WO2021238446A1 - 文本识别方法、设备及存储介质 - Google Patents

文本识别方法、设备及存储介质 Download PDF

Info

Publication number
WO2021238446A1
WO2021238446A1 PCT/CN2021/086198 CN2021086198W WO2021238446A1 WO 2021238446 A1 WO2021238446 A1 WO 2021238446A1 CN 2021086198 W CN2021086198 W CN 2021086198W WO 2021238446 A1 WO2021238446 A1 WO 2021238446A1
Authority
WO
WIPO (PCT)
Prior art keywords
coordinates
track
points
point
value
Prior art date
Application number
PCT/CN2021/086198
Other languages
English (en)
French (fr)
Inventor
张欢欢
Original Assignee
京东方科技集团股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 京东方科技集团股份有限公司 filed Critical 京东方科技集团股份有限公司
Publication of WO2021238446A1 publication Critical patent/WO2021238446A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition
    • G06V30/32Digital ink
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/24Aligning, centring, orientation detection or correction of the image
    • G06V10/243Aligning, centring, orientation detection or correction of the image by compensating for image skew or non-uniform image deformations
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition
    • G06V30/24Character recognition characterised by the processing or recognition method
    • G06V30/242Division of the character sequences into groups prior to recognition; Selection of dictionaries
    • G06V30/244Division of the character sequences into groups prior to recognition; Selection of dictionaries using graphical properties, e.g. alphabet type or font
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10Character recognition

Definitions

  • This application relates to the field of text recognition technology, and in particular to a text recognition method, electronic equipment and storage medium.
  • Handwritten text recognition refers to the process of converting the orderly track points generated when writing text with a finger or a pen on a handwriting device with touch function into text.
  • Some embodiments of this application propose a text recognition method, including:
  • the scaled coordinates of all track points are processed in turn, the rounded coordinates of each track point are determined, and the track points whose rounded coordinates overlap are filtered out, and the average value is taken according to the scaled coordinates of the overlapped track points as a substitute
  • the first writing state value is used as the writing state value of the new track point, wherein the rounded coordinates are not
  • the zoom coordinates of the overlapping track points are directly used as the coordinates of the new track points;
  • the text recognition method of the embodiment of the present application corrects the oblique handwritten text to improve the quality of the track point coordinates, and processes multiple track points with the same coordinates after scaling and rounding into one track point, so as to reduce the loss of the track point. Therefore, it is possible to avoid the influence of too many track points on the time and efficiency of handwritten text recognition, shorten the time of text recognition, and improve the efficiency of text recognition.
  • Some embodiments of the present application provide an electronic device including a memory, a processor, and a computer program stored in the memory and capable of running on the processor.
  • the processor executes the program, the following text recognition method is implemented, including: :
  • the scaled coordinates of all track points are processed in turn, the rounded coordinates of each track point are determined, and the track points whose rounded coordinates overlap are filtered out, and the average value is taken according to the scaled coordinates of the overlapped track points as a substitute
  • the first writing state value is used as the writing state value of the new track point, wherein the rounded coordinates are not
  • the zoom coordinates of the overlapping track points are directly used as the coordinates of the new track points;
  • the electronic device of some embodiments of the present application corrects the oblique handwritten text to improve the quality of the track point coordinates, and processes multiple track points with the same coordinates after scaling and rounding into one track point to reduce the amount of track points. Therefore, it is possible to avoid the influence of too many track points on the time and efficiency of handwritten text recognition, shorten the time of text recognition, and improve the efficiency of text recognition.
  • Some embodiments of the present application propose a non-volatile computer-readable storage medium on which a computer program is stored.
  • the program is executed by a processor, the following text recognition method is implemented, including:
  • the scaled coordinates of all track points are processed in turn, the rounded coordinates of each track point are determined, and the track points whose rounded coordinates overlap are filtered out, and the average value is taken according to the scaled coordinates of the overlapped track points as a substitute
  • the first writing state value is used as the writing state value of the new track point, wherein the rounded coordinates are not
  • the zoom coordinates of the overlapping track points are directly used as the coordinates of the new track points;
  • the non-volatile computer-readable storage medium of the embodiment of the present application corrects the oblique handwritten text to improve the quality of the track point coordinates, and processes multiple track points with the same coordinates after scaling and rounding into one track point , In order to reduce the number of track points, which can avoid the influence of too many track points on the time and efficiency of handwritten text recognition, shorten the time of text recognition, and improve the efficiency of text recognition.
  • FIG. 1 is a schematic flowchart of a text recognition method provided by some embodiments of this application.
  • FIG. 3 is a schematic diagram of a network structure of an encoder provided by some embodiments of this application.
  • FIG. 1 is a schematic flowchart of a text recognition method provided by an embodiment of this application.
  • the writing track point is obtained from the touch display area of the terminal and provided to the processor, and the processor executes the recognition method.
  • the processor may be set in the terminal provided in the embodiment of the application, or Set in electronic devices such as servers in the cloud. Specific implementations of the terminal include terminal devices with handwriting recognition functions such as notebooks, mobile phones, conference machines, and educational machines.
  • the embodiment of the application provides a text recognition method, which can avoid the influence of too many track points on the time and efficiency of handwritten text recognition, shorten the time of text recognition, and improve the efficiency of text recognition.
  • the text recognition method It includes the following steps:
  • Step 101 Obtain the initial coordinates and writing state values of the track points generated when the handwritten text is written, where:
  • the stroke For each stroke of writing, the stroke has three stages: pen up, pen movement, and pen down.
  • the corresponding user's writing action is pressing pen or pen up
  • pen up and pen movement correspond to pen pressing
  • pen movement correspond to pen up.
  • the first track point in the pen-up phase is the starting point
  • the last track point in the pen-down phase is the end point.
  • the writing state value of the track point at the end point of each stroke is the first writing state value
  • the writing state value of the remaining track points of each stroke excluding the end point is the second writing state value.
  • the first writing state value is not equal to the second writing state value.
  • the writing state value can be distinguished by whether to write with the pen or by raising the pen when writing, and can also be distinguished by whether or not the pen is down in a stroke.
  • the writing state of the last trajectory point of the stroke is the pen-up writing
  • the writing state of the remaining trajectory points of the stroke is the pen writing
  • the writing state value of the pen-up writing can be the first writing state value
  • the writing state value of pen writing can be the second writing state value, for example, the first writing state value can be 1, and the second writing state value can be 0; of course, the first writing state value can also be 0, and the second writing state value can be 0.
  • the writing state value can be 1, which is not specifically limited.
  • the text may be Chinese characters or other characters.
  • a point set on the touch display area of the terminal may be used as the coordinate origin, and the set point may be any point.
  • a two-dimensional rectangular coordinate system is established with the horizontal direction of the touch display area as the x-axis and the vertical direction as the y-axis.
  • the initial coordinates include the abscissa and the ordinate.
  • the touch display area of the terminal also automatically obtains the writing state value of the track point generated when the handwritten text is written.
  • the embodiment of the present application obtains the handwritten text from the touch display area of the terminal when it is written. At the same time as the coordinates of the track point, the writing state value of the point is also obtained.
  • Step 102 Perform tilt correction processing on the handwritten text, and obtain the coordinates of the corrected track points.
  • skew correction processing of the handwritten text can be performed in the following manners, which specifically include:
  • x_max is the maximum value of the abscissa
  • x_min is the minimum value of the abscissa
  • y_max is the maximum value of the ordinate
  • y_min is the minimum value of the ordinate
  • Step 103 Perform scaling processing on the coordinates of all the trajectory points after correction in sequence, and determine the scaling coordinates of each trajectory point.
  • the scaled coordinates are the scaled coordinates of the trajectory points after correction, and the present application sequentially performs scaling processing on the coordinates of all trajectory points after correction according to the time sequence generated when the trajectory points are written.
  • the present application may perform scaling processing on the coordinates of all the track points after correction in a variety of ways, which is not specifically limited here.
  • the coordinates of all the track points after correction may be scaled in the following manner, which specifically includes:
  • the normal distribution of all trajectory points after correction in the longitudinal direction is obtained; wherein, the distribution of all trajectory points after correction in the longitudinal direction obeys the normal distribution, and the sample mean value of the ordinate is :
  • the sample variance is
  • the zoom ratio is obtained according to the length of the horizontal axis and the preset zoom height; among them, the preset zoom height h ref can be set according to the actual distribution of the track points, and cannot be set too small, otherwise the number of handwritten track points will be reduced too much, affecting Subsequent text recognition effect; it can not be set too large, otherwise the number of track points will not be reduced, and the effect of the number of track points will not be compressed.
  • the scaling ratio of the coordinates of the track point of the handwritten text line is:
  • Step 104 Perform rounding of the scaled coordinates of all track points in turn to determine the rounded coordinates of each track point, where,
  • the rounded coordinates are the coordinates after the zoomed coordinates are rounded.
  • the overlapped track points are averaged according to the scaled coordinates of the overlapped track points, which are used as the coordinates of the new track point to replace the overlapped track points.
  • the first writing state value is taken as the new The writing state value of the track point.
  • the scaled coordinates of the track point are directly used as the coordinates of the new track point, and the writing state value of the track point is directly used as the writing state value of the new track point.
  • the scaled coordinates of all track points are sequentially processed to integers
  • xf is the abscissa in the zoom coordinates
  • yf is the ordinate in the zoom coordinates
  • Step 105 Obtain the handwritten trajectory point characteristics according to the coordinates and writing state values of all new trajectory points.
  • the handwritten track point feature can be extracted in the following ways, which specifically include:
  • the writing state value of the new track point two characteristics of the writing state characterization value and the writing initial characterization value of the locus point are obtained.
  • the writing state characterization value of a track point if the locus point is located at the end of a stroke, the writing state characterization value is 0, and the writing state characterization value is 1 in other cases.
  • the writing start characterization value of a track point if the locus point is located at the beginning of a stroke, its writing start characterization value is 1, and in other cases, its writing start characterization value is 0.
  • a handwriting track point feature includes four dimensions: abscissa, ordinate, characterization value of writing state, and starting value of writing state.
  • Step 106 Recognize the text information of the handwritten track point feature to obtain the recognition result of the handwritten text.
  • handwritten trajectory point features can be recognized through a handwriting recognition neural network to obtain the recognition result of handwritten text.
  • the method of recognizing handwriting trajectory point features through the handwriting recognition neural network will be described in detail in the following embodiments, and will not be described here too much.
  • the oblique handwritten text is corrected to improve the quality of the track point coordinates, and multiple track points with the same coordinates after scaling and rounding are processed into one track point to reduce the track points Therefore, it is possible to avoid the influence of too many track points on the time and efficiency of handwritten text recognition, shorten the time of text recognition, and improve the efficiency of text recognition.
  • the method before performing the tilt correction processing on the handwritten text, the method further includes: filtering out track points with overlapping initial coordinates, and performing deduplication processing on the overlapping track points, wherein, when the writing state values of the overlapping track points are different , And use the first writing state value as the writing state value of the track point after deduplication.
  • the present application directly performs de-duplication processing on the track points with the same coordinates in the track points generated during writing, so as to reduce the number of track points.
  • this embodiment provides another text recognition method to illustrate how to recognize handwritten trajectory point features.
  • This embodiment and the previous embodiment have their own emphasis on the description content. The detailed steps can be referred to each other.
  • the text recognition method includes:
  • Step 201 Obtain the initial coordinates and writing state values of the trajectory points generated during the writing of the handwritten text, where the writing state value of the trajectory point corresponding to the end point of each stroke in the writing stroke is the first writing state value, and the writing stroke is in the first writing state value.
  • the writing state value of each stroke except the track point corresponding to the end point is the second writing state value.
  • Step 202 Perform tilt correction processing on the handwritten text, and obtain the coordinates of the corrected track points.
  • Step 203 Perform scaling processing on the coordinates of all the trajectory points after correction in sequence, and determine the scaling coordinates of each trajectory point.
  • Step 204 rounding coordinates of all track points are processed in turn, the rounding coordinates of each track point are determined, and the track points with overlapping rounding coordinates are filtered out, and the average value is taken according to the zooming coordinates of the overlapping track points, as Instead of the coordinates of the new track point of the overlapped track point, when the writing state value of the overlapped track point is different, the first writing state value is taken as the writing state value of the new track point, where the coordinates of the track point that do not overlap are rounded The zoom coordinates are directly used as the coordinates of the new track point.
  • Step 205 Obtain the handwriting track point characteristics according to the coordinates and writing state values of all new track points.
  • steps 201-205 can refer to the explanations of steps 101-105, in order to avoid redundancy, details are not described here.
  • Step 206 Recognize the handwritten trajectory point features through the handwritten recognition neural network to obtain the recognition result of the handwritten text.
  • the handwriting recognition neural network includes an encoder and a decoder, as follows:
  • the network structure includes an output layer, a hidden layer, and a linear transformation layer.
  • the hidden layer includes a long and short-term memory network and a random working layer (Dropout), and the output of the long and short-term memory network is used as the input of the random working layer.
  • the long and short-term memory network and the random working layer are set as a group, and the hidden layer of the network structure can include multiple groups of long and short-term memory networks and random working layers.
  • some implementations include a first long-short-term memory network, a first random working layer, a second long-short-term memory network, a second random working layer, and a linear transformation layer that are sequentially arranged, in which one LSTM and one random
  • the number of working layers is one group, and the number of groups is variable and can be adjusted according to the actual situation.
  • Some embodiments include four groups, and the long and short-term memory network and the random working layer are arranged at intervals in sequence.
  • LSTM can be set to a bidirectional network
  • the hidden layer can be set to n layers, for example, to 128 layers.
  • Linea linear transformation layer
  • the encoder needs to be trained, and the loss function used for model training is the connection time series classification loss function.
  • the input data of the input layer of the encoder is handwritten trajectory point features. Assuming that the recognizable character type is N and the length of the input handwritten trajectory point feature is M, the output of the encoder is a two-dimensional matrix value of M*N. For a certain row of the matrix, each output value represents the probability value of recognizing the output as each character in a time interval.
  • the tag value of a character is generally an integer from 1 to N.
  • the input of the decoder is the output of the encoder, that is, a two-dimensional matrix of M*N, and the output is a sequence of recognized character mark values.
  • the specific identification process is as follows:
  • the output of the encoder calculate the character mark value with the highest probability of recognizing the output character in each time interval; traverse in sequence according to the time interval, and merge the same character mark value output in each time interval into one output character mark value , And remove the empty characters in the character mark value to obtain the recognized character mark value.
  • the text information is obtained according to the corresponding relationship between the recognized character tag value and the character, and the recognition result of the handwritten text is obtained.
  • the text recognition method of the embodiment of the present application corrects the oblique handwritten text to improve the quality of the track point coordinates, and processes multiple track points with the same coordinates after scaling and rounding into one track point, so as to reduce the loss of the track point. Therefore, it is possible to avoid the influence of too many track points on the time and efficiency of handwritten text recognition, shorten the time of text recognition, and improve the efficiency of text recognition.
  • this application also proposes an electronic device.
  • the electronic device includes a memory, a processor, and a computer program that is stored on the memory and can run on the processor.
  • the processor executes the program, the following text recognition method is implemented, including:
  • the scaled coordinates of all track points are processed in turn, the rounded coordinates of each track point are determined, and the track points whose rounded coordinates overlap are filtered out, and the average value is taken according to the scaled coordinates of the overlapped track points as a substitute for overlapping The coordinates of the new track point of the track point.
  • the first writing state value is used as the writing state value of the new track point.
  • the scaled coordinates of the track points that are not overlapped by rounding the coordinates are directly As the coordinates of the new track point;
  • performing tilt correction processing on the handwritten text and obtaining the coordinates of the corrected track point includes:
  • x_max is the maximum value of the abscissa
  • x_min is the minimum value of the abscissa
  • y_max is the maximum value of the ordinate
  • y_min is the minimum value of the ordinate
  • the coordinates of all the track points after correction are sequentially scaled, including
  • the coordinates of all the track points after correction are sequentially zoomed.
  • it further includes:
  • the handwriting recognition neural network is used to recognize handwritten trajectory point features to obtain the recognition result of handwritten text.
  • the handwritten recognition neural network includes:
  • the encoder includes a first long-short-term memory network, a first random working layer, a second long-short-term memory network, a second random working layer, and a linear transformation layer, which are used to encode handwritten trajectory point features to obtain a two-dimensional matrix value, Among them, each column value of each row of the two-dimensional matrix value represents the probability value of the recognition output as each character at the time step.
  • the handwriting recognition neural network further includes:
  • the decoder is used to calculate the character mark value with the highest probability of recognizing the output character at each time step. According to the sequence of the time step, the output of the time step with the same character mark value is merged into one character mark value, and the character mark is removed The null character in the value to get a sequence of character token values.
  • it further includes:
  • the character sequence is recognized according to the corresponding relationship between the character tag value sequence and the character.
  • the handwritten track point features are obtained according to the coordinates and writing state values of all new track points, including:
  • the method before performing the tilt correction processing on the handwritten text, the method further includes:
  • the track points with overlapping initial coordinates are filtered out, and the overlapping track points are deduplicated.
  • the first writing state value is used as the writing state value of the track point after deduplication.
  • the electronic device of the embodiment of the present application corrects the oblique handwritten text to improve the quality of the track point coordinates, and processes multiple track points with the same coordinates after scaling and rounding into one track point to reduce the number of track points In this way, the influence of too many track points on the time and efficiency of handwritten text recognition can be avoided, the time of text recognition can be shortened, and the efficiency of text recognition can be improved.
  • this application also proposes a non-volatile computer-readable storage medium.
  • the non-volatile computer-readable storage medium has a computer program stored thereon, and when the program is executed by a processor, the following text recognition method is realized, including:
  • the scaled coordinates of all track points are processed in turn, the rounded coordinates of each track point are determined, and the track points whose rounded coordinates overlap are filtered out, and the average value is taken according to the scaled coordinates of the overlapped track points as a substitute for overlapping The coordinates of the new track point of the track point.
  • the first writing state value is used as the writing state value of the new track point.
  • the scaled coordinates of the track points that are not overlapped by rounding the coordinates are directly As the coordinates of the new track point;
  • performing tilt correction processing on the handwritten text and obtaining the coordinates of the corrected track point includes:
  • x_max is the maximum value of the abscissa
  • x_min is the minimum value of the abscissa
  • y_max is the maximum value of the ordinate
  • y_min is the minimum value of the ordinate
  • the coordinates of all the track points after correction are sequentially scaled, including
  • the coordinates of all the track points after correction are sequentially zoomed.
  • it further includes:
  • the handwriting recognition neural network is used to recognize handwritten trajectory point features to obtain the recognition result of handwritten text.
  • the handwritten recognition neural network includes:
  • the encoder includes a first long-short-term memory network, a first random working layer, a second long-short-term memory network, a second random working layer, and a linear transformation layer, which are used to encode handwritten trajectory point features to obtain a two-dimensional matrix value, Among them, each column value of each row of the two-dimensional matrix value represents the probability value of the recognition output as each character at the time step.
  • the handwriting recognition neural network further includes:
  • the decoder is used to calculate the character mark value with the highest probability of recognizing the output character at each time step. According to the sequence of the time step, the output of the time step with the same character mark value is merged into one character mark value, and the character mark is removed The null character in the value to get a sequence of character token values.
  • it further includes:
  • the character sequence is recognized according to the corresponding relationship between the character tag value sequence and the character.
  • the handwritten track point features are obtained according to the coordinates and writing state values of all new track points, including:
  • the method before performing the tilt correction processing on the handwritten text, the method further includes:
  • the track points with overlapping initial coordinates are filtered out, and the overlapping track points are deduplicated.
  • the first writing state value is used as the writing state value of the track point after deduplication.
  • the non-volatile computer-readable storage medium of the embodiment of the present application corrects the oblique handwritten text to improve the quality of the track point coordinates, and processes multiple track points with the same coordinates after scaling and rounding into one track point , In order to reduce the number of track points, which can avoid the influence of too many track points on the time and efficiency of handwritten text recognition, shorten the time of text recognition, and improve the efficiency of text recognition.
  • first and second are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with “first” and “second” may explicitly or implicitly include at least one of the features. In the description of the present application, "a plurality of” means at least two, such as two, three, etc., unless specifically defined otherwise.
  • a "computer-readable medium” can be any device that can contain, store, communicate, propagate, or transmit a program for use by an instruction execution system, device, or device or in combination with these instruction execution systems, devices, or devices.
  • computer-readable media include the following: electrical connections (electronic devices) with one or more wiring, portable computer disk cases (magnetic devices), random access memory (RAM), Read only memory (ROM), erasable and editable read only memory (EPROM or flash memory), fiber optic devices, and portable compact disk read only memory (CDROM).
  • the computer-readable medium may even be paper or other suitable medium on which the program can be printed, because it can be used for example by optically scanning the paper or other medium, followed by editing, interpretation or other suitable media if necessary. The program is processed in a way to obtain the program electronically and then stored in the computer memory.
  • each part of this application can be implemented by hardware, software, firmware, or a combination thereof.
  • multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system.
  • Discrete logic gate circuits for implementing logic functions on data signals
  • Logic circuits application specific integrated circuits with suitable combinational logic gates
  • PGA programmable gate arrays
  • FPGA field programmable gate arrays
  • each functional unit in each embodiment of the present application may be integrated into one processing module, or each unit may exist alone physically, or two or more units may be integrated into one module.
  • the above-mentioned integrated modules can be implemented in the form of hardware or software function modules. If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it may also be stored in a computer readable storage medium.
  • the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Character Discrimination (AREA)

Abstract

一种文本识别方法、设备及存储介质,其中,方法包括:获取手写体文本在书写时产生的轨迹点的初始坐标和书写状态值;对手写体文本进行倾斜校正处理,然后依次进行缩放处理和取整数处理,确定每个轨迹点的缩放坐标和取整坐标,并筛选出取整坐标重叠的轨迹点,将重叠的轨迹点的缩放坐标的均值作为新轨迹点的坐标,取整坐标未重叠的轨迹点的缩放坐标直接作为新轨迹点的坐标;根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征(105);识别手写轨迹点特征的文本信息,以得到手写体文本的识别结果(106),提高识别效率。

Description

文本识别方法、设备及存储介质 技术领域
本申请涉及文本识别技术领域,尤其涉及一种文本识别方法、电子设备及存储介质。
背景技术
手写体文本识别是指将在具有触控功能的手写设备上通过手指或笔书写文本时产生的有序轨迹点转化为文本的过程。
然而,即使书写同一文本,比如书写同一文字时,不同书写风格、不同的具有触控功能的手写设备产生的轨迹点数量可能不同,一旦轨迹点数量过多,将会增加手写体文本识别的时间。
发明内容
该部分公开的内容用于提供部分本申请的实施例方式,不用于限制本申请的发明内容。详细的实施方式在后面具体实施方式部分描述。
本申请一些实施例提出了一种文本识别方法,包括:
获取手写体文本在书写时产生的轨迹点的初始坐标和书写状态值,其中,书写笔画中的每一笔画的终点对应的轨迹点的书写状态值为第一书写状态值,书写笔画中的每一笔画的除所述终点对应的轨迹点的其余点的书写状态值为第二书写状态值;
对所述手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标;
对校正后的所有轨迹点的坐标依次进行缩放处理,确定每个轨迹点的缩放坐标;
对所有轨迹点的缩放坐标依次进行取整数处理,确定每个轨迹点的取整坐标,并筛选出取整坐标重叠的轨迹点,并根据所述重叠的轨迹点的缩放坐标取均值,作为替代所述重叠的轨迹点的新轨迹点的坐标,在所述重叠的轨迹点的书写状态值不同时,将第一书写状态值作为所述新轨迹点的书写状态值,其中,取整坐标未重叠的轨迹点的缩放坐标直接作为新轨迹点的坐标;
根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征;
识别所述手写轨迹点特征的文本信息,以得到所述手写体文本的识别结果。
本申请实施例的文本识别方法,对倾斜手写体文本进行校正,以提高轨迹点坐标的质量,并将缩放及取整后具有相同坐标的多个轨迹点处理为一个轨迹点,以减少轨迹点的数量,从而可以避免轨迹点数量过多对手写体文本识别时间和效率的影响,缩短文本识别的 时间,并提高文本识别的效率。
本申请一些实施例提供一种电子设备,包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序,所述处理器执行所述程序时,以实现如下文本识别方法,包括:
获取手写体文本在书写时产生的轨迹点的初始坐标和书写状态值,其中,书写笔画中的每一笔画的终点对应的轨迹点的书写状态值为第一书写状态值,书写笔画中的每一笔画的除所述终点对应的轨迹点的其余点的书写状态值为第二书写状态值;
对所述手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标;
对校正后的所有轨迹点的坐标依次进行缩放处理,确定每个轨迹点的缩放坐标;
对所有轨迹点的缩放坐标依次进行取整数处理,确定每个轨迹点的取整坐标,并筛选出取整坐标重叠的轨迹点,并根据所述重叠的轨迹点的缩放坐标取均值,作为替代所述重叠的轨迹点的新轨迹点的坐标,在所述重叠的轨迹点的书写状态值不同时,将第一书写状态值作为所述新轨迹点的书写状态值,其中,取整坐标未重叠的轨迹点的缩放坐标直接作为新轨迹点的坐标;
根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征;
识别所述手写轨迹点特征的文本信息,以得到所述手写体文本的识别结果。
本申请一些实施例的电子设备,对倾斜手写体文本进行校正,以提高轨迹点坐标的质量,并将缩放及取整后具有相同坐标的多个轨迹点处理为一个轨迹点,以减少轨迹点的数量,从而可以避免轨迹点数量过多对手写体文本识别时间和效率的影响,缩短文本识别的时间,并提高文本识别的效率。
本申请一些实施例提出了一种非易失性计算机可读存储介质,其上存储有计算机程序,所述程序被处理器执行时,实现如下的文本识别方法,包括:
获取手写体文本在书写时产生的轨迹点的初始坐标和书写状态值,其中,书写笔画中的每一笔画的终点对应的轨迹点的书写状态值为第一书写状态值,书写笔画中的每一笔画的除所述终点对应的轨迹点的其余点的书写状态值为第二书写状态值;
对所述手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标;
对校正后的所有轨迹点的坐标依次进行缩放处理,确定每个轨迹点的缩放坐标;
对所有轨迹点的缩放坐标依次进行取整数处理,确定每个轨迹点的取整坐标,并筛选出取整坐标重叠的轨迹点,并根据所述重叠的轨迹点的缩放坐标取均值,作为替代所述重叠的轨迹点的新轨迹点的坐标,在所述重叠的轨迹点的书写状态值不同时,将第一书写状态值作为所述新轨迹点的书写状态值,其中,取整坐标未重叠的轨迹点的缩放坐标直接作为新轨迹点的坐标;
根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征;
识别所述手写轨迹点特征的文本信息,以得到所述手写体文本的识别结果。
本申请实施例的非易失性计算机可读存储介质,对倾斜手写体文本进行校正,以提高轨迹点坐标的质量,并将缩放及取整后具有相同坐标的多个轨迹点处理为一个轨迹点,以减少轨迹点的数量,从而可以避免轨迹点数量过多对手写体文本识别时间和效率的影响,缩短文本识别的时间,并提高文本识别的效率。
本申请附加的方面和优点将在下面的描述中部分给出,部分将从下面的描述中变得明显,或通过本申请的实践了解到。
附图说明
本申请上述的和/或附加的方面和优点从下面结合附图对实施例的描述中将变得明显和容易理解,其中:
图1为本申请一些实施例所提供的一种文本识别方法的流程示意图;
图2为本申请一些实施例所提供的另一种文本识别方法的流程示意图;
图3为本申请一些实施例所提供的一种编码器的网络结构示意图。
具体实施方式
下面详细描述本申请构思的不同实施例,以便更清楚的理解本申请的申请构思。所述实施例的示例在附图中示出,其中自始至终相同或类似的标号表示相同或类似的元件或具有相同或类似功能的元件。下面通过参考附图描述的实施例是示例性的,旨在用于解释本申请,而不能理解为对本申请的限制。
下面参考附图描述本申请实施例的文本识别方法。
图1为本申请实施例所提供的一种文本识别方法的流程示意图。
在进行手写体文本识别时,即使书写同一文本,比如书写同一文字时(如一个单词、一个汉字等),不同书写风格、不同的具有触控功能的手写设备获取的轨迹点数量可能不同,一旦轨迹点数量过多,将会增加手写体文本的识别时间,降低识别效率。
本申请实施例提供的文本识别方法,书写轨迹点从终端的触控显示区域上获取并提供给处理器,由处理器执行识别方法,该处理器可以设置在本申请实施例提供的终端,或设置在云端的服务器等电子设备中。终端的具体实施方式包括如书写本、手机、会议机、教育机等具有手写识别功能的终端设备。
本申请实施例提供了文本识别方法,可以避免轨迹点数量过多对手写体文本识别时间和效率的影响,缩短文本识别的时间,并提高文本识别的效率,如图1所示,该文本识别 方法包括以下步骤:
步骤101,获取手写体文本在书写时产生的轨迹点的初始坐标和书写状态值,其中,
针对书写的每一笔画,该笔画有起笔、运笔和落笔三个阶段,对应用户的书写动作为按笔或抬笔,起笔和运笔对应按笔,运笔对应抬笔。
起笔阶段的第一个轨迹点为起点,落笔阶段的最后一个轨迹点为终点。每一笔画的终点的轨迹点的书写状态值为第一书写状态值,每一笔画的除去终点的其余轨迹点的书写状态值为第二书写状态值。第一书写状态值不等于第二书写状态值。
其中,书写状态值可以通过书写时是按笔书写还是抬笔书写区别,也可以用一笔画中的是否落笔阶段区别。比如,在书写一个笔画时,笔画最后轨迹点的书写状态即为抬笔书写,该笔画的其余轨迹点的书写状态即为按笔书写;抬笔书写的书写状态值可以为第一书写状态值,按笔书写的书写状态值可以为第二书写状态值,例如,第一书写状态值可以为1,第二书写状态值可以为0;当然也可以第一书写状态值可以为0,第二书写状态值可以为1,对此不作具体限定。
比如,在具体应用时,以用户在终端的触控显示区域书写文本为例,文本可以为汉字等文字。其中,本申请实施例可以以终端的触控显示区域上设定一点作为坐标原点,该设定点可以是任一点。以触控显示区域的水平方向为x轴,以竖直方向为y轴建立二维直角坐标系。初始坐标包括横坐标和纵坐标,当用户通过手或笔等触控工具在触控显示区域书写时,终端获取书写对应的轨迹点的初始坐标,本申请实施例可以从终端的触控显示区域上获取手写体文本在书写时产生的轨迹点的初始坐标。
另外,当轨迹点产生时,终端的触控显示区域还自动获取手写体文本在书写时产生的轨迹点的书写状态值,本申请实施例从终端的触控显示区域上获取手写体文本在书写时产生的轨迹点的坐标的同时,也获取该点的书写状态值。
步骤102,对手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标。
本申请可以通过多种方式对手写体文本进行倾斜校正处理,在此不做具体限定。在一些实施方式中,可以通过以下方式对手写体文本进行倾斜校正处理,具体包括:
获取所有轨迹点的初始坐标的横坐标集合和纵坐标集合,并通过最小二乘直线拟合计算手写体文本的倾斜角度;比如,可以使用numpy的polyfit函数计算倾斜角度,a,b=numpy.polyfit(X,Y,1),ang=a*180/3.1415926,X为横坐标集合,Y为纵坐标集合;
计算手写体文本的中心点坐标,其中,中心点的横坐标mx=(x_max-x_min)*0.5,x_max为横坐标最大值,x_min为横坐标最小值;中心点的纵坐标my=(y_max-y_min)*0.5,y_max为纵坐标最大值,y_min为纵坐标最小值;
根据倾斜角度旋转手写体文本的中心点坐标得到旋转后的中心点坐标,其中,旋转后 的中心点的横坐标cx和纵坐标cy相同,cx=cy=0.5*sqrt((y_max-y_min)*(y_max-y_min)+(x_max-x_min)*(x_max-x_min));
根据倾斜角度、手写体文本的中心点坐标和旋转后的中心点坐标对所有轨迹点的初始坐标进行校正,其中,校正后的所有轨迹点的横坐标x=(j-x_min-mx)*cos(ang)+(i-y_min-my)*sin(ang)+cx,校正后的所有轨迹点的纵坐标y=-(j-x_min-mx)*cos(ang)+(i-y_min-my)*sin(ang)+cy,i为轨迹点的初始横坐标,j为轨迹点的初始纵坐标,ang为倾斜角度。
步骤103,对校正后的所有轨迹点的坐标依次进行缩放处理,确定每个轨迹点的缩放坐标。
缩放坐标为校正后的轨迹点缩放后的坐标,本申请根据轨迹点在书写时产生的时间顺序对校正后的所有轨迹点的坐标依次进行缩放处理。其中,本申请可以通过多种方式对校正后的所有轨迹点的坐标进行缩放处理,在此不做具体限定。
在一些实施方式中,可以通过以下方式对对校正后的所有轨迹点的坐标进行缩放处理,具体包括:
根据校正后的所有轨迹点的纵坐标得到校正后的所有轨迹点在纵向上的正态分布;其中,校正后的所有轨迹点在纵向上的分布服从正态分布,其纵坐标的样本均值为:
Figure PCTCN2021086198-appb-000001
样本方差为
Figure PCTCN2021086198-appb-000002
获取手写体文本的图像,确定图像高度为正态分布对应分位点的横轴长度;假设该手写体文本行图像的高度对应着正态分布2.58分位点时的横轴长度,手写体文本行图像高度标记为h new,h new=2.58*2*s=s/0.19,即
Figure PCTCN2021086198-appb-000003
s样本标准差,其中,也可以假设该手写体文本行图像的高度对应着正态分布其他分位点时的横轴长度,可以根据需要缩放的情况进行设定,在此不做具体限定;
根据横轴长度和预设缩放高度得到缩放比;其中,预设缩放高度h ref可以根据轨迹点实际分布的情况进行设置,不能设置得过小,否则会使手写轨迹点数量减少过多,影响后续文字识别效果;也不可以设置得过大,否则会使轨迹点数量没有任何减少,起不到轨迹点数量压缩的效果。例如,当h new=2.58*2*s=s/0.19,手写体文本行轨迹点坐标的缩放比例为:
Figure PCTCN2021086198-appb-000004
根据缩放比依次缩放校正后的所有轨迹点的坐标;其缩放坐标的横坐标xf=(j-xr_min)*r,缩放坐标的纵坐标yf=(i-yr_min)*r,i为校正后的所有轨迹点的坐标中的横坐标,j为校正后的所有轨迹点的坐标中的纵坐标,xr_min为校正后的所有轨迹点的坐标中 的横坐标最小值,yr_min为校正后的所有轨迹点的坐标中的纵坐标最小值,r为缩放比。
步骤104,对所有轨迹点的缩放坐标依次进行取整数处理,确定每个轨迹点的取整坐标,其中,
取整坐标为对缩放坐标取整后的坐标,所有轨迹点的取整坐标可能存在一个坐标对应多个轨迹点的情况,即存在整坐标重叠的多个轨迹点,本申请筛选出取整坐标重叠的轨迹点,并根据重叠的轨迹点的缩放坐标取均值,作为替代重叠的轨迹点的新轨迹点的坐标,在重叠的轨迹点的书写状态值不同时,将第一书写状态值作为新轨迹点的书写状态值。
对于取整坐标未重叠的轨迹点,直接该轨迹点的缩放坐标作为新轨迹点的坐标,该轨迹点的书写状态值直接作为新轨迹点的书写状态值。
其中,对所有轨迹点的缩放坐标依次进行取整数处理,所述取整坐标中的横坐标为xi=int(xf+0.5),所述取整坐标中的纵坐标为yi=int(yf+0.5),xf为所述缩放坐标中的横坐标,yf为所述缩放坐标中的纵坐标
步骤105,根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征。
本申请可以通过多种方式提取手写轨迹点特征,在此不做具体限定。在一些实施方式中,可以通过以下方式提取手写轨迹点特征,具体包括:
根据新轨迹点的书写状态值得到该轨迹点的书写状态表征值和书写起始表征值两个特征。对于一个轨迹点的书写状态表征值,如果该轨迹点位于一个笔画的终点,则其书写状态表征值为0,其余情况其书写状态表征值为1。对于一个轨迹点的书写起始表征值,如果该轨迹点位于一个笔画的起点,则其书写起始表征值为1,其余情况其书写起始表征值为0。一个手写轨迹点特征包括横坐标、纵坐标、书写状态表征值和书写状态起始值4个维度。由此,可以根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征。
步骤106,识别手写轨迹点特征的文本信息,以得到手写体文本的识别结果。
本申请可以通过多种方式识别手写轨迹点特征的文本信息,在此不做具体限定,在本申请一些实施例中,可以通过手写体识别神经网络识别手写轨迹点特征,从而得到手写体文本的识别结果。其中,通过手写体识别神经网络识别手写轨迹点特征的方式将在下面实施例进行详细说明,在此不做过多描述。
根据本申请实施例的文本识别方法,对倾斜手写体文本进行校正,以提高轨迹点坐标的质量,并将缩放及取整后具有相同坐标的多个轨迹点处理为一个轨迹点,以减少轨迹点的数量,从而可以避免轨迹点数量过多对手写体文本识别时间和效率的影响,缩短文本识别的时间,并提高文本识别的效率。
在一些实施方式中,对手写体文本进行倾斜校正处理之前,还包括:筛选出初始坐标重叠的轨迹点,对重叠的轨迹点进行去重处理,其中,在重叠的轨迹点的书写状态值不同 时,将第一书写状态值作为去重后轨迹点的书写状态值。
可以理解的是,对于手写体文本在书写时产生的所有轨迹点的初始坐标可能存在一个坐标对应多个轨迹点的情况,直接进行去重处理,从而在一个坐标处仅保留一个轨迹点。在多个轨迹点中有一个轨迹点为第一书写状态值时,将第一书写状态值作为去重后轨迹点的写状态值。如果多个轨迹点的写状态值相同,比如,均为第二写状态值,则直接将第二写状态值作为重后轨迹点的写状态值。由此,本申请对书写时产生的轨迹点中存在坐标相同的轨迹点直接进行去重处理,以减少轨迹点数量。
基于上一实施例,本实施例提供了另一种文本识别方法用以说明如何识别手写轨迹点特征,本实施例和上一实施例在描述内容上各有侧重,各实施例之间对于未尽述步骤可相互参考。本实施例中,如图2所示,该文本识别方法包括:
步骤201,获取手写体文本在书写时产生的轨迹点的初始坐标和书写状态值,其中,书写笔画中的每一笔画的终点对应的轨迹点的书写状态值为第一书写状态值,书写笔画中的每一笔画的除终点对应的轨迹点的其余点的书写状态值为第二书写状态值。
步骤202,对手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标。
步骤203,对校正后的所有轨迹点的坐标依次进行缩放处理,确定每个轨迹点的缩放坐标。
步骤204,对所有轨迹点的缩放坐标依次进行取整数处理,确定每个轨迹点的取整坐标,并筛选出取整坐标重叠的轨迹点,并根据重叠的轨迹点的缩放坐标取均值,作为替代重叠的轨迹点的新轨迹点的坐标,在重叠的轨迹点的书写状态值不同时,将第一书写状态值作为新轨迹点的书写状态值,其中,取整坐标未重叠的轨迹点的缩放坐标直接作为新轨迹点的坐标。
步骤205,根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征。
其中,步骤201-205可以参见步骤101-105的解释,为避免冗余,在此不做赘述。
步骤206,通过手写体识别神经网络识别手写轨迹点特征,以得到手写体文本的识别结果。
其中,手写体识别神经网络包括编码器和解码器,具体如下:
(1)构建基于LSTM(Long Short-Term Memory,长短期记忆网络)的编码器,网络结构包括输出层、隐藏层和线性变换层。
隐藏层包括长短期记忆网络和随机工作层(Dropout),长短期记忆网络的输出作为随机工作层的输入。长短期记忆网络和随机工作层作为一组,网络结构的隐藏层可以包括多组长短期记忆网络和随机工作层。
如图3所示,一些实施方式中包括依次设置的第一长短期记忆网络、第一随机工作层、第二长短期记忆网络、第二随机工作层和线性变换层,其中一个LSTM和一个随机工作层的层数是一组,组数是可变的,可根据实际情况进行调节。一些实施方式中包括四组,长短期记忆网络和随机工作层依次间隔设置。
其中LSTM可设置为双向网路,隐藏层可设置为n层,比如,设置为128层。最后连接一个线性变换层(Linea)。编码器需要训练,模型训练所用的损失函数为连接时序分类损失函数。编码器的输入层的输入数据为手写轨迹点特征,假设可识别的字符种类为N,输入的手写轨迹点特征的长度为M,则编码器的输出为M*N的二维矩阵值。对于矩阵的某一行,每个输出值表示在一个时间间隔识别输出为各个字符的概率值。字符的标记值一般为从1到N的整数。
(2)解码器:不需要训练。可采用贪心算法基于最大概率进行解码。其中,贪心算法是指在对问题求解时,总是做出在当前看来是最好的选择;也就是说,不从整体最优上加以考虑,算法得到的是在某种意义上的局部最优解。
解码器的输入为编码器的输出,即M*N的二维矩阵,输出为识别出的字符标记值序列。具体识别过程如下:
根据编码器的输出,计算每个时间间隔识别输出字符概率值最大的字符标记值;按照时间的先后顺序,依次遍历,将每个时间间隔输出的相同字符标记值合并为1个输出字符标记值,并去除字符标记值中的空字符,以得到识别出的字符标记值。
然后,根据识别出的字符标记值与字符对应关系得到文本信息,从而得到手写体文本的识别结果。
本申请实施例的文本识别方法,对倾斜手写体文本进行校正,以提高轨迹点坐标的质量,并将缩放及取整后具有相同坐标的多个轨迹点处理为一个轨迹点,以减少轨迹点的数量,从而可以避免轨迹点数量过多对手写体文本识别时间和效率的影响,缩短文本识别的时间,并提高文本识别的效率。
为了实现上述实施例,本申请还提出一种电子设备。
该电子设备包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序,处理器执行程序时,以实现如下的文本识别方法,包括:
获取手写体文本在书写时产生的轨迹点的初始坐标和书写状态值,其中,书写笔画中的每一笔画的终点对应的轨迹点的书写状态值为第一书写状态值,书写笔画中的每一笔画的除终点对应的轨迹点的其余点的书写状态值为第二书写状态值;
对手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标;
对校正后的所有轨迹点的坐标依次进行缩放处理,确定每个轨迹点的缩放坐标;
对所有轨迹点的缩放坐标依次进行取整数处理,确定每个轨迹点的取整坐标,并筛选出取整坐标重叠的轨迹点,并根据重叠的轨迹点的缩放坐标取均值,作为替代重叠的轨迹点的新轨迹点的坐标,在重叠的轨迹点的书写状态值不同时,将第一书写状态值作为新轨迹点的书写状态值,其中,取整坐标未重叠的轨迹点的缩放坐标直接作为新轨迹点的坐标;
根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征;以及
识别手写轨迹点特征的文本信息,以得到手写体文本的识别结果。
在一些实施方式中,对手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标,包括:
获取所有轨迹点的初始坐标的横坐标集合和纵坐标集合,并通过最小二乘直线拟合计算手写体文本的倾斜角度;
计算手写体文本的中心点坐标,其中,中心点的横坐标mx=(x_max-x_min)*0.5,x_max为横坐标最大值,x_min为横坐标最小值;中心点的纵坐标my=(y_max-y_min)*0.5,y_max为纵坐标最大值,y_min为纵坐标最小值;
根据倾斜角度旋转手写体文本的中心点坐标得到旋转后的中心点坐标,其中,旋转后的中心点的横坐标cx和纵坐标cy相同,cx=cy=0.5*sqrt((y_max-y_min)*(y_max-y_min)+(x_max-x_min)*(x_max-x_min));
根据倾斜角度、手写体文本的中心点坐标和旋转后的中心点坐标对所有轨迹点的初始坐标进行校正,其中,校正后的所有轨迹点的横坐标x=(j-x_min-mx)*cos(ang)+(i-y_min-my)*sin(ang)+cx,校正后的所有轨迹点的纵坐标y=-(j-x_min-mx)*cos(ang)+(i-y_min-my)*sin(ang)+cy,i为轨迹点的初始横坐标,j为轨迹点的初始纵坐标,ang为倾斜角度。
在一些实施方式中,对校正后的所有轨迹点的坐标依次进行缩放处理,包括
根据校正后的所有轨迹点的纵坐标得到校正后的所有轨迹点在纵向上的正态分布;
获取手写体文本的图像,确定图像高度为正态分布对应分位点的横轴长度;
根据横轴长度和预设缩放高度得到缩放比;
根据缩放比依次缩放校正后的所有轨迹点的坐标。
在一些实施方式中,其中,
缩放坐标的横坐标xf=(j-xr_min)*r,缩放坐标的纵坐标yf=(i-yr_min)*r,i为校正后的所有轨迹点的坐标中的横坐标,j为校正后的所有轨迹点的坐标中的纵坐标,xr_min为校正后的所有轨迹点的坐标中的横坐标最小值,yr_min为校正后的所有轨迹点的坐标中的纵坐标最小值,r为缩放比。
在一些实施方式中,其中,
取整坐标中的横坐标xi=int(xf+0.5),xf为缩放坐标中的横坐标;取整坐标中的纵坐标yi=int(yf+0.5),yf为缩放坐标中的纵坐标。
在一些实施方式中,还包括:
通过手写体识别神经网络识别手写轨迹点特征,以得到手写体文本的识别结果,其中,手写体识别神经网络包括:
编码器,编码器包括第一长短期记忆网络、第一随机工作层、第二长短期记忆网络、第二随机工作层和线性变换层,用于对手写轨迹点特征编码得到二维矩阵值,其中,二维矩阵值每行的各个列值表示在该时间步识别输出为各个字符的概率值。
在一些实施方式中,手写体识别神经网络还包括:
解码器,用于计算每个时间步识别输出字符概率值最大的字符标记值,按照时间步的先后顺序,对于字符标记值相同的时间步的输出合并为1个字符标记值,并去除字符标记值中的空字符,以得到字符标记值序列。
在一些实施方式中,还包括:
根据字符标记值序列与字符对应关系识别得到字符序列。
在一些实施方式中,根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征,包括:
在一些实施方式中,对手写体文本进行倾斜校正处理之前,还包括:
筛选出初始坐标重叠的轨迹点,对重叠的轨迹点进行去重处理,其中,在重叠的轨迹点的书写状态值不同时,将第一书写状态值作为去重后轨迹点的书写状态值。
需要说明的是,前述对文本识别方法实施例的解释说明也适用于该实施例的电子设备,此处不再赘述。
本申请实施例的电子设备,对倾斜手写体文本进行校正,以提高轨迹点坐标的质量,并将缩放及取整后具有相同坐标的多个轨迹点处理为一个轨迹点,以减少轨迹点的数量,从而可以避免轨迹点数量过多对手写体文本识别时间和效率的影响,缩短文本识别的时间,并提高文本识别的效率。
为了实现上述实施例,本申请还提出一种非易失性计算机可读存储介质。
该非易失性计算机可读存储介质,其上存储有计算机程序,程序被处理器执行时,以实现如下的文本识别方法,包括:
获取手写体文本在书写时产生的轨迹点的初始坐标和书写状态值,其中,书写笔画中的每一笔画的终点对应的轨迹点的书写状态值为第一书写状态值,书写笔画中的每一笔画的除终点对应的轨迹点的其余点的书写状态值为第二书写状态值;
对手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标;
对校正后的所有轨迹点的坐标依次进行缩放处理,确定每个轨迹点的缩放坐标;
对所有轨迹点的缩放坐标依次进行取整数处理,确定每个轨迹点的取整坐标,并筛选出取整坐标重叠的轨迹点,并根据重叠的轨迹点的缩放坐标取均值,作为替代重叠的轨迹点的新轨迹点的坐标,在重叠的轨迹点的书写状态值不同时,将第一书写状态值作为新轨迹点的书写状态值,其中,取整坐标未重叠的轨迹点的缩放坐标直接作为新轨迹点的坐标;
根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征;以及
识别手写轨迹点特征的文本信息,以得到手写体文本的识别结果。
在一些实施方式中,对手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标,包括:
获取所有轨迹点的初始坐标的横坐标集合和纵坐标集合,并通过最小二乘直线拟合计算手写体文本的倾斜角度;
计算手写体文本的中心点坐标,其中,中心点的横坐标mx=(x_max-x_min)*0.5,x_max为横坐标最大值,x_min为横坐标最小值;中心点的纵坐标my=(y_max-y_min)*0.5,y_max为纵坐标最大值,y_min为纵坐标最小值;
根据倾斜角度旋转手写体文本的中心点坐标得到旋转后的中心点坐标,其中,旋转后的中心点的横坐标cx和纵坐标cy相同,cx=cy=0.5*sqrt((y_max-y_min)*(y_max-y_min)+(x_max-x_min)*(x_max-x_min));
根据倾斜角度、手写体文本的中心点坐标和旋转后的中心点坐标对所有轨迹点的初始坐标进行校正,其中,校正后的所有轨迹点的横坐标x=(j-x_min-mx)*cos(ang)+(i-y_min-my)*sin(ang)+cx,校正后的所有轨迹点的纵坐标y=-(j-x_min-mx)*cos(ang)+(i-y_min-my)*sin(ang)+cy,i为轨迹点的初始横坐标,j为轨迹点的初始纵坐标,ang为倾斜角度。
在一些实施方式中,对校正后的所有轨迹点的坐标依次进行缩放处理,包括
根据校正后的所有轨迹点的纵坐标得到校正后的所有轨迹点在纵向上的正态分布;
获取手写体文本的图像,确定图像高度为正态分布对应分位点的横轴长度;
根据横轴长度和预设缩放高度得到缩放比;
根据缩放比依次缩放校正后的所有轨迹点的坐标。
在一些实施方式中,其中,
缩放坐标的横坐标xf=(j-xr_min)*r,缩放坐标的纵坐标yf=(i-yr_min)*r,i为校正后的所有轨迹点的坐标中的横坐标,j为校正后的所有轨迹点的坐标中的纵坐标,xr_min为校正后的所有轨迹点的坐标中的横坐标最小值,yr_min为校正后的所有轨迹点的坐标中的纵坐标最小值,r为缩放比。
在一些实施方式中,其中,
取整坐标中的横坐标xi=int(xf+0.5),xf为缩放坐标中的横坐标;取整坐标中的纵坐标yi=int(yf+0.5),yf为缩放坐标中的纵坐标。
在一些实施方式中,还包括:
通过手写体识别神经网络识别手写轨迹点特征,以得到手写体文本的识别结果,其中,手写体识别神经网络包括:
编码器,编码器包括第一长短期记忆网络、第一随机工作层、第二长短期记忆网络、第二随机工作层和线性变换层,用于对手写轨迹点特征编码得到二维矩阵值,其中,二维矩阵值每行的各个列值表示在该时间步识别输出为各个字符的概率值。
在一些实施方式中,手写体识别神经网络还包括:
解码器,用于计算每个时间步识别输出字符概率值最大的字符标记值,按照时间步的先后顺序,对于字符标记值相同的时间步的输出合并为1个字符标记值,并去除字符标记值中的空字符,以得到字符标记值序列。
在一些实施方式中,还包括:
根据字符标记值序列与字符对应关系识别得到字符序列。
在一些实施方式中,根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征,包括:
在一些实施方式中,对手写体文本进行倾斜校正处理之前,还包括:
筛选出初始坐标重叠的轨迹点,对重叠的轨迹点进行去重处理,其中,在重叠的轨迹点的书写状态值不同时,将第一书写状态值作为去重后轨迹点的书写状态值。
需要说明的是,前述对文本识别方法实施例的解释说明也适用于该实施例的非易失性计算机可读存储介质,此处不再赘述。
本申请实施例的非易失性计算机可读存储介质,对倾斜手写体文本进行校正,以提高轨迹点坐标的质量,并将缩放及取整后具有相同坐标的多个轨迹点处理为一个轨迹点,以减少轨迹点的数量,从而可以避免轨迹点数量过多对手写体文本识别时间和效率的影响,缩短文本识别的时间,并提高文本识别的效率。
在本说明书的描述中,参考术语“一个实施例”、“一些实施例”、“示例”、“具体示例”、或“一些示例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或者特点包含于本申请的至少一个实施例或示例中。在本说明书中,对上述术语的示意性表述不必须针对的是相同的实施例或示例。而且,描述的具体特征、结构、材料或者特点可以在任一个或多个实施例或示例中以合适的方式结合。此外,在不相互矛盾的情况下,本领域的技术人员可以将本说明书中描述的不同实施例或示例以及不同实施例或示例的特征进行结合 和组合。
此外,术语“第一”、“第二”仅用于描述目的,而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此,限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。在本申请的描述中,“多个”的含义是至少两个,例如两个,三个等,除非另有明确具体的限定。
流程图中或在此以其他方式描述的任何过程或方法描述可以被理解为,表示包括一个或更多个用于实现定制逻辑功能或过程的步骤的可执行指令的代码的模块、片段或部分,并且本申请的优选实施方式的范围包括另外的实现,其中可以不按所示出或讨论的顺序,包括根据所涉及的功能按基本同时的方式或按相反的顺序,来执行功能,这应被本申请的实施例所属技术领域的技术人员所理解。
在流程图中表示或在此以其他方式描述的逻辑和/或步骤,例如,可以被认为是用于实现逻辑功能的可执行指令的定序列表,可以具体实现在任何计算机可读介质中,以供指令执行系统、装置或设备(如基于计算机的系统、包括处理器的系统或其他可以从指令执行系统、装置或设备取指令并执行指令的系统)使用,或结合这些指令执行系统、装置或设备而使用。就本说明书而言,"计算机可读介质"可以是任何可以包含、存储、通信、传播或传输程序以供指令执行系统、装置或设备或结合这些指令执行系统、装置或设备而使用的装置。计算机可读介质的更具体的示例(非穷尽性列表)包括以下:具有一个或多个布线的电连接部(电子装置),便携式计算机盘盒(磁装置),随机存取存储器(RAM),只读存储器(ROM),可擦除可编辑只读存储器(EPROM或闪速存储器),光纤装置,以及便携式光盘只读存储器(CDROM)。另外,计算机可读介质甚至可以是可在其上打印所述程序的纸或其他合适的介质,因为可以例如通过对纸或其他介质进行光学扫描,接着进行编辑、解译或必要时以其他合适方式进行处理来以电子方式获得所述程序,然后将其存储在计算机存储器中。
应当理解,本申请的各部分可以用硬件、软件、固件或它们的组合来实现。在上述实施方式中,多个步骤或方法可以用存储在存储器中且由合适的指令执行系统执行的软件或固件来实现。如,如果用硬件来实现和在另一实施方式中一样,可用本领域公知的下列技术中的任一项或他们的组合来实现:具有用于对数据信号实现逻辑功能的逻辑门电路的离散逻辑电路,具有合适的组合逻辑门电路的专用集成电路,可编程门阵列(PGA),现场可编程门阵列(FPGA)等。
本技术领域的普通技术人员可以理解实现上述实施例方法携带的全部或部分步骤是可以通过程序来指令相关的硬件完成,所述的程序可以存储于一种计算机可读存储介质中,该程序在执行时,包括方法实施例的步骤之一或其组合。
此外,在本申请各个实施例中的各功能单元可以集成在一个处理模块中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个模块中。上述集成的模块既可以采用硬件的形式实现,也可以采用软件功能模块的形式实现。所述集成的模块如果以软件功能模块的形式实现并作为独立的产品销售或使用时,也可以存储在一个计算机可读取存储介质中。
上述提到的存储介质可以是只读存储器,磁盘或光盘等。尽管上面已经示出和描述了本申请的实施例,可以理解的是,上述实施例是示例性的,不能理解为对本申请的限制,本领域的普通技术人员在本申请的范围内可以对上述实施例进行变化、修改、替换和变型。

Claims (15)

  1. 一种文本识别方法,其特征在于,包括:
    获取手写体文本在书写时产生的轨迹点的初始坐标和书写状态值,其中,书写笔画中的每一笔画的终点对应的轨迹点的书写状态值为第一书写状态值,书写笔画中的每一笔画的除所述终点对应的轨迹点的其余点的书写状态值为第二书写状态值;
    对所述手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标;
    对校正后的所有轨迹点的坐标依次进行缩放处理,确定每个轨迹点的缩放坐标;
    对所有轨迹点的缩放坐标依次进行取整数处理,确定每个轨迹点的取整坐标,并筛选出取整坐标重叠的轨迹点,并根据所述重叠的轨迹点的缩放坐标取均值,作为替代所述重叠的轨迹点的新轨迹点的坐标,在所述重叠的轨迹点的书写状态值不同时,将第一书写状态值作为所述新轨迹点的书写状态值,其中,取整坐标未重叠的轨迹点的缩放坐标直接作为新轨迹点的坐标;
    根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征;以及
    识别所述手写轨迹点特征的文本信息。
  2. 根据权利要求1所述的方法,其特征在于,所述对所述手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标,包括:
    获取所有轨迹点的初始坐标的横坐标集合和纵坐标集合,计算所述手写体文本的倾斜角度;
    计算所述手写体文本的中心点坐标,其中,所述中心点的横坐标mx=(x_max-x_min)*0.5,x_max为横坐标最大值,x_min为横坐标最小值;所述中心点的纵坐标my=(y_max-y_min)*0.5,y_max为纵坐标最大值,y_min为纵坐标最小值;
    根据所述倾斜角度旋转所述手写体文本的中心点坐标得到旋转后的中心点坐标,其中,所述旋转后的中心点的横坐标cx和纵坐标cy相同,cx=cy=0.5*sqrt((y_max-y_min)*(y_max-y_min)+(x_max-x_min)*(x_max-x_min));
    根据所述倾斜角度、所述手写体文本的中心点坐标和所述旋转后的中心点坐标对所有轨迹点的初始坐标进行校正,其中,校正后的所有轨迹点的横坐标x=(j-x_min-mx)*cos(ang)+(i-y_min-my)*sin(ang)+cx,所述校正后的所有轨迹点的纵坐标y=-(j-x_min-mx)*cos(ang)+(i-y_min-my)*sin(ang)+cy,i为轨迹点的初始横坐标,j为轨迹点的初始纵坐标,ang为所述倾斜角度。
  3. 根据权利要求1或2所述的方法,其特征在于,所述对校正后的所有轨迹点的坐标依次进行缩放处理,包括
    根据校正后的所有轨迹点的纵坐标得到所述校正后的所有轨迹点在纵向上的正态分布;
    获取所述手写体文本的图像,确定所述图像高度为所述正态分布对应分位点的横轴长度;
    根据所述横轴长度和预设缩放高度得到缩放比;
    根据所述缩放比依次缩放所述校正后的所有轨迹点的坐标。
  4. 根据权利要求3所述的方法,其特征在于,其中,
    所述缩放坐标的横坐标xf=(j-xr_min)*r,所述缩放坐标的纵坐标yf=(i-yr_min)*r,i为所述校正后的所有轨迹点的坐标中的横坐标,j为所述校正后的所有轨迹点的坐标中的纵坐标,xr_min为所述校正后的所有轨迹点的坐标中的横坐标最小值,yr_min为校正后的所有轨迹点的坐标中的纵坐标最小值,r为所述缩放比。
  5. 根据权利要求1所述的方法,其特征在于,其中,
    对所有轨迹点的缩放坐标依次进行取整数处理,所述取整坐标中的横坐标为xi=int(xf+0.5),所述取整坐标中的纵坐标为yi=int(yf+0.5),xf为所述缩放坐标中的横坐标,yf为所述缩放坐标中的纵坐标。
  6. 根据权利要求1所述的方法,其特征在于,还包括:
    通过手写体识别神经网络识别所述手写轨迹点特征,以得到所述手写体文本的识别结果,其中,所述手写体识别神经网络包括:
    编码器,所述编码器包括第一长短期记忆网络、第一随机工作层、第二长短期记忆网络、第二随机工作层和线性变换层,用于对所述手写轨迹点特征编码得到二维矩阵值,其中,所述二维矩阵值每行的各个列值表示在该时间步识别输出为各个字符的概率值。
  7. 根据权利要求6所述的方法,其特征在于,所述手写体识别神经网络还包括:
    解码器,用于计算每个时间步识别输出字符概率值最大的字符标记值,按照时间步的先后顺序,对于所述字符标记值相同的时间步的输出合并为1个字符标记值,并去除字符标记值中的空字符,以得到字符标记值序列。
  8. 根据权利要求7所述的方法,其特征在于,还包括:
    根据所述字符标记值序列与字符对应关系识别得到字符序列。
  9. 一种电子设备,其特征在于,包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序,所述处理器执行所述程序时,以实现如下的文本识别方法,包括:
    获取手写体文本在书写时产生的轨迹点的初始坐标和书写状态值,其中,书写笔画中的每一笔画的终点对应的轨迹点的书写状态值为第一书写状态值,书写笔画中的每一笔画的除所述终点对应的轨迹点的其余点的书写状态值为第二书写状态值;
    对所述手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标;
    对校正后的所有轨迹点的坐标依次进行缩放处理,确定每个轨迹点的缩放坐标;
    对所有轨迹点的缩放坐标依次进行取整数处理,确定每个轨迹点的取整坐标,并筛选出取整坐标重叠的轨迹点,并根据所述重叠的轨迹点的缩放坐标取均值,作为替代所述重叠的轨迹点的新轨迹点的坐标,在所述重叠的轨迹点的书写状态值不同时,将第一书写状态值作为所述新轨迹点的书写状态值,其中,取整坐标未重叠的轨迹点的缩放坐标直接作为新轨迹点的坐标;
    根据所有新轨迹点的坐标和书写状态值得到手写轨迹点特征;以及
    识别所述手写轨迹点特征的文本信息,以得到所述手写体文本的识别结果。
  10. 根据权利要求9所述的电子设备,其特征在于,所述对所述手写体文本进行倾斜校正处理,并获取校正后的轨迹点的坐标,包括:
    获取所有轨迹点的初始坐标的横坐标集合和纵坐标集合,并通过最小二乘直线拟合计算所述手写体文本的倾斜角度;
    计算所述手写体文本的中心点坐标,其中,所述中心点的横坐标mx=(x_max-x_min)*0.5,x_max为横坐标最大值,x_min为横坐标最小值;所述中心点的纵坐标my=(y_max-y_min)*0.5,y_max为纵坐标最大值,y_min为纵坐标最小值;
    根据所述倾斜角度旋转所述手写体文本的中心点坐标得到旋转后的中心点坐标,其中,所述旋转后的中心点的横坐标cx和纵坐标cy相同,cx=cy=0.5*sqrt((y_max-y_min)*(y_max-y_min)+(x_max-x_min)*(x_max-x_min));
    根据所述倾斜角度、所述手写体文本的中心点坐标和所述旋转后的中心点坐标对所有轨迹点的初始坐标进行校正,其中,校正后的所有轨迹点的横坐标x=(j-x_min-mx)*cos(ang)+(i-y_min-my)*sin(ang)+cx,所述校正后的所有轨迹点的纵坐标y=-(j-x_min-mx)*cos(ang)+(i-y_min-my)*sin(ang)+cy,i为轨迹点的初始横坐标,j为轨迹点的初始纵坐标,ang为所述倾斜角度。
  11. 根据权利要求9或10所述的电子设备,其特征在于,所述对校正后的所有轨迹点的坐标依次进行缩放处理,包括
    根据校正后的所有轨迹点的纵坐标得到所述校正后的所有轨迹点在纵向上的正态分布;
    获取所述手写体文本的图像,确定所述图像高度为所述正态分布对应分位点的横轴长度;
    根据所述横轴长度和预设缩放高度得到缩放比;
    根据所述缩放比依次缩放所述校正后的所有轨迹点的坐标。
  12. 根据权利要求11所述的电子设备,其特征在于,其中,
    所述缩放坐标的横坐标xf=(j-xr_min)*r,所述缩放坐标的纵坐标yf=(i-yr_min)*r,i为所述校正后的所有轨迹点的坐标中的横坐标,j为所述校正后的所有轨迹点的坐标中的纵坐标,xr_min为所述校正后的所有轨迹点的坐标中的横坐标最小值,yr_min为校正后的所有轨迹点的坐标中的纵坐标最小值,r为所述缩放比。
  13. 根据权利要求9所述的电子设备,其特征在于,还包括:
    通过手写体识别神经网络识别所述手写轨迹点特征,以得到所述手写体文本的识别结果,其中,所述手写体识别神经网络包括:
    编码器,所述编码器包括第一长短期记忆网络、第一随机工作层、第二长短期记忆网络、第二随机工作层和线性变换层,用于对所述手写轨迹点特征编码得到二维矩阵值,其中,所述二维矩阵值每行的各个列值表示在该时间步识别输出为各个字符的概率值。
  14. 根据权利要求13所述的电子设备,其特征在于,所述手写体识别神经网络还包括:
    解码器,用于计算每个时间步识别输出字符概率值最大的字符标记值,按照时间步的先后顺序,对于所述字符标记值相同的时间步的输出合并为1个字符标记值,并去除字符标记值中的空字符,以得到字符标记值序列。
  15. 一种非易失性计算机可读存储介质,其上存储有计算机程序,其特征在于,所述程序被处理器执行时,实现如权利要求1-9任一项所述的文本识别方法。
PCT/CN2021/086198 2020-05-29 2021-04-09 文本识别方法、设备及存储介质 WO2021238446A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010476181.0 2020-05-29
CN202010476181.0A CN111626238B (zh) 2020-05-29 2020-05-29 文本识别方法、电子设备及存储介质

Publications (1)

Publication Number Publication Date
WO2021238446A1 true WO2021238446A1 (zh) 2021-12-02

Family

ID=72271122

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/086198 WO2021238446A1 (zh) 2020-05-29 2021-04-09 文本识别方法、设备及存储介质

Country Status (2)

Country Link
CN (1) CN111626238B (zh)
WO (1) WO2021238446A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114463760A (zh) * 2022-04-08 2022-05-10 华南理工大学 一种基于双流编码的文字图像书写轨迹恢复方法
CN115758476A (zh) * 2022-11-11 2023-03-07 重庆傲雄在线信息技术有限公司 基于手写签名的电子名章制作方法、设备及介质
WO2024103292A1 (zh) * 2022-11-16 2024-05-23 京东方科技集团股份有限公司 手写体识别方法、手写体识别模型的训练方法及装置

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111626238B (zh) * 2020-05-29 2023-08-04 京东方科技集团股份有限公司 文本识别方法、电子设备及存储介质
WO2022160619A1 (zh) * 2021-02-01 2022-08-04 京东方科技集团股份有限公司 手写体识别方法及装置、手写体识别系统和交互平板
WO2022222096A1 (zh) * 2021-04-22 2022-10-27 京东方科技集团股份有限公司 手绘图形识别方法、装置和系统,以及计算机可读存储介质
CN113807295B (zh) * 2021-09-24 2023-10-27 科大讯飞股份有限公司 手写识别方法、装置、电子设备和存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120020566A1 (en) * 2010-07-26 2012-01-26 Casio Computer Co., Ltd. Character recognition device and recording medium
CN105117054A (zh) * 2015-08-12 2015-12-02 珠海优特电力科技股份有限公司 一种手写输入的识别方法及系统
CN108921155A (zh) * 2018-04-23 2018-11-30 新疆大学 一种联机手写维吾尔文单词倾斜矫正方法
CN109147004A (zh) * 2018-08-20 2019-01-04 广州视源电子科技股份有限公司 笔迹显示方法、装置、设备及存储介质
CN111626238A (zh) * 2020-05-29 2020-09-04 京东方科技集团股份有限公司 文本识别方法、电子设备及存储介质

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108596167B (zh) * 2018-04-19 2020-08-21 上海亿锎智能科技有限公司 一种变长光学字符文字快速识别系统及方法、装置
CN108764155B (zh) * 2018-05-30 2021-10-12 新疆大学 一种手写维吾尔文单词切分识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120020566A1 (en) * 2010-07-26 2012-01-26 Casio Computer Co., Ltd. Character recognition device and recording medium
CN105117054A (zh) * 2015-08-12 2015-12-02 珠海优特电力科技股份有限公司 一种手写输入的识别方法及系统
CN108921155A (zh) * 2018-04-23 2018-11-30 新疆大学 一种联机手写维吾尔文单词倾斜矫正方法
CN109147004A (zh) * 2018-08-20 2019-01-04 广州视源电子科技股份有限公司 笔迹显示方法、装置、设备及存储介质
CN111626238A (zh) * 2020-05-29 2020-09-04 京东方科技集团股份有限公司 文本识别方法、电子设备及存储介质

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114463760A (zh) * 2022-04-08 2022-05-10 华南理工大学 一种基于双流编码的文字图像书写轨迹恢复方法
CN115758476A (zh) * 2022-11-11 2023-03-07 重庆傲雄在线信息技术有限公司 基于手写签名的电子名章制作方法、设备及介质
CN115758476B (zh) * 2022-11-11 2024-02-09 重庆亲笔签数字科技有限公司 基于手写签名的电子名章制作方法、设备及介质
WO2024103292A1 (zh) * 2022-11-16 2024-05-23 京东方科技集团股份有限公司 手写体识别方法、手写体识别模型的训练方法及装置

Also Published As

Publication number Publication date
CN111626238A (zh) 2020-09-04
CN111626238B (zh) 2023-08-04

Similar Documents

Publication Publication Date Title
WO2021238446A1 (zh) 文本识别方法、设备及存储介质
US7349576B2 (en) Method, device and computer program for recognition of a handwritten character
US7302099B2 (en) Stroke segmentation for template-based cursive handwriting recognition
US7369702B2 (en) Template-based cursive handwriting recognition
KR102463657B1 (ko) 다중 오브젝트 구조를 인식하기 위한 시스템 및 방법
US20150294184A1 (en) Pattern recognition based on information integration
WO2021196013A1 (zh) 单词识别方法、设备及存储介质
US8494277B2 (en) Handwritten character recognition based on frequency variations in characters
US11474627B2 (en) Method, apparatus and device for erasing handwriting on electronic whiteboard
KR20220038477A (ko) 텍스트 라인 추출
CN115461792A (zh) 手写文本识别方法、装置和系统,手写文本搜索方法和系统,以及计算机可读存储介质
CN115589786A (zh) 手绘图形识别方法、装置和系统,以及计算机可读存储介质
CN111492407A (zh) 用于绘图美化的系统和方法
JP2019194788A (ja) 学習装置、認識装置、学習方法及びコンピュータプログラム
US20050089225A1 (en) Method for aligning gesture features of image
WO2021220589A1 (ja) インクデータの修正方法、情報処理装置、及びプログラム
JP2761679B2 (ja) オンライン手書き文字認識装置
WO2022180725A1 (ja) 文字認識装置、プログラム及び方法
JP6437208B2 (ja) 手書き音楽記号認識装置及び手書き音楽記号認識プログラム
JP2001184457A (ja) オンライン手書き文字認識装置、方法及びオンライン手書き文字認識プログラムを記録した記録媒体
CN112596603A (zh) 核电站控制系统的手势操控方法、装置、设备及存储介质
JPH04353964A (ja) 文書作成装置
JPH012186A (ja) 手書き文字認識方式
JPH09269974A (ja) オンライン文字認識装置
CN110647245A (zh) 基于dtw算法的手写输入法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21812426

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21812426

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 21812426

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 29.06.2023)