identify the image for including table.

It is general to the identification process of table in image at present are as follows: horizontal line and vertical line first in extraction image, if nothing Horizontal line and vertical line, then without table in determinating area；If having horizontal line and vertical line, using region growing method come really Determine the position of table in image, and then according to the position of table in image, text identification is carried out to table in image, obtains image In table in word content.

In above-mentioned image in the identification process of table, obtained recognition result is only the word content in table, information It is less, it is highly detrimental to subsequent table be carried out restoring etc. to be further processed.

Summary of the invention

The embodiment of the present invention is designed to provide the recognition methods of form data, device, electronic equipment in a kind of image And storage medium is further processed with improving the diversity of the Table recognition result in image using subsequent.Particular technique Scheme is as follows:

In a first aspect, the embodiment of the invention provides a kind of recognition methods of form data in image, which comprises

Receive the target image with table；

The form image comprising table is determined from the target image；

Line of text detection is carried out to the form image, determines the position of line of text in the form image；

The form image is identified according to the position of the line of text, obtains the table letter of the form image Breath, wherein the form data includes text information and tableau format information.

Optionally, the form image is identified in the position according to the line of text, obtains the table Before the step of form data of image, the method also includes:

Remove all table lines of the form image；

The position according to the line of text identifies the form image, obtains the table of the form image The step of information, comprising:

According to the position of the line of text, text image is partitioned into from the form image after removal table line；

The text image being partitioned into is identified, the text information of the form image is obtained；

Determine whether the table line of the form image is complete；

If the table line of the form image is imperfect, by the table line completion of the form image；

Table recognition is carried out to the complete form image of table line, obtains the tableau format information of the form image.

Optionally, the whether complete step of the table line of the determination form image, comprising:

Based on the position of line of text in the form image, the character in the form image is removed；

Number of intersections and the quantity of closed cell lattice in form image after determining removal character；

The cell quantity of the table is determined according to the number of intersections of the table line；

Judge whether quantity and the cell quantity of the closed cell lattice are equal；

If the quantity of the closed cell lattice is equal with the cell quantity, the table line of the form image is determined Completely；

If the quantity of the closed cell lattice and the cell quantity are unequal, the table of the form image is determined Line is imperfect.

Optionally, the step of number of intersections and the quantity of closed cell lattice in the form image after the determining removal character Suddenly, comprising:

Corrosion treatment is carried out to the intermediate image, obtains corrosion image；

Expansion process is carried out to the corrosion image, obtains expanding image；

Transverse direction and longitudinal direction table line is carried out to the expanding image separately to handle, and obtains horizontal line image and vertical line image；

The horizontal line image and the vertical line image are carried out that union is taken to handle, obtain table line image；

The horizontal line image and the vertical line image are carried out that intersection is taken to handle, obtain intersection point image；

According to the intersection point image, number of intersections in the form image after determining removal character；

According to the table line image, the quantity of closed cell lattice in the form image after determining removal character.

Optionally, described that the text image being partitioned into is identified, obtain the step of the text information of the table Suddenly, comprising:

Text region is carried out to the text image being partitioned into, obtains the Text region result of the form image；

Semantic analysis is carried out to the Text region result, obtains the corresponding semanteme of each line of text；

According to the corresponding semanteme of each line of text, classify to the Text region result, obtains each text and know The corresponding classification of other result；

According to the corresponding classification of the Text region result, the Text region result is stored, the table is obtained The text information of table images.

Optionally, described the step of determination includes the form image of table from the target image, comprising:

The target image is inputted into the deep learning model that training is completed in advance, obtains table in the target image Target position；

According to the target position, judge whether the corresponding table area in the target position distorts；

If so, carrying out affine transformation processing to the table area, the corresponding form image of the target image is obtained.

Second aspect, the embodiment of the invention provides a kind of identification device of form data in image, described device includes:

Target image receiving module, for receiving the target image with table；

Form image determining module, for determining the form image comprising table from the target image；

Line of text position determination module determines the form image for carrying out line of text detection to the form image The position of middle line of text；

Information identification module obtains described for being identified according to the position of the line of text to the form image The form data of form image, wherein the form data includes text information and tableau format information.

Optionally, described device further include:

Table line removes module, for being identified in the position according to the line of text to the form image, Before obtaining the form data of the form image,

Remove all table lines of the form image；

The information identification module includes:

Image segmentation unit is divided from the form image after removal table line for the position according to the line of text Text image out；

Word recognition unit obtains the text of the form image for identifying to the text image that is partitioned into Information；

Whether table line determination unit, the table line for determining the form image are complete；

Table line completion unit, if the table line for the form image is imperfect, by the table of the form image Ruling completion；

Table recognition unit obtains the form image for carrying out Table recognition to the complete form image of table line Tableau format information.

Optionally, the table line determining module includes:

Character removal unit removes in the form image for the position based on line of text in the form image Character；

First quantity determination unit, for determining number of intersections and closed cell lattice in the form image after removing character Quantity；

Second quantity determination unit, for determining the cell number of the table according to the number of intersections of the table line Amount；

Whether quantity judging unit, the quantity and the cell quantity for judging the closed cell lattice are equal；

First table line determination unit, if the quantity for the closed cell lattice is equal with the cell quantity, Determine that the table line of the form image is complete；

Second table line determination unit, if quantity and the cell quantity for the closed cell lattice not phase Deng determining that the table line of the form image is imperfect.

Optionally, the first quantity determination unit includes:

Binary conversion treatment subelement, for the form image after the removal character to be carried out binary conversion treatment and to pixel Value carries out negating processing, obtains intermediate image；

Image erosion subelement obtains corrosion image for carrying out corrosion treatment to the intermediate image；

Image expansion subelement obtains expanding image for carrying out expansion process to the corrosion image；

Table line separates subelement, separately handles, obtains for carrying out transverse direction and longitudinal direction table line to the expanding image Horizontal line image and vertical line image；

Table line image determines subelement, for the horizontal line image and the vertical line image carrying out that union is taken to handle, Obtain table line image；

Intersection point image determines subelement, for the horizontal line image and the vertical line image carrying out that intersection is taken to handle, obtains To intersection point image；

Number of intersections determines subelement, for being handed in the form image after determining removal character according to the intersection point image Point quantity；

Cell quantity determines subelement, for the form image according to the table line image, after determining removal character The quantity of middle closed cell lattice.

Optionally, the word recognition unit includes:

Text region subelement obtains the tabular drawing for carrying out Text region to the text image being partitioned into The Text region result of picture；

It is corresponding to obtain each line of text for carrying out semantic analysis to the Text region result for semantic analysis subelement It is semantic；

Classify subelement, for classifying to the Text region result according to the corresponding semanteme of each line of text, Obtain the corresponding classification of each Text region result；

Recognition result storing sub-units are used for according to the corresponding classification of the Text region result, to the Text region As a result it is stored, obtains the text information of the form image.

Optionally, the form image determining module includes:

Target position determination unit is obtained for the target image to be inputted the deep learning model that training is completed in advance The target position of table into the target image；

Judging unit is distorted, for whether judging the corresponding table area in the target position according to the target position Distortion；

Form image determination unit, if for the corresponding table area distortion in the target position, to the table area Domain carries out affine transformation processing, obtains the corresponding form image of the target image.

The third aspect, the embodiment of the invention provides a kind of electronic equipment, including processor, communication interface, memory and Communication bus, wherein processor, communication interface, memory complete mutual communication by communication bus；

Memory, for storing computer program；

Processor when for executing the program stored on memory, realizes table letter in any of the above-described image The recognition methods step of breath.

Fourth aspect, the embodiment of the invention provides a kind of computer readable storage medium, the computer-readable storage Dielectric memory contains computer program, and the computer program realizes table in any of the above-described image when being executed by processor The recognition methods step of lattice information.

In scheme provided by the embodiment of the present invention, electronic equipment can receive the target image with table first, so The form image comprising table is determined from target image afterwards, then line of text detection is carried out to form image, determines form image The position of middle line of text, and then described image is identified according to the position of line of text, the form data of form image is obtained, Wherein, form data includes text information and tableau format information.Due to identify obtained form data include text information and Tableau format information rather than just the word content in table, therefore improves the multiplicity of the Table recognition result in image Property, it is further processed using subsequent progress table recovery etc..

Detailed description of the invention

In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with It obtains other drawings based on these drawings.

Fig. 1 is the flow chart of the recognition methods of form data in a kind of image provided by the embodiment of the present invention；

Fig. 2 (a) is a kind of schematic diagram of artificial marquee provided by the embodiment of the present invention；

Fig. 2 (b) is the schematic diagram of the artificial marquee of another kind provided by the embodiment of the present invention；

Fig. 3 is a kind of schematic diagram of the position of line of text in the form image based on embodiment illustrated in fig. 1；

Fig. 4 is a kind of specific flow chart of step S104 in embodiment illustrated in fig. 1；

Fig. 5 is a kind of specific flow chart of step S403 in embodiment illustrated in fig. 4；

Fig. 6 is a kind of schematic diagram of the intersection point of the table line based on embodiment illustrated in fig. 5；

Fig. 7 is a kind of specific flow chart of step S502 in embodiment illustrated in fig. 5；

Fig. 8 (a) is a kind of schematic diagram of form image based on embodiment illustrated in fig. 1；

Fig. 8 (b) is a kind of schematic diagram of intermediate image based on embodiment illustrated in fig. 1；

Fig. 8 (c) is a kind of schematic diagram of horizontal line image based on embodiment illustrated in fig. 1；

Fig. 8 (d) is a kind of schematic diagram of vertical line image based on embodiment illustrated in fig. 1；

Fig. 8 (e) is a kind of schematic diagram of table line image based on embodiment illustrated in fig. 1；

Fig. 8 (f) is a kind of schematic diagram of intersection point image based on embodiment illustrated in fig. 1；

Fig. 9 is a kind of specific flow chart of step S104 in embodiment illustrated in fig. 1；

Figure 10 is a kind of flow chart of the training method of the deep learning model based on embodiment illustrated in fig. 1；

Figure 11 is the structural schematic diagram of the identification device of form data in a kind of image provided by the embodiment of the present invention；

Figure 12 is the structural schematic diagram of a kind of electronic equipment provided by the embodiment of the present invention.

Specific embodiment

Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other Embodiment shall fall within the protection scope of the present invention.

In order to improve the accuracy of the Table recognition in image, the embodiment of the invention provides form datas in a kind of image Recognition methods, device, electronic equipment and computer readable storage medium.

The recognition methods for being provided for the embodiments of the invention form data in a kind of image first below is introduced.

The recognition methods of form data can be applied to any need pair in a kind of image provided by the embodiment of the present invention The electronic equipment that form data is identified in image is not done specific herein for example, can be computer, mobile phone, smartwatch etc. It limits.For ease of description, hereinafter referred to as electronic equipment.

As shown in Figure 1, in a kind of image form data recognition methods, the described method comprises the following steps:

S101 receives the target image with table；

S102 determines the form image comprising table from the target image；

S103 carries out line of text detection to the form image, determines the position of line of text in the form image；

S104 identifies the form image according to the position of the line of text, obtains the table of the form image Lattice information.

Wherein, the form data includes text information and tableau format information.

As it can be seen that electronic equipment can receive the target figure with table first in scheme provided by the embodiment of the present invention Then picture determines the form image comprising table from target image, then carries out line of text detection to form image, determine table The position of line of text in image, and then described image is identified according to the position of line of text, obtain the table of form image Information, wherein form data includes text information and tableau format information.Due to identifying that obtained form data includes text letter Breath and tableau format information rather than just the word content in table, therefore improve Table recognition result in image Diversity is further processed using subsequent progress table recovery etc..

In above-mentioned steps S101, electronic equipment can receive the target image with table, which is to need Carry out the image of form data identification.The available image with table being locally stored of electronic equipment, as target figure Picture.Also the image with table that can receive the transmission of other electronic equipments, as target image.Itself can certainly be passed through The image acquisition device of installation obtains the image with table, as target image, for example, being obtained by the camera that itself is installed Take the image with table.This is all reasonably, to be not specifically limited herein.

Electronic equipment can be shown within display screen when the image acquisition device installed by itself obtains target image Work of leting others have a look at marquee, such as shown in Fig. 2 (a) and Fig. 2 (b), user can change artificial marquee by pulling artificial marquee Shape, can be the shapes such as rectangle, trapezoidal, triangle.Image acquisition device acquires the region for including in artificial marquee i.e. Available target image.

After obtaining above-mentioned target image, in order to identify to the table in target image, electronic equipment can be from target The form image comprising table is determined in image.Electronic equipment can use the modes such as deep learning model, image detection and determine It include the form image of table in target image.In order to scheme understand and be laid out it is clear, it is subsequent will to from target image really Surely the mode of the form image comprising table carries out citing introduction.After obtaining above table image, electronic equipment can be to this Form image carries out line of text detection, determines the position of line of text in form image, i.e. execution above-mentioned steps S103.In a kind of reality It applies in mode, electronic equipment can use pixel link algorithm and carry out line of text detection to above table image, not do herein It illustrates and limits.

In order to improve line of text identification accuracy and more adapt to practical application scene, pixel link can be calculated Deep learning model used in method carries out adaptation adjustment, for example, adjusting its parameter, loss function etc., specific adjustment mode can To be not specifically limited and illustrate herein using the relevant way in deep learning model field.

The position of line of text is the position of all line of text in table in above table image, can be using line of text The apex coordinate of minimum circumscribed rectangle indicates, can use the coordinate representation on four vertex, naturally it is also possible to the seat of angular vertex Mark indicates.For example, as shown in figure 3, the coordinate representation of point 301- point 304 can be used, point 301 can also be used and put 303 Coordinate representation, naturally it is also possible to using the coordinate representation of point 302 and point 304.It is only exemplary in Fig. 3 and shows " age " institute The position of the position of corresponding line of text, other line of text is not shown.

Next, electronic equipment can carry out form image according to the position of line of text in above-mentioned steps S104 Identification, and then obtain the form data of form image.Wherein, form data may include text information and tableau format information.

Wherein, tableau format information may include ranks number, cell span information, cell borders information, cell Fill Color, the high information of cell width.Text information may include the information such as content of text, type, font, font size, color, It is not specifically limited herein.

As a kind of embodiment of the embodiment of the present invention, in the above-mentioned position according to the line of text to the tabular drawing Before as the step of being identified, obtain the form data of the form image, the above method can also include: the removal table All table lines of table images.

In order to remove influence of the table line to Text region, electronic equipment can be removed all table lines of form image It removes, in this way, just not will receive the influence of table line when carrying out Text region.

As an implementation, the color filling of table line can be the background colour of form image by electronic equipment, with Achieve the purpose that remove table line.For example, the background colour of form image is white, table line and character therein are black, So all table lines can be filled with white by electronic equipment, just leave behind the character of black in this way.

Correspondingly, obtaining as shown in figure 4, the above-mentioned position according to the line of text identifies the form image The step of form data of the form image, may include:

S401 is partitioned into text image from the form image after removal table line according to the position of the line of text；

In order to carry out Text region, electronic equipment can be according to the position for the line of text that line of text detects, from removal Text image is partitioned into form image after table line.For example, the position of above-mentioned line of text is the diagonal apex coordinate of rectangle (5,7.5) and (35,15) just obtain a text then electronic equipment can split the rectangular area from table This image.

The corresponding rectangular area of all texts in form image is partitioned by electronic equipment according to the position of line of text Come, the corresponding all text images of form image can be obtained.Since the table line of form image at this time has been removed, institute Even if table line will not be divided into text image in the case where table line and very close text distance.

S402 carries out Text region to the text image that is partitioned into, obtains the text information of the form image；

In turn, electronic equipment carries out Text region to text image is partitioned into, and can obtain the text letter of form image Breath.

S403 determines whether the table line of the form image is complete, if the table line of the form image is imperfect, Execute step S404；

In order to enable the tableau format information arrived is more accurate, electronic equipment can determine the table line of above table image It is whether complete.In one embodiment, electronic equipment can be by the quantity of closed cell lattice in detection form image come really Whether the table line for determining form image is complete, subsequent to carry out citing introduction to specific implementation.

If the table line of form image is complete, step S405 can be continued to execute.

S404, by the table line completion of the form image；

If the table line of form image is imperfect, electronic equipment can execute step S404, i.e., by tabular drawing Then the table line completion of picture executes step S405；

S405 carries out Table recognition to the complete form image of table line, obtains the tableau format letter of the form image Breath.

Electronic equipment can carry out Table recognition to the complete form image of table line, and then obtain the table of form image Lattice structural information.After obtaining other tableau format information, carries out recovery in order to subsequent and handle to obtain table, can store the table knot Structure information.

As it can be seen that in the present embodiment, the table image segmentation for removing all table lines can be text diagram by electronic equipment Picture, so that will not be comprising table line, in turn, so that obtained text information is more accurate in the text image that segmentation obtains.Together When completion processing can be carried out to the incomplete form image of table line, and then standard is obtained according to the complete form image of table line True tableau format information.

As a kind of embodiment of the embodiment of the present invention, as shown in figure 5, the table line of the above-mentioned determination form image Whether complete step, may include:

S501 removes the character in the form image based on the position of line of text in the form image；

The position of line of text in form image has been determined it, electronic equipment can be according to the position of line of text by form image In character removal, in order to not influence the quantity of subsequent determining closed cell lattice and the number of intersections of table line, electronic equipment can All to remove all characters, that is, only retain the table line of table.

In one embodiment, electronic equipment can fill out the corresponding rectangular area in the position of line of text in form image The background colour for form image is filled, to achieve the purpose that remove character.For example, the background colour of form image is white, table line And character therein is black, then all Character Fillings can be white by electronic equipment, just leaves behind black in this way Table line.

S502 determines number of intersections and the quantity of closed cell lattice in the form image after removing character；

In turn, electronic equipment can determine the quantity and table of closed cell lattice in the form image after removing character The number of intersections of line.As an implementation, after electronic equipment can be using findContours algorithm detection removal character Form image in the quantity of closed cell lattice and the number of intersections of table line.

Wherein, the intersection point of table line is the intersection point that two table lines are crossed to form, for example, as shown in fig. 6, shown in Fig. 6 It is the table of 2 rows 3 column, wherein point 610 is the intersection point of table line, shares 12.

S503 determines the cell quantity of the table according to the number of intersections of the table line；

The number of intersections of table line in form image is determined, electronic equipment also can be according to the number of intersections of table line Determine the cell quantity of table.

For example, the number of intersections of table line is 30, it then can determine that the table is the table of 4 rows 5 column, or is 5 rows 4 The table of column, then can determine that the cell quantity of table is 20.

S504 judges whether the quantity of the closed cell lattice and the cell quantity are equal, if the closing is single The quantity of first lattice is equal with the cell quantity, executes step S505；If the quantity of the closed cell lattice and the list First lattice quantity is unequal, executes step S506；

Next, electronic equipment can judge above-mentioned closed cell lattice quantity and identified cell quantity whether It is equal, if the quantity of closed cell lattice is equal with cell quantity, illustrate that all cells are all closed in form image, So that is to say, the table line of the table of bright form image is completely, there is no the lines of missing, then step can be executed Rapid S505 determines that the table line of form image is complete.

If the quantity of closed cell lattice and cell quantity are unequal, illustrate that cell is not all in form image Cell be all closed, then that is to say, the table line of the table of bright form image be it is incomplete, there are the lines of missing Item determines that the table line of form image is imperfect then step S506 can be executed.

For example, the quantity of closed cell lattice is 28, the cell quantity determined in step S503 is 30, then just illustrating table It is not closed that the cell of table, which has 2, in table images, then the table line of table is incomplete in form image.

S505 determines that the table line of the form image is complete；

S506 determines that the table line of the form image is imperfect.

As it can be seen that in the present embodiment, electronic equipment can remove form image based on the position of line of text in form image In character, determine removal character after form image in the quantity of closed cell lattice and the number of intersections of table line, in turn Then the cell quantity that table is determined according to the number of intersections of table line judges the quantity and cell quantity of closed cell lattice It is whether equal, if equal, it is determined that the table line of form image is complete, if unequal, it is determined that the table of form image Line is imperfect.Whether the table line that form image can be accurately determined in this way is complete, and then improves subsequent to table content knowledge Other accuracy.

As a kind of embodiment of the embodiment of the present invention, as shown in fig. 7, the form image after above-mentioned determining removal character The step of middle number of intersections and the quantity of closed cell lattice, may include:

Form image after the removal character is carried out binary conversion treatment and carries out negating processing to pixel value by S701, Obtain intermediate image；

In one embodiment, electronic equipment can use adaptiveThreshold algorithm will remove character after Form image carries out binary conversion treatment, and then electronic equipment can take the pixel value of the form image after binary conversion treatment Inverse processing obtains intermediate image.

For example, the form image as shown in Fig. 8 (a) carries out at binaryzation form image after character therein removal Reason, and pixel value is carried out to negate processing, it obtains shown in intermediate image such as Fig. 8 (b).As it can be seen that character and table in form image Line is black, carries out binary conversion treatment to form image, and carry out negating table in the intermediate image that processing obtains to pixel value Ruling is white, and rest part is black.

S702 carries out corrosion treatment to the intermediate image, obtains corrosion image；

Next, having duplicate part since some character potential range table lines are closer, or with table line, will cause It may include some pixels for being not belonging to table line in intermediate image, such as white dotted in Fig. 8 (b).So in order to more Number of intersections in form image is accurately determined, electronic equipment can be handled above-mentioned intermediate image using corrosion treatment, And then obtain corrosion image.

Corrosion treatment and expansion process are a kind of morphological operations to image, are substantially the shapes for changing objects in images Shape.Corrosion treatment and expansion process general action are in binary image, for connecting adjacent element or being separated into independent member Element.Corrosion treatment and expansion process are generally directed to the white portion in image.

Since corrosion treatment is to take local minimum in the zonule to image.Because above-mentioned intermediate image is binary picture Picture, pixel value only have 0 and 255, so it is 0 that the pixel value in zonule, which has one, then all pixels point in the zonule Become 0, so can will lose apart from the farther away character of table line when handling using corrosion treatment above-mentioned intermediate image The pixel stayed erodes.

S703 carries out expansion process to the corrosion image, obtains expanding image；

Next, electronic equipment can carry out expansion process to corrosion image, and then obtain expanding image.At expansion Reason is to take local maximum in the zonule to image.Because above-mentioned intermediate image is binary image, pixel value only has 0 He 255, so it is 255 that the pixel value in zonule, which has one, then all pixels point in the zonule becomes 255, so It can will be incorporated in table line apart from the pixel that the closer character of table line is left by the expansion process of table line.

S704 carries out transverse direction and longitudinal direction table line to the expanding image and separately handles, obtains horizontal line image and vertical line charting Picture；

After obtaining above-mentioned expanding image, electronic equipment can carry out transverse direction and longitudinal direction table line point to above-mentioned expanding image Processing is opened, obtained horizontal line image and vertical line image.Due to having carried out excessive erosion and expansion process, so obtained horizontal line figure There was only table line in picture and vertical line image.

For example, carrying out transverse direction and longitudinal direction table line after using corrosion and expansion process to intermediate image shown in Fig. 8 (b) Separately processing, obtained horizontal line image and vertical line image can be respectively as shown in Fig. 8 (c) and Fig. 8 (d).

S705 to the horizontal line image and the vertical line image carries out that union is taken to handle, obtains table line image；

In turn, electronic equipment to above-mentioned horizontal line image and above-mentioned vertical line image can carry out that union is taken to handle, it can To table line image.For example, such as Fig. 8 (c) and Fig. 8 (d) is shown respectively for horizontal line image and vertical line image, then to Fig. 8 (c) and figure 8 (d) carry out that union is taken to handle, and can obtain table line image 8 (e).

S706 to the horizontal line image and the vertical line image carries out that intersection is taken to handle, obtains intersection point image；

Electronic equipment to above-mentioned horizontal line image and above-mentioned vertical line image can also carry out that intersection is taken to handle, it can be handed over Point image.For example, horizontal line image and vertical line image be respectively as shown in Fig. 8 (c) and Fig. 8 (d), then to Fig. 8 (c) and Fig. 8 (d) into Row takes intersection to handle, and can obtain intersection point image graph 8 (f).

S707, according to the intersection point image, number of intersections in the form image after determining removal character；

After obtaining above-mentioned intersection point image, electronic equipment can determine number of intersections in the form image after removing character. For example, shown in intersection point image such as Fig. 8 (f), then can determine that number of intersections is 56.

S708, according to the table line image, the quantity of closed cell lattice in the form image after determining removal character.

After obtaining above table line image, electronic equipment can determine closed cell in the form image after removing character The quantity of lattice.For example, shown in intersection point image such as Fig. 8 (e), then can determine that the quantity of closed cell lattice in form image is 42。

As it can be seen that in the present embodiment, the form image after removal character can be carried out binary conversion treatment simultaneously by electronic equipment Pixel value is carried out to negate processing, obtains intermediate image, so using corrosion and expansion process to intermediate image carry out laterally and Vertical table ruling is separately handled, and obtains horizontal line image and vertical line image, as character is lost in obtained horizontal line image and vertical line image The pixel stayed, so that the number of intersections of subsequent determination and the quantity of closed cell lattice are more accurate.

For the ease of subsequent query and restore table content, as a kind of embodiment of the embodiment of the present invention, such as Fig. 9 It is shown, it the above-mentioned the step of text image being partitioned into is identified, obtains the text information of the table, can wrap It includes:

S901 carries out Text region to the text image being partitioned into, obtains the Text region knot of the form image Fruit；

Electronic equipment can carry out Text region to the text image being partitioned into, and then obtain the Text region of form image As a result.It wherein, can be using any Text region of field of character recognition in image for the specific implementation of Text region Mode is not specifically limited herein and illustrates as long as the word content in text image can be identified.

S902 carries out semantic analysis to the Text region result, obtains the corresponding semanteme of each line of text；

After obtaining above-mentioned Text region result, in order to carry out structured storage to Text region result, electronic equipment can be with Semantic analysis is carried out to the text recognition result, obtains the corresponding semanteme of each line of text.Wherein, language is carried out to Text region result Justice analysis specific implementation can use semantic analysis field any semantic analysis mode, be not specifically limited herein and Explanation.

S903 classifies to the Text region result, obtains each text according to the corresponding semanteme of each line of text The corresponding classification of word recognition result；

In turn, electronic equipment can classify to above-mentioned Text region result, obtain according to the corresponding semanteme of each line of text To the corresponding classification of each Text region result.For example, Text region result be " name ", " Zhang San ", " Li Si ", " age ", " 25 years old ", " 28 years old ", then " Zhang San ", " Li Si " corresponding semanteme are the name of people, " 25 years old ", " 28 years old " are corresponding semantic equal For the age of people, then Text region result " Zhang San " and " Li Si " and " name " can be divided into name one by electronic equipment table Text region result " 25 years old " and " 28 years old " and " age " are divided into age one kind by class.

S904 stores the Text region result, obtains institute according to the corresponding classification of the Text region result State the text information of form image.

After obtaining the corresponding classification of Text region result.Electronic equipment can will carry out Text region result according to classification Storage, obtains the text information of form image.

In one embodiment, electronic equipment can be with JSON (JavaScript Object Notation, object letter Spectrum) format key-value pair mode to Text region result carry out structured storage.Or it is illustrated by taking above-mentioned example as an example, Electronic equipment can by " name ", " age " as storage key, " Zhang San ", " Li Si " as value corresponding to key " name " into Row storage.Similarly, key by " age " as storage, " 25 years old ", " 28 years old " are deposited as value corresponding to key " age " Storage.

In order to more intuitively show the table in above table image, electronic equipment can be complete by above table line Form image after form image or completion table line is also stored.

Electronic equipment can also be by information such as the type of the character in form image, font, font size, color and above-mentioned Tableau format information is also stored, and later use text information and tableau format Information recovering is facilitated to obtain table.

As it can be seen that in the present embodiment, electronic equipment can carry out semantic analysis to Text region result, each line of text is obtained Corresponding semanteme, and then according to the corresponding semanteme of each line of text, classify to Text region result, according to classification results to text Word recognition result is stored.It can also be by the form image after the complete form image of above table line or completion table line And tableau format information etc. is also stored.In this way, can have been checked when user checks the corresponding information of the form image At form image and table content, more intuitive and convenient improves user experience, later use text can also be facilitated to believe Breath and tableau format Information recovering obtain table.

It is above-mentioned that the table comprising table is determined from the target image as a kind of embodiment of the embodiment of the present invention The step of image may include:

The target image is inputted into the deep learning model that training is completed in advance, obtains table in the target image Target position；According to the target position, judge whether the corresponding table area in the target position distorts；If so, to institute It states table area and carries out affine transformation processing, obtain the corresponding form image of the target image.

In order to determine the position of table in acquired target image, to identify to table, electronic equipment can lead to Cross the target position that the deep learning model that training is completed in advance determines table in target image.The deep learning model be based on What the form image sample obtained in advance was trained initial depth learning model, it can be with by the deep learning model Obtain the position of table in target image, that is, above-mentioned target position.

Wherein, deep learning model can be convolutional neural networks etc., and the specific structure present invention of deep learning model exists This is not specifically limited, as long as can obtain obtaining the deep learning model of the position of table in form image by training ?.The initial parameter of initial depth learning model can be set at random, be not specifically limited herein.In order to which scheme understands and cloth Office is clear, it is subsequent will the training method to deep learning model carry out citing introduction.

It has determined in above-mentioned target image behind the target position of table, electronic equipment can be according to the target position, really The table area to set the goal in image.For example, target position is four vertex of table in target image, then in target image Table area be four vertex determine region.

And then electronic equipment may determine that whether the corresponding table area in target position distorts, if non-warping, just Table area can not be processed, the corresponding image of the table section is above table image.Wherein, electronic equipment can root Determine whether table area distorts according to the coordinate of target position, for example, if the coordinate representation table area of target position is one Parallelogram, then can determine that table area is distortion；If the coordinate representation table area of target position is one Rectangle, then can determine that table area is non-warping.

If table area distorts, electronic equipment can carry out affine transformation processing to determining table area, obtain The corresponding form image of target image.Under many actual conditions, the table in target image that electronic equipment obtains is distortion , in order to still be accurately identified in this case to table content,

Electronic equipment can carry out affine transformation processing to table area, and then obtain the corresponding tabular drawing of target image Picture.

It is understood that table is usually rectangle, but table when scalloping, in target image Region may not be rectangle, but the shapes such as trapezoidal, then electronic equipment can carry out affine transformation to the table area Processing, and then the corresponding form image of target image is obtained, which is the form image after twist correcting.

Wherein, the specific implementation of affine transformation processing is carried out to table area, it can be using at any affine transformation Reason mode, as long as form image can be carried out twist correcting.For example, it is assumed that target position is the table in target image Apex coordinate, which indicates that table area is one trapezoidal, then electronic equipment can determine its corresponding square Four apex coordinates of shape, and then determine affine transformation matrix between the two, it can will be turned round according to the affine transformation matrix Bent table area carries out affine transformation processing, also can be obtained by the corresponding form image of target image.

As a kind of embodiment of the embodiment of the present invention, above-mentioned deep learning model may include form image and table The corresponding relationship of apex coordinate.It is above-mentioned that the target image is inputted to the depth that training is completed in advance in response to this Learning model the step of obtaining the target position of table in the target image, may include:

The target image is inputted into the deep learning model that training is completed in advance, obtains table in the target image Table apex coordinate.

In this embodiment, above-mentioned deep learning model may include form image pass corresponding with table apex coordinate System, wherein table apex coordinate is four apex coordinates of table, which has determined in image locating for table Region.

Since deep learning model can determine table in image according to the corresponding relationship of form image and table apex coordinate The apex coordinate in lattice region, so above-mentioned target image is inputted the deep learning model that training is completed in advance, the deep learning Model can be handled target image, and then output formats apex coordinate, the table apex coordinate i.e. target figure The table apex coordinate of table as in.

As it can be seen that in the present embodiment, target image can be inputted the deep learning mould that training is completed in advance by electronic equipment Type, and then the table apex coordinate of table in target image is obtained, it can accurately determine the table vertex of table in target image Coordinate, that is, the accurate specific region for determining the table of table in target image, can be further improved subsequent in table Hold the accuracy of identification.

As a kind of embodiment of the embodiment of the present invention, as shown in Figure 10, the training method of above-mentioned deep learning model, May include:

S1001 obtains form image sample and initial depth learning model；

Above-mentioned deep learning model in order to obtain, form image sample available first and initial depth learning model. Wherein, which can pre-establish, and can also obtain from other electronic equipments, this is all reasonable.

Form image sample is the image for including table, can only include table in form image sample, also may include Other content in addition to table, for example, picture, the text outside table, number etc..The quantity of form image sample is more A, particular number can be determines according to actual conditions.

S1002 marks the position of table area in the form image sample；

After obtaining form image sample, the position of table area in each form image sample can be marked.In a kind of reality It applies in mode, position of four apex coordinates of table area as table area can be used.

Form image sample after label is inputted the initial depth learning model, to the initial depth by S1003 Model is practised to be trained；

In flag table image pattern behind the position of table area, the form image sample after marking can be inputted Above-mentioned initial depth learning model is trained the initial depth learning model.In the training process, initial depth learns mould Type constantly learns the corresponding relationship of the position of form image feature and table area, constantly adjusts the parameter of itself.

The specific training method being trained to initial depth learning model can be common using gradient descent algorithm etc. Training method is not specifically limited herein.

S1004, when the accuracy of the output result of the initial depth learning model reaches preset value or the tabular drawing When decent training the number of iterations reaches preset times, deconditioning obtains the deep learning model.

When the accuracy of the output result of initial deep learning model reaches preset value, alternatively, the instruction of form image sample When white silk the number of iterations reaches preset times, illustrate that initial depth learning model at this time has been able to the various figures with table As being handled, the position of accurate table area is obtained.So can deconditioning, obtain above-mentioned deep learning model.

Wherein, above-mentioned preset value can be determined according to the requirement of the accuracy of the output result to deep learning model, example It such as, can be 90%, 95%, 98% etc..Above-mentioned preset times equally can be according to the output result to deep learning model The requirement of accuracy determines that, if accuracy is more demanding, preset times can be more, for example, can be 50,000 It is secondary, 80,000 times, it is 100,000 inferior；If the requirement of accuracy is lower, preset times can be less, for example, can be 10,000 It is secondary, 20,000 times, it is 30,000 inferior.

As it can be seen that in the present embodiment, the available form image sample of electronic equipment and initial depth learning model, label Then form image sample after label is inputted initial depth learning model by the position of table area in form image sample, Initial depth learning model is trained, when the accuracy of the output result of initial deep learning model reaches preset value, or When the training the number of iterations of form image sample reaches preset times, deconditioning obtains deep learning model.In this way, can be with The deep learning model that can accurately determine the position of table area in image is obtained, can be further improved form data identification Accuracy.

Corresponding to the recognition methods of form data in above-mentioned image, the embodiment of the invention also provides tables in a kind of image The identification device of information.

The identification device for being provided for the embodiments of the invention form data in a kind of image below is introduced.

As shown in figure 11, in a kind of image table identification device, described device includes:

Target image receiving module 1110, for receiving the target image with table；

Form image determining module 1120, for determining the form image comprising table from the target image；

Line of text position determination module 1130 determines the table for carrying out line of text detection to the form image The position of line of text in image；

Information identification module 1140 is obtained for being identified according to the position of the line of text to the form image The form data of the form image.

As a kind of embodiment of the embodiment of the present invention, above-mentioned apparatus can also include:

Table line remove module (being not shown in Figure 11), in the position according to the line of text to the table Image is identified, before obtaining the form data of the form image, removes all table lines of the form image；

Above- mentioned information identification module 1140 may include:

Image segmentation unit (is not shown) in Figure 11, for the position according to the line of text, after removal table line Text image is partitioned into form image；

Word recognition unit (is not shown) in Figure 11, for carrying out Text region to the text image that is partitioned into, obtains The text information of the form image；

Whether table line determination unit (being not shown in Figure 11), the table line for determining the form image are complete；

Table line completion unit (being not shown in Figure 11), if the table line for the form image is imperfect, by institute State the table line completion of form image；

Table recognition unit (is not shown) in Figure 11, for carrying out Table recognition to the complete form image of table line, obtains To the tableau format information of the form image.

As a kind of embodiment of the embodiment of the present invention, above table line determining module may include:

Character removal unit (is not shown) in Figure 11, for the position based on line of text in the form image, removes institute State the character in form image；

First quantity determination unit (being not shown in Figure 11), for determining number of intersections in the form image after removing character And the quantity of closed cell lattice；

Second quantity determination unit (being not shown in Figure 11), for determining the table according to the number of intersections of the table line The cell quantity of lattice；

Quantity judging unit (is not shown) in Figure 11, for judge the closed cell lattice quantity and the cell number It whether equal measures；

First table line determination unit (being not shown in Figure 11), if quantity and the list for the closed cell lattice First lattice quantity is equal, determines that the table line of the form image is complete；

Second table line determination unit (being not shown in Figure 11), if quantity and the list for the closed cell lattice First lattice quantity is unequal, determines that the table line of the form image is imperfect.

As a kind of embodiment of the embodiment of the present invention, above-mentioned first quantity determination unit may include:

Binary conversion treatment subelement (is not shown) in Figure 11, for the form image after the removal character to be carried out two-value Change and handle and pixel value is carried out to negate processing, obtains intermediate image；

Image erosion subelement (is not shown) in Figure 11, for carrying out corrosion treatment to the intermediate image, is corroded Image；

Image expansion subelement (is not shown) in Figure 11, for carrying out expansion process to the corrosion image, is expanded Image；

Table line separates subelement (being not shown in Figure 11), for carrying out transverse direction and longitudinal direction table line to the expanding image Separately processing, obtains horizontal line image and vertical line image；

Table line image determines subelement (being not shown in Figure 11), for the horizontal line image and the vertical line image into Row takes union to handle, and obtains table line image；

Intersection point image determines subelement (being not shown in Figure 11), for carrying out to the horizontal line image and the vertical line image It takes intersection to handle, obtains intersection point image；

Number of intersections determines subelement (being not shown in Figure 11), for determining after removing character according to the intersection point image Form image in number of intersections；

Cell quantity determines subelement (being not shown in Figure 11), for determining removal word according to the table line image The quantity of closed cell lattice in form image after symbol.

As a kind of embodiment of the embodiment of the present invention, above-mentioned word recognition unit may include:

Text region subelement (is not shown) in Figure 11, for carrying out Text region to the text image being partitioned into, Obtain the Text region result of the form image；

Semantic molecular cell (being not shown in Figure 11) obtains each for carrying out semantic analysis to the Text region result The corresponding semanteme of line of text；

Classify subelement (being not shown in Figure 11), for being known to the text according to the corresponding semanteme of each line of text Other result is classified, and the corresponding classification of each Text region result is obtained；

Recognition result storing sub-units (are not shown) in Figure 11, are used for according to the corresponding classification of the Text region result, The Text region result is stored, the text information of the form image is obtained.

As a kind of embodiment of the embodiment of the present invention, above table image determining module 1120 may include:

Target position determination unit (is not shown) in Figure 11, for the target image to be inputted the depth that training is completed in advance Learning model is spent, the target position of table in the target image is obtained；

It distorts judging unit (being not shown in Figure 11), for judging that the target position is corresponding according to the target position Table area whether distort；

Form image determination unit (is not shown) in Figure 11, if turned round for the corresponding table area in the target position Song carries out affine transformation processing to the table area, obtains the corresponding form image of the target image.

The embodiment of the invention also provides a kind of electronic equipment, and as shown in figure 12, electronic equipment may include processor 1201, communication interface 1202, memory 1203 and communication bus 1204, wherein processor 1201, communication interface 1202, storage Device 1203 completes mutual communication by communication bus 1204,

Memory 1203, for storing computer program；

Processor 1201 when for executing the program stored on memory 1203, realizes following steps:

Receive the target image with table；

The form image comprising table is determined from the target image；

The form image is identified according to the position of the line of text, obtains the table letter of the form image Breath.

The communication bus that above-mentioned electronic equipment is mentioned can be Peripheral Component Interconnect standard (Peripheral Component Interconnect, PCI) bus or expanding the industrial standard structure (Extended Industry Standard Architecture, EISA) bus etc..The communication bus can be divided into address bus, data/address bus, control bus etc..For just It is only indicated with a thick line in expression, figure, it is not intended that an only bus or a type of bus.

Communication interface is for the communication between above-mentioned electronic equipment and other equipment.

Memory may include random access memory (Random Access Memory, RAM), also may include non-easy The property lost memory (Non-Volatile Memory, NVM), for example, at least a magnetic disk storage.Optionally, memory may be used also To be storage device that at least one is located remotely from aforementioned processor.

Above-mentioned processor can be general processor, including central processing unit (Central Processing Unit, CPU), network processing unit (Network Processor, NP) etc.；It can also be digital signal processor (Digital Signal Processing, DSP), it is specific integrated circuit (Application Specific Integrated Circuit, ASIC), existing It is field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete Door or transistor logic, discrete hardware components.

Wherein, the form image is identified in the position according to the line of text, obtains the tabular drawing Before the step of form data of picture, the method also includes:

Remove all table lines of the form image；

Determine whether the table line of the form image is complete；

Wherein, the whether complete step of the table line of the determination form image, comprising:

Wherein, number of intersections and the step of the quantity of closed cell lattice in the form image after the determining removal character, Include:

Form image after the removal character carried out binary conversion treatment and carrying out negating processing to pixel value, is obtained Between image；

Wherein, the described the step of text image being partitioned into is identified, obtains the text information of the table, Include:

Wherein, described the step of determination includes the form image of table from the target image, comprising:

The embodiment of the invention also provides a kind of computer readable storage medium, the computer readable storage medium memory Computer program is contained, the computer program performs the steps of when being executed by processor

Receive the target image with table；

The form image comprising table is determined from the target image；

As it can be seen that when computer program is executed by processor, can receive first in scheme provided by the embodiment of the present invention Then target image with table determines the form image comprising table from target image, then carries out text to form image Current row detection, determines the position of line of text in form image, and then identify to described image according to the position of line of text, obtains To the form data of form image, wherein form data includes text information and tableau format information.Due to identifying obtained table Lattice information includes text information and tableau format information, rather than just the word content in table, therefore is improved in image Table recognition result diversity, be further processed using subsequent progress table recovery etc..

Remove all table lines of the form image；

Determine whether the table line of the form image is complete；

It should be noted that for above-mentioned apparatus, electronic equipment and computer readable storage medium embodiment, due to It is substantially similar to embodiment of the method, so being described relatively simple, related place is referring to the part explanation of embodiment of the method It can.

Need further exist for explanation, herein, relational terms such as first and second and the like be used merely to by One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged Except there is also other identical elements in the process, method, article or apparatus that includes the element.

Each embodiment in this specification is all made of relevant mode and describes, same and similar portion between each embodiment Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.

The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the scope of the present invention.It is all Any modification, equivalent replacement, improvement and so within the spirit and principles in the present invention, are all contained in protection scope of the present invention It is interior.

Claims

1. the recognition methods of form data in a kind of image, which is characterized in that the described method includes:

Receive the target image with table；

The form image comprising table is determined from the target image；

The form image is identified according to the position of the line of text, obtains the form data of the form image, In, the form data includes text information and tableau format information.

2. the method as described in claim 1, which is characterized in that in the position according to the line of text to the tabular drawing Before as the step of being identified, obtain the form data of the form image, the method also includes:

Remove all table lines of the form image；

The position according to the line of text identifies the form image, obtains the form data of the form image The step of, comprising:

Determine whether the table line of the form image is complete；

3. method according to claim 2, which is characterized in that whether the table line of the determination form image is complete Step, comprising:

If the quantity of the closed cell lattice is equal with the cell quantity, determine that the table line of the form image is complete It is whole；

If the quantity of the closed cell lattice and the cell quantity are unequal, the table line of the form image is determined not Completely.

4. method as claimed in claim 3, which is characterized in that number of intersections in the form image after the determining removal character And closed cell lattice quantity the step of, comprising:

Form image after the removal character is subjected to binary conversion treatment and pixel value is carried out to negate processing, obtains middle graph Picture；

5. method according to claim 2, which is characterized in that it is described that the text image being partitioned into is identified, it obtains To the table text information the step of, comprising:

According to the corresponding semanteme of each line of text, classifies to the Text region result, obtain each Text region knot The corresponding classification of fruit；

According to the corresponding classification of the Text region result, the Text region result is stored, the tabular drawing is obtained The text information of picture.

6. the method according to claim 1 to 5, which is characterized in that described determine from the target image includes table The step of form image of lattice, comprising:

The target image is inputted into the deep learning model that training is completed in advance, obtains the target of table in the target image Position；

7. the identification device of form data in a kind of image, which is characterized in that described device includes:

Target image receiving module, for receiving the target image with table；

Line of text position determination module determines the form image Chinese for carrying out line of text detection to the form image The position of current row；

Information identification module obtains the table for identifying according to the position of the line of text to the form image The form data of image, wherein the form data includes text information and tableau format information.

8. device as claimed in claim 7, which is characterized in that described device further include:

Table line removal module is obtained for identifying in the position according to the line of text to the form image Before the form data of the form image, all table lines of the form image are removed；

The information identification module includes:

Image segmentation unit is partitioned into text from the form image after removal table line for the position according to the line of text This image；

Word recognition unit obtains the text information of the form image for identifying to the text image that is partitioned into；

Table line completion unit, if the table line for the form image is imperfect, by the table line of the form image Completion；

Table recognition unit obtains the table of the form image for carrying out Table recognition to the complete form image of table line Lattice structural information.

9. device as claimed in claim 8, which is characterized in that the table line determining module includes:

Character removal unit removes the character in the form image for the position based on line of text in the form image；

First quantity determination unit, for determining the number of number of intersections and closed cell lattice in the form image after removing character Amount；

Second quantity determination unit, for determining the cell quantity of the table according to the number of intersections of the table line；

First table line determination unit determines if the quantity for the closed cell lattice is equal with the cell quantity The table line of the form image is complete；

Second table line determination unit, if unequal for the quantity of the closed cell lattice and the cell quantity, really The table line of the fixed form image is imperfect.

10. device as claimed in claim 9, which is characterized in that the first quantity determination unit includes:

Binary conversion treatment subelement, for by it is described removal character after form image carry out binary conversion treatment and to pixel value into Row negates processing, obtains intermediate image；

Table line separates subelement, separately handles for carrying out transverse direction and longitudinal direction table line to the expanding image, obtains horizontal line Image and vertical line image；

Table line image determines subelement, for the horizontal line image and the vertical line image carrying out that union is taken to handle, obtains Table line image；

Intersection point image determines subelement, for the horizontal line image and the vertical line image carrying out that intersection is taken to handle, is handed over Point image；

Number of intersections determines subelement, for determining number of hits in the form image after removing character according to the intersection point image Amount；

Cell quantity determines subelement, for being sealed in the form image after determining removal character according to the table line image Close the quantity of cell.

11. device as claimed in claim 8, which is characterized in that the word recognition unit includes:

Text region subelement obtains the form image for carrying out Text region to the text image being partitioned into Text region result；

Semantic analysis subelement obtains the corresponding semanteme of each line of text for carrying out semantic analysis to the Text region result；

Classify subelement, for classifying to the Text region result, obtaining according to the corresponding semanteme of each line of text The corresponding classification of each Text region result；

Recognition result storing sub-units are used for according to the corresponding classification of the Text region result, to the Text region result It is stored, obtains the text information of the form image.

12. such as the described in any item devices of claim 8-11, which is characterized in that the form image determining module includes:

Target position determination unit obtains institute for the target image to be inputted the deep learning model that training is completed in advance State the target position of table in target image；

Judging unit is distorted, for judging whether the corresponding table area in the target position distorts according to the target position；

Form image determination unit, if for the distortion of the corresponding table area in the target position, to the table area into The processing of row affine transformation, obtains the corresponding form image of the target image.

13. a kind of electronic equipment, which is characterized in that including processor, communication interface, memory and communication bus, wherein processing Device, communication interface, memory complete mutual communication by communication bus；

Memory, for storing computer program；

Processor when for executing the program stored on memory, realizes any method and step of claim 1-6.

14. a kind of computer readable storage medium, which is characterized in that be stored with computer in the computer readable storage medium Program realizes claim 1-6 any method and step when the computer program is executed by processor.