WO2011074067A1 - 文字認識方法、文字認識装置および文字認識プログラム - Google Patents
文字認識方法、文字認識装置および文字認識プログラム Download PDFInfo
- Publication number
- WO2011074067A1 WO2011074067A1 PCT/JP2009/070886 JP2009070886W WO2011074067A1 WO 2011074067 A1 WO2011074067 A1 WO 2011074067A1 JP 2009070886 W JP2009070886 W JP 2009070886W WO 2011074067 A1 WO2011074067 A1 WO 2011074067A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- character
- type
- handwriting
- image
- score
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/18—Extraction of features or characteristics of the image
- G06V30/18105—Extraction of features or characteristics of the image related to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/24—Character recognition characterised by the processing or recognition method
- G06V30/242—Division of the character sequences into groups prior to recognition; Selection of dictionaries
- G06V30/244—Division of the character sequences into groups prior to recognition; Selection of dictionaries using graphical properties, e.g. alphabet type or font
- G06V30/2455—Discrimination between machine-print, hand-print and cursive writing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Definitions
- the present invention relates to a character recognition method, a character recognition device, and a character recognition program, and more particularly to a character recognition method, a character recognition device, and a character recognition program for discriminating characters printed on forms used in financial institutions and the like and characters entered by handwriting. About.
- a character recognition device that recognizes characters on a form is used to automate the input of characters printed and entered on the form.
- this character recognition device in order to improve the accuracy of character recognition, a definition body that defines in what order the headings to be printed on the form and the corresponding data are written in what order is created in advance. The character string is recognized based on the definition information.
- the form includes type characters printed in advance and handwritten characters written by the user, and the definition body is set to indicate whether the character recognition target item is handwritten or typed.
- the character recognition device refers to the definition body, uses the handwriting character recognition engine if the character recognition target item is set to handwriting, and sets the character recognition engine for printing if the character setting is set. Use to recognize characters.
- the character string is extracted from the image data of the document, the center position in the height direction of each character is calculated, whether the handwritten character or the printed character is determined based on the regularity of the center position, and character recognition is performed based on the determination result.
- a method of performing this has been proposed (see, for example, Patent Document 1).
- Patent Document 1 the regularity of the center position varies when the character string includes a cloud point or a roar. For this reason, there is also known a method of extracting characters from a character string without including mazes and stuttering, and determining whether the characters are handwritten characters or printed characters from the regularity of the center positions of all extracted characters (for example, Patent Documents). 2).
- As the feature amount density uniformity, pixel value variation, character stroke linearity, character height, width uniformity, character line width uniformity, and the like are used.
- any of the conventional character recognition methods it is determined only whether the character string of the character recognition target item is a handwritten character or a printed character. For this reason, even in a method of performing character recognition using a definition body, it is impossible to deal with a character recognition target item in which both handwritten and typed characters may be entered. Therefore, at the time of character recognition, since both the handwriting and the print type character recognition engines are called, there is a problem that processing time is required.
- the present invention has been made in view of the above points, and has improved the accuracy of handwriting / type discrimination in character recognition processing, shortening the processing time by character recognition, and improving the character recognition accuracy, and the character
- An object is to provide a recognition device and a character recognition program.
- an item to be recognized is specified from the image data of the form. Calculating the position and size of the image character of the identified recognition target item, analyzing the characteristics of the handwriting and type of the image character, calculating the score of the probability of handwriting or type, and calculating the score based on the score. Whether the image character is handwritten or typed, calculates the average score of each character in the recognition target item to check the mixture of handwritten and typed, and whether the character of each recognition target item is handwritten or typed Character recognition is performed using a character recognition engine in accordance with a determination result and a mixed handwriting and mixed character check result.
- a first score obtained by analyzing the gradation distribution of the di-character and quantifying the probability of handwriting or type is calculated, and a second score obtained by analyzing the character color of the image character and quantifying the probability of hand-writing or type Calculating the ratio of the gradation of the inflection point at which the character stroke changes from the character stroke extracted from the image character and the gradation of the character stroke other than the inflection point to determine the certainty of handwriting or type
- a third score that is digitized is calculated, and a fourth score is calculated by analyzing the character stroke edge strength representing the intensity of the change in shading at the boundary between the character stroke and the background, and quantifying the probability of handwriting or type.
- the fifth score is calculated by analyzing the vertical size and center position of the character from the position and size of the image character to quantify the probability of handwriting or type. Analyzing the character size and pitch from the position and size of the image character to calculate a sixth score obtained by quantifying the probability of handwriting or type, and determining whether the image character is handwritten or type, There is provided a character recognition method characterized in that the first to sixth scores are summed up in consideration of weights set in accordance with their importance.
- the gradation distribution of the image character of the recognition target item specified from the image data of the form is analyzed to check whether the handwritten or printed character is correct.
- a tone distribution analysis unit that calculates a first score that quantifies the likelihood; a character color analysis unit that calculates a second score that analyzes the character color of the image character to quantify the likelihood of handwriting or type;
- the ratio of the inflection point gradation at which the character stroke changes from the character stroke extracted from the image character and the gradation of the character stroke other than the inflection point are analyzed to quantify the probability of handwriting or type.
- Character stroke gradation ratio analysis unit for calculating a score, and a character stroke edge strength representing the intensity of shade change at the boundary between the character stroke and the background
- a character stroke edge strength analysis unit that calculates a fourth score obtained by quantifying the probability of handwriting or type by analyzing, and analyzing the vertical size and center position of the character from the position and size of the image character
- a character size / center position analysis unit for calculating a fifth score in which the probability is quantified, and a sixth score in which the character size and pitch are analyzed from the position and size of the image character to quantify the probability of handwriting or type
- a character size / pitch analysis unit for calculating the handwriting
- a handwriting / printing mixed checking unit for calculating an average value of the first to sixth scores of each character in the recognition target item and checking a mixture of handwriting and type A total score obtained by counting the first to sixth scores in consideration of weights set according to importance, and the handwriting Based on the check result by the mixed character check unit, the handwriting / printing discriminating unit that
- a character recognition program for causing a computer to execute a process for discriminating between handwritten characters and printed characters on a form, and the computer is configured to recognize items from image data of the form. Identifying the position and size of the image character of the identified recognition target item, analyzing the gradation distribution of the image character to calculate a first score that quantifies the probability of handwriting or type, A second score obtained by analyzing the character color of the image character to quantify the probability of handwriting or type is calculated, and the gradation and inflection of the inflection point where the character stroke changes from the character stroke extracted from the image character A third score is calculated by analyzing a ratio with the gradation of a character stroke other than a point, and quantifying the probability of handwriting or type.
- a fourth score is calculated by analyzing the character stroke edge intensity representing the intensity of the change in shading at the boundary between the roke and the background, and quantifying the probability of handwriting or type, and from the position and size of the image character, Analyzing the size and center position to calculate the 5th score that digitizes the likelihood of handwriting or type, and analyzing the character size and pitch from the position and size of the image character to quantify the probability of handwriting or type
- the sixth score is calculated, and the first to sixth scores are counted in consideration of the weight set according to the importance, thereby determining whether the image character is a handwritten character or a printed character. Calculates the average score of each character, checks for mixed handwriting and type, and determines whether each recognition target character is handwritten or typed. And handwriting and in accordance with the printing type of the mixed results of the check performed for character recognition, the character recognition program for causing to execute a process is provided.
- a character recognition method a character recognition device, and a character recognition program
- a combination of a plurality of analysis results and determination of handwriting or type by a mixed handwriting and type check are performed, and a character recognition engine corresponding to the determination result
- the processing time for character recognition is shortened and the character recognition accuracy is improved.
- the character recognition method, the character recognition device, and the character recognition program configured as described above determine whether the character described in the character recognition item is handwritten or printed by combining a plurality of analyzes, the accuracy of the determination is increased.
- the handwritten character can be recognized as a handwritten character, and the typed character can be recognized as a typed character, thereby improving the character recognition accuracy.
- handwriting / type determination is performed for each character, so that handwritten characters can be recognized as handwritten characters and type characters can be recognized as type characters. Character recognition accuracy is improved.
- a handwritten character string can be excluded from the headline extraction target by utilizing the function of the handwriting / type discrimination unit. This leads to an improvement in the performance of the process of extracting the headline.
- a transfer request document used in a financial institution or the like is used as a target form, and a device that recognizes a character string printed and handwritten on the transfer request document The case where this is applied will be described in detail with reference to the drawings.
- FIG. 1 is a block diagram showing a character recognition apparatus according to an embodiment of the present invention.
- the character recognition device 10 includes a control unit 11 that controls the entire device, and a scanner control unit 12, a character recognition position specifying unit 13, a handwriting / type discrimination unit 14 and a recognition result display unit 15 are connected to the control unit 11.
- the handwriting / type discrimination unit 14 is connected to a character recognition processing unit 16, and the character recognition processing unit 16 is associated with a handwritten character recognition dictionary 17 and a type character recognition dictionary 18 referred to in the character recognition processing. Yes.
- the handwriting / type discrimination unit 14 is an element having a function of discriminating whether the character of the recognition target item is handwritten or type. Based on the discrimination result, the character recognition processing unit 16 selects an appropriate character recognition engine. Character recognition processing using is possible. That is, the handwriting / type discrimination unit 14 includes a tone distribution analysis unit 19, a character color analysis unit 20, a character stroke tone ratio analysis unit 21, a character stroke edge strength analysis unit 22, and a character size / center position. An analysis unit 23, a character size / pitch analysis unit 24, and a handwritten / printed mixed check unit 25 are provided.
- the gradation distribution analysis unit 19 uses characteristics that tend to have a large density distribution in the case of handwritten characters whose density distribution is small for prints with a constant print density and whose density is not constant due to writing pressure. Yes.
- the gradation distribution analysis unit 19 calculates a score (first score) obtained by quantifying the probability of handwriting or type from the analysis result.
- the character color analysis unit 20 calculates a score (second score) obtained by quantifying the probability of handwriting or type depending on whether the character is an achromatic color or a chromatic color from the analysis result.
- the character stroke gradation ratio analysis unit 21 is seen in a type in which the character density at the inflection point increases because the writing speed decreases at the point where the character stroke changes (inflection point). Features that tend not to be used for analysis.
- the character stroke gradation ratio analysis unit 21 analyzes the ratio between the gradation of the character stroke and the gradation of the inflection point, and calculates a score (third score) obtained by quantifying the probability of handwriting or type.
- the character stroke edge strength analysis unit 22 is characterized in that the edge strength (strength change strength) at the boundary between the printed character and the background is high in the case of printed characters, and the edge strength is weak in the case of handwritten characters. Is used for analysis.
- the character stroke edge strength analysis unit 22 analyzes the strength of the edge and calculates a score (fourth score) obtained by digitizing the probability of handwriting or type.
- the character size / center position analysis unit 23 analyzes the vertical size and center position of the character. That is, the character size entered in the form is printed with a certain vertical width in the case of printed characters, but the vertical width varies in the case of handwriting. As for the character entry position, the type is printed at a certain center position, but handwriting cannot be entered at a certain position like the type, so the center position varies. The character size / center position analysis unit 23 uses such tending characteristics for analysis, and calculates a score (fifth score) obtained by quantifying the probability of handwriting or type.
- the character size / pitch analysis unit 24 has a function to compensate for determining that characters smaller than usual such as stuttering are variations in the analysis by the character size / center position analysis unit 23.
- the character size / pitch analysis unit 24 determines that the character determined to be a variation between the same type of characters is likely to be a type, and obtains a score (sixth score) obtained by quantifying the probability of handwriting or type. calculate.
- the handwriting / printing mixed check unit 25 checks whether or not the character string in the recognition target item includes both handwritten characters and printed characters. Thereby, even if it is a character in the same recognition object item, if a handwritten character and a printed character are mixed, the character recognition process part 16 will use the character recognition engine suitable for character recognition processing for every character. It becomes possible.
- the handwriting / printing type discriminating unit 14 calculates a total score obtained by counting the first to sixth scores calculated by the respective analyzing units in consideration of weights according to importance, and a check result by the handwriting / printing mixed checking unit 25. Based on this, it is determined whether the image character is handwritten or printed.
- FIG. 2 is a flowchart showing the overall flow of character recognition processing
- FIG. 3 is a diagram showing an example of a form to be recognized
- FIG. 4 is a diagram showing a state in which a part to be character recognition is specified from a form image
- FIG. 5 is a diagram illustrating an example of weights attached to each score in the score totaling process.
- the character recognition device 10 first inputs an image of a form read by the scanner by the scanner control unit 12 (step S1). As shown in FIG. 2, the form to be read is, for example, a transfer request form. Character strings other than the title are printed and handwritten in a ruled line frame.
- the control unit 11 of the character recognition device 10 specifies a location for character recognition from the form image (step S2).
- the recognition target portion can be a ruled line frame that is an area in the ruled line frame and does not include a heading character. In FIG. This is indicated by a bold frame.
- the control unit 11 performs a labeling process or the like for cutting out image characters from the image of the identified recognition target portion, specifies the position and size of the image characters (step S3), and passes them to the handwriting / type discrimination unit 14. .
- the gradation distribution analysis unit 19 analyzes the distribution of gradation values representing the density of characters, and calculates a first score of a value corresponding to the degree of gradation distribution (step) S4), the character color analysis unit 20 analyzes the color of the character and calculates a second score having a value corresponding to the character color (step S5).
- the character stroke gradation ratio analysis unit 21 analyzes the ratio between the gradation of the character stroke and the gradation of the inflection point at which the character stroke changes, and calculates a third score corresponding to the magnitude of the ratio.
- the character stroke edge strength analysis unit 22 analyzes the edge strength indicating the change in shading of the character stroke edge, and calculates a fourth score corresponding to the blur of the character stroke edge (Step S7).
- the character size / center position analysis unit 23 analyzes the vertical size and center position of the character, calculates a fifth score according to the variation (step S8), and the character size / pitch analysis unit 24 recognizes it. Comparison of the vertical size of the target character and the adjacent characters and the character pitch size of the adjacent characters are analyzed, and a sixth score corresponding to a specific condition is calculated with respect to the vertical size and pitch of the character (step S9). ).
- the handwriting / type discrimination unit 14 aggregates the first to sixth scores calculated as described above, and discriminates whether the recognition target character is a handwritten character or a type character (step S10).
- the first to sixth scores are weighted according to their importance as shown in FIG. In the example of FIG. 5, the first score and the fifth score indicate the highest importance related to the handwriting / type discrimination process.
- the total score is calculated by the sum of values obtained by multiplying the first to sixth scores by the corresponding weights.
- the first to sixth scores calculated in each analysis process are quantified as ⁇ 50 to +50 according to the likelihood of handwriting or type. Whether it is handwritten or typed is based on 0, and when the total score is a positive number, it is determined as a typed character, and when it is a negative number, it is determined as a handwritten character.
- the handwriting / printing mixed check unit 25 checks whether handwritten characters and printed characters are mixed in the character string in the recognition target item (step S11).
- the character recognition processing unit 16 uses the handwritten character recognition dictionary 17 or the character recognition dictionary 18 based on the determination result in the handwriting / printing determination unit 14 and the check result in the handwriting / printing mixed check unit 25.
- a recognition process is performed (step S12), and a character recognition result is notified to a device that performs, for example, a transfer process (step S13).
- FIG. 6 is a flowchart showing the flow of handwritten type analysis processing based on gradation distribution
- FIG. 7 is a diagram for explaining gradation distribution analysis processing
- (A) shows a histogram of gradation values of handwritten characters
- (B ) Shows a histogram of gradation values of type characters.
- the gradation distribution analysis unit 19 creates a gray image from the image of the image character whose position and size are specified in the character recognition target item (step S21), and the character of the character obtained by excluding the background pixels from the gray image. For example, a histogram of the number of pixels (gradation value) when the pixels are scanned one by one is created (step S22). Next, the half width is calculated from the created histogram (step S23), and the first score is calculated using the threshold of the half width (step S24). This process is performed individually for all characters in the character recognition target item.
- the gradation value is dispersed and the distribution is widened.
- the half-value width of the gradation value distribution is also large and larger than the threshold value.
- the half-value width represents a width that is set so as to take half of the pixels by excluding background pixels from the histogram and scanning left and right from the point where the number of pixels is maximum in the remaining histograms.
- a negative number generated based on a value obtained by calculating the difference between the half width and the threshold value is set as the first score.
- FIG. 8 is a flowchart showing the flow of handwritten type analysis processing by character color
- FIG. 9 is a diagram for explaining character color analysis processing
- (A) shows a histogram of saturation of handwritten characters
- (B) shows handwriting A lightness histogram of characters is shown
- (C) shows a histogram of the saturation of type characters
- (D) shows a histogram of the lightness of type characters.
- the character color analyzing unit 20 converts an image character image whose position and size are specified in the character recognition target item into an HSV color space composed of components of hue (H), saturation (S), and lightness (V). (Step S31). Next, histograms for saturation (S) and lightness (V) are created to obtain respective distributions (step S32), and the presence / absence of a character color is calculated from the distribution of saturation (S) and lightness (V). (Step S33).
- the saturation distribution value uses a brighter color as the numerical value increases, and if a brighter color is not used, The degree is concentrated around zero.
- the lightness distribution value uses a brighter color as the numerical value increases, and is 0 when a brighter color is not used. Lightness is distributed to close positions.
- a threshold value is set for each of the saturation distribution value and the lightness distribution value.
- the saturation distribution value is smaller than the threshold value and the lightness distribution value is smaller than the threshold value, it can be determined that the color is achromatic.
- both the saturation distribution value and the lightness distribution value are larger than the threshold value, it can be determined that the color is chromatic.
- step S34 After calculating the character color in the process of step S33, it is determined whether or not the character color is an achromatic color (step S34). If the character color is an achromatic color, the second score is set to -50, and a hue histogram is not created. (Step S35). If the character color is not an achromatic color, +50 is set to the second score (step S36).
- a process for determining that the color is blue is added. That is, if the saturation distribution value and the lightness distribution value are larger than the threshold values and it is determined in step S34 that the color is not an achromatic color, a hue (H) histogram is created and the used color is determined.
- the hue (H) histogram represents the color distribution used in the image.
- the background pixels are excluded from the hue (H) histogram, and the proportion of pixels in the blue region among the total pixels in the remaining histogram is calculated. If the threshold value of the ratio of pixels existing in the blue region is set and the ratio exceeds the threshold value, it is determined that the character is blue.
- the score in this case is the difference between the ratio of pixels existing in the blue region and the threshold value, and a numerical value created based on the difference is set as the second score.
- FIG. 10 is a flowchart showing the flow of handwritten type analysis processing by character strokes
- FIG. 11 is a diagram for explaining each stroke and inflection points
- (A) shows an example of inflection points of handwritten characters
- (B ) Shows examples of inflection points of type characters.
- the character stroke gradation ratio analysis unit 21 extracts character strokes by searching for continuous black pixels from the image of the image character whose position and size are specified in the character recognition target item.
- a character stroke vector is obtained (step S41).
- a place where the vector changes is obtained, and the place is set as an inflection point (step S42).
- This inflection point is a portion where character stroke vectors intersect, and in the case of a handwritten character, as shown by a circle in FIG. In the case of the type), there is no such shade.
- step S43 The ratio between the average gradation value of the inflection point and the average gradation value of the entire character stroke other than the inflection point tends to be large when handwritten and small when printed.
- a threshold value of the gradation value ratio is set, a difference between the threshold value and the gradation value ratio is calculated, and a numerical value created based on the difference is used as a score as a third score. In this case, a positive number is set (step S44).
- FIG. 12 is a flowchart showing the flow of the handwritten type analysis process based on the character stroke edge strength
- FIG. 13 is a diagram for explaining the character stroke edge strength
- (A) shows the change in the gradation value in the case of a handwritten character
- (B) shows a change in gradation value in the case of type characters.
- the character stroke edge strength analysis unit 22 uses the characteristic that the strength of the change in shading, that is, the character stroke edge strength differs between the handwritten character and the printed character at the edge of the boundary between the character and the background.
- the character stroke edge strength differs between the handwritten character and the printed character at the edge of the boundary between the character and the background.
- all the strokes of a character are divided into three in the left-right direction as seen from the direction of vector movement, and the gradation value average (first gradation value) of the edges serving as the left boundary and the right boundary is calculated (step S51).
- the gradation value average (referred to as the second gradation value) of the stroke of the entire character is calculated (step S52), and the difference between the first gradation value and the second gradation value is calculated (step S53).
- the gradation value average referred to as the second gradation value
- the first gradation value is lowered due to the blurring of the character stroke edge portion, so the difference from the second gradation value is large.
- the difference between the first gradation value and the second gradation value becomes small.
- a threshold value of the difference between the first gradation value and the second gradation value is set, and a score is calculated using the difference between the first gradation value and the second gradation value and the set threshold value, and the handwritten In the case of a negative number, and in the case of type, a positive number is set as the fourth score (step S54).
- the above steps S52 and S53 calculate the average of the background gradation values around the edge (the third gradation value, which is the second gradation value in the claims), and the first floor It is replaced with the step of calculating the difference between the tone value and the third gradation value. Since the difference between the first gradation value and the third gradation value is small for handwritten characters and large for printed characters, the sign of the score set for the fourth score is determined based on this determination. It is done.
- FIG. 14 is a flowchart showing the flow of handwritten type analysis processing based on the character size / center position.
- FIG. 15 is a diagram for explaining an analysis example based on the character size / center position of the handwritten character.
- (B) shows the calculation result of classification by character vertical size
- (C) shows the center coordinate average of each character
- (D) shows the center coordinate average and center coordinate of each character.
- 16 is a diagram for explaining an analysis example based on the character size / center position of type characters, (A) shows the calculation result of the circumscribed rectangle of the set of black pixels, and (B) Indicates a calculation result of classification by character vertical size, (C) indicates a center coordinate average of each character, and (D) indicates a calculation result of a difference between the center coordinate average of each character and the center coordinate.
- the character size / center position analysis unit 23 calculates the vertical size of each character in the character recognition target item (step S61), and collects characters having the same size to create a character string set (step S62).
- the same size is a size that allows a certain degree of error, including complete identity.
- the calculation result of the vertical size of the character is managed together with the data of the character order, the character frame center coordinates, and the character frame width in the character recognition target item. Is done.
- the character string sets are classified according to the vertical size of characters from within the character recognition target items. There are three classifications.
- the ratio of the average vertical size, average center coordinates, and vertical size is also calculated for each classified character string.
- the ratio of the vertical size is calculated using the calculated average vertical size as S1, S2,. . . , Sn, where Smax is the largest of S1 to Sn, and S is representative of the other character strings, and the average vertical size Smax of the largest character string and the average vertical length of the other character strings. It is obtained by the ratio (S / Smax) with the size S.
- the center position of the character string in which characters having the same vertical size are collected is evaluated (step S63). That is, the center coordinate average M of the recognition item is calculated by dividing the sum of the products of the average center coordinates of each character string and the square of the number of characters by the sum of the squares of the number of characters.
- the center coordinate average M of the recognition item is calculated by dividing the sum of the products of the average center coordinates of each character string and the square of the number of characters by the sum of the squares of the number of characters.
- 33.5 ⁇ 4 + 64 ⁇ 1 + 43 ⁇ 1) / (4 + 1 + 1) 40.2 (1)
- the central coordinate average M and the characters of the character recognition target item have a positional relationship as shown in (C) of FIG. 15 and (C) of FIG.
- character strings having a vertical size ratio equal to or less than a certain ratio are not evaluated by the center position.
- the character string of the second classification is excluded from the calculation formula (2) of the central coordinate average M because the vertical size ratio is as low as 56%.
- step S64 the difference between the center coordinate average M and the center coordinates of each character is calculated to investigate the variation, and the fifth score is calculated using the threshold value (step S64). If the variation is large, the character is handwritten, and if it is small, the character is determined as a type.
- FIG. 15D and FIG. 16D show examples of character variations and scores in the character recognition target item. The handwritten type analysis process based on the character size / center position is performed for each character recognition target item, and the score performed for each item is set as the score for each character.
- FIG. 17 is a flowchart showing a flow of handwritten type analysis processing between character frames
- FIG. 18 is a diagram for explaining an example of handwritten type analysis between character frames.
- the character size / pitch analysis unit 24 calculates the character size of the character positioned on both sides of the recognition target character (step S71), calculates the character pitch between the recognition target character and the adjacent character (step S72), and the corresponding character. Determines whether or not a predetermined condition is satisfied, and sets a sixth score (step S73).
- This specific condition is that the vertical size of the character is equal to or smaller than the character size on both sides, and the pitch between the characters on both sides is approximately equal.
- the character size of “” is within the height of both adjacent characters (“B” and “”), and the position is also inferred from both adjacent characters, so it is determined as a type. . Thereby, in the stuttering sound between the same type of characters, a character that matches this specific condition is determined as a type.
- the sixth score is set to +50 when a character satisfying the above specific condition is determined as a printed character, and is set as ⁇ 50 when it is determined that the character is not a handwritten character.
- the handwriting / printing type discriminating unit 14 totals the first to sixth scores analyzed as described above, and discriminates between handwritten characters and printed characters. Since a plurality of analysis results are combined and based on this, it is discriminated comprehensively whether it is a handwritten character or a printed character, so that discrimination with higher accuracy is possible.
- FIG. 19 is a flowchart showing a flow of mixed handwriting / printing check processing
- FIG. 20 is an explanatory diagram of processing when handwritten characters and printed characters are mixed in the recognition target item.
- the handwriting / printing mixed check unit 25 calculates an average score of all characters of the recognition target item and checks a difference from each character (step S81), excludes characters having an extreme difference, and averages again. A value is calculated and it is checked whether there is a change in the difference to determine whether it is handwritten or typed (step S82).
- Each character in the recognition target item is analyzed and a total score is calculated.
- a difference between the first average value and the total score of each character is calculated, and it is determined whether or not the absolute value of the difference is larger than a set threshold value. If the threshold value is “10”, for example, in the example of FIG. 20, since the absolute value of the difference between the first average value and the total score of each character is larger than the set threshold value in each example, the determination at this stage Are all x.
- the difference between the first average value and the second average value is large, it is determined that the character type of the excluded character is different from other characters. In the illustrated example, since the difference between the first average value and the second average value is large, it is determined that the excluded character and the remaining character have different character types.
- the difference between the second average value and the total score of each remaining character is calculated, and it is determined whether or not the absolute value of the difference is larger than the set threshold value.
- the absolute value of the difference between the second average value and the total score of each character is smaller than the set threshold value “10” and can be estimated as the same character type in all the characters, all the determinations are ⁇ . Yes.
- the characters “flat”, “sei”, “month”, and “day” are determined as type characters, and the excluded characters “2”, “1”, “1”, “1”, “3”, and “3” “0” is determined as a handwritten character.
- the determination process by the above analysis demonstrated the case where a definition body is not used, the precision of a handwriting / type determination process can be improved by using together the process using a definition body.
- any one of the tone distribution analysis unit 19, the character color analysis unit 20, the character stroke tone ratio analysis unit 21, the character stroke edge strength analysis unit 22, the character size / center position analysis unit 23, and the character size / pitch analysis unit 24 In such an analysis process, if it is determined that the certainty of the type character exceeds a certain predetermined threshold value, the remaining analysis regarding the image character may be skipped and the analysis may proceed to the next image character. .
- FIG. 21 is a diagram illustrating a hardware configuration example of a computer used in the character recognition device.
- a part of the character recognition device 10 is configured by a computer 50.
- the entire computer 50 is controlled by a CPU (Central Processing Unit) 51.
- a random access memory (RAM) 52, a hard disk drive (HDD) 53, a graphic processing device 54, an input interface 55, and a communication interface 56 are connected to the CPU 51 via a bus 57.
- RAM Central Processing Unit
- HDD hard disk drive
- FIG. 21 is a diagram illustrating a hardware configuration example of a computer used in the character recognition device.
- a part of the character recognition device 10 is configured by a computer 50.
- the entire computer 50 is controlled by a CPU (Central Processing Unit) 51.
- a random access memory (RAM) 52, a hard disk drive (HDD) 53, a graphic processing device 54, an input interface 55, and a communication interface 56 are connected to the CPU 51 via a bus 57.
- HDD hard disk drive
- the RAM 52 temporarily stores at least a part of an OS (Operating System) program to be executed by the CPU 51 and an application program necessary for character recognition processing.
- the RAM 52 stores various data necessary for processing by the CPU 51.
- the hard disk drive 53 stores an OS, an application program for character recognition, a dictionary used for character recognition, and the like.
- a monitor 58 is connected to the graphic processing device 54.
- the graphic processing device 54 displays an image on the screen of the monitor 58 in accordance with a command from the CPU 51.
- a scanner 59, a keyboard 60 and a mouse 61 are connected to the input interface 55.
- the input interface 55 transmits signals sent from the scanner 59, keyboard 60, and mouse 61 to the CPU 51 via the bus 57.
- the communication interface 56 is connected to a network in the financial institution.
- the communication interface 56 transmits / receives data to / from other computers via a network.
- the processing functions related to the character recognition device 10 of the present embodiment can be realized.
- a program describing the processing contents of the functions that the character recognition apparatus 10 should have is provided.
- the above processing functions are realized on the computer.
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Character Input (AREA)
- Character Discrimination (AREA)
Abstract
Description
文字認識装置10は、装置全体を制御する制御部11を備え、この制御部11には、スキャナ制御部12、文字認識位置特定部13、手書き/活字判別部14および認識結果表示部15が接続されている。手書き/活字判別部14は、文字認識処理部16が接続され、この文字認識処理部16には、文字認識処理のときに参照される手書き文字認識辞書17および活字文字認識辞書18が関連されている。
次に、手書き/活字判別部14での判別結果および手書き/活字混在チェック部25によるチェック結果を基に、文字認識処理部16が手書き文字認識辞書17または活字文字認識辞書18を使用して文字認識処理を行い(ステップS12)、文字認識結果をたとえば振込処理を行う装置に通知する(ステップS13)。
図6は階調分布による手書き活字解析処理の流れを示すフローチャート、図7は階調分布解析処理を説明する図であって、(A)は手書き文字の階調値のヒストグラムを示し、(B)は活字文字の階調値のヒストグラムを示している。
M=(33.5×4+64×1+43×1)/(4+1+1)=40.2・・(1)
となり、図16に例示した例では、
M=(87.8×25+86.8×16)/(25+16)=87.4・・(2)
となる。この中心座標平均Mおよび文字認識対象項目の文字は、図15の(C)および図16の(C)に示したような位置関係になる。なお、中心座標平均Mの算出において、縦サイズの割合が一定の割合以下の文字列は、ここの中心位置による評価は行わないようにしている。たとえば、図16の(B)の例では、第2分類の文字列は、その縦サイズの割合が56%と低いので、中心座標平均Mの算出式(2)から除外している。
文字サイズ/ピッチ解析部24は、認識対象文字の両隣に位置する文字の文字サイズを算出し(ステップS71)、認識対象文字と両隣の文字との文字ピッチを算出し(ステップS72)、該当文字が所定の条件を満たすかどうか判定し、第6スコアを設定する(ステップS73)。文字サイズ/中心位置解析部23による文字の縦サイズおよび中心位置に基づいた解析では、拗音等、通常文字より小さいものをばらつきと判断していたが、文字サイズ/ピッチ解析部24による解析では、特定の条件に一致したものは、活字と判断することにしている。
手書き/活字判別部14では、以上のようにして解析された第1ないし第6スコアを集計して手書き文字か活字文字かを判別している。複数の解析結果を組み合わせ、これを基に総合的に手書き文字か活字文字かを判別しているので、より精度の高い判別を可能にしている。
手書き/活字混在チェック部25は、認識対象項目の全文字についてスコアの平均値を算出して各文字との差をチェックし(ステップS81)、極端な差がある文字は除外して、再度平均値を算出し、差に変化がないかをチェックして手書きか活字かを判断する(ステップS82)。
なお、以上の解析による判別処理は、定義体を使用しない場合について説明したが、定義体を使用した処理を併用することにより、手書き/活字判別処理の精度を向上させることができる。また、階調分布解析部19、文字色解析部20、文字ストローク階調比解析部21、文字ストロークエッジ強度解析部22、文字サイズ/中心位置解析部23および文字サイズ/ピッチ解析部24のいずれかの解析処理において、活字の確からしさが確実な所定の閾値を超えていると判断された場合、当該イメージ文字に関する残りの解析をスキップして次のイメージ文字の解析に進めるようにしてもよい。
文字認識装置10は、その一部がコンピュータ50によって構成されている。コンピュータ50は、CPU(Central Processing Unit)51によって装置全体が制御されている。CPU51には、バス57を介してRAM(Random Access Memory)52、ハードディスクドライブ(HDD:Hard Disk Drive)53、グラフィック処理装置54、入力インタフェース55、および通信インタフェース56が接続されている。
11 制御部
12 スキャナ制御部
13 文字認識位置特定部
14 手書き/活字判別部
15 認識結果表示部
16 文字認識処理部
17 手書き文字認識辞書
18 活字文字認識辞書
19 階調分布解析部
20 文字色解析部
21 文字ストローク階調比解析部
22 文字ストロークエッジ強度解析部
23 文字サイズ/中心位置解析部
24 文字サイズ/ピッチ解析部
25 手書き/活字混在チェック部
50 コンピュータ
51 CPU
52 RAM
53 ハードディスクドライブ
54 グラフィック処理装置
55 入力インタフェース
56 通信インタフェース
57 バス
58 モニタ
59 スキャナ
60 キーボード
61 マウス
Claims (12)
- 帳票上の手書きの文字と活字の文字とを判別して認識する文字認識方法において、
前記帳票のイメージデータから文字認識の対象となる項目を特定し、
特定された認識対象項目のイメージ文字の位置およびサイズを算出し、
前記イメージ文字の手書きおよび活字の特性を解析して手書きまたは活字の確からしさを数値化したスコアを算出し、
前記スコアを基に前記イメージ文字が手書きか活字かを判定し、
認識対象項目内の各文字のスコアの平均値を算出して手書きおよび活字の混在をチェックし、
各認識対象項目の文字を手書きか活字かの判定結果および手書きおよび活字の混在のチェック結果に応じた文字認識エンジンを用いて文字認識を行う、
ことからなり、
前記スコアの算出は、
前記イメージ文字の階調分布を解析して手書きまたは活字の確からしさを数値化した第1スコアを算出し、
前記イメージ文字の文字色を解析して手書きまたは活字の確からしさを数値化した第2スコアを算出し、
前記イメージ文字から抽出した文字ストロークから前記文字ストロークが変化する変曲点の階調と変曲点以外の文字ストロークの階調との比を解析して手書きまたは活字の確からしさを数値化した第3スコアを算出し、
前記文字ストロークと背景との境界における濃淡変化の強さを表す文字ストロークエッジ強度を解析して手書きまたは活字の確からしさを数値化した第4スコアを算出し、
前記イメージ文字の位置およびサイズから文字の縦サイズおよび中心位置を解析して手書きまたは活字の確からしさを数値化した第5スコアを算出し、
前記イメージ文字の位置およびサイズから文字サイズおよびピッチを解析して手書きまたは活字の確からしさを数値化した第6スコアを算出する、
ことを含み、
前記イメージ文字が手書きか活字かの判定では、前記第1ないし第6スコアをそれらの重要度に応じて設定されたウエイトを考慮して前記スコアを集計している、
ことを特徴とする文字認識方法。 - 前記階調分布の解析、前記文字色の解析、前記文字ストロークの解析、前記文字ストロークエッジ強度の解析、前記文字の縦サイズおよび中心位置の解析、および前記文字サイズおよびピッチの解析において、前記第1ないし第6スコアのいずれかが確実に活字と判断できる所定の閾値を超えている場合、当該イメージ文字に関する残りの解析をスキップして次の前記イメージ文字の解析に進めることを特徴とする請求の範囲第1項記載の文字認識方法。
- 前記階調分布の解析は、前記イメージ文字のグレー画像を作成し、1文字単位の階調値のヒストグラムを作成し、前記ヒストグラムよりその半値幅を算出し、前記半値幅と所定の閾値との差を算出して前記第1スコアとすることを特徴とする請求の範囲第1項記載の文字認識方法。
- 前記文字色の解析は、前記イメージ文字を色相、彩度および明度の成分からなる色空間に変換し、前記彩度および前記明度についてのヒストグラムを作成し、前記彩度および前記明度の分布値により無彩色か有彩色かを判断し、無彩色または有彩色に応じた値を前記第2スコアに設定することを特徴とする請求の範囲第1項記載の文字認識方法。
- 有彩色の中から特定の文字色を判別する場合、前記色相についてのヒストグラムを作成し、前記色相の分布値に応じた値を前記第2スコアに設定することを特徴とする請求の範囲第4項記載の文字認識方法。
- 前記文字ストロークの解析は、前記イメージ文字から抽出した文字ストロークのベクトルを求め、前記ベクトルの交点である変曲点を算出し、前記変曲点の平均階調値および前記変曲点以外の前記文字ストローク全体の平均階調値を算出し、前記変曲点の平均階調値と前記変曲点以外の前記文字ストローク全体の平均階調値との比に応じた値を前記第3スコアとして設定することを特徴とする請求の範囲第1項記載の文字認識方法。
- 前記文字ストロークエッジ強度の解析は、前記イメージ文字から文字ストロークを抽出し、前記文字ストロークのエッジの階調値の平均である第1階調値を算出し、前記文字ストロークの全体の階調値の平均である第2階調値を算出し、前記第1階調値と前記第2階調値との差と所定の閾値との差を前記第4スコアとして設定することを特徴とする請求の範囲第1項記載の文字認識方法。
- 前記文字ストロークエッジ強度の解析は、前記イメージ文字から文字ストロークおよびその背景を抽出し、前記文字ストロークのエッジの階調値の平均である第1階調値を算出し、前記エッジの周辺における前記背景の階調値の平均である第2階調値を算出し、前記第1階調値と前記第2階調値との差と所定の閾値との差を前記第4スコアとして設定することを特徴とする請求の範囲第1項記載の文字認識方法。
- 前記文字の縦サイズおよび中心位置の解析は、縦サイズが同一の文字を集めた文字列についての文字列中心位置を算出し、文字の中心位置と前記文字列中心位置との差と所定の閾値との差を前記第5スコアとして設定することを特徴とする請求の範囲第1項記載の文字認識方法。
- 前記文字サイズおよびピッチの解析は、認識対象項目にて判定対象となる文字およびその両隣の文字に対して縦サイズ差と文字間ピッチとを算出し、前記縦サイズ差および前記文字間ピッチが所定の関係にあるかどうかに応じた値を前記第6スコアに設定することを特徴とする請求の範囲第1項記載の文字認識方法。
- 帳票上の手書きおよび活字の文字を認識する文字認識装置において、
前記帳票のイメージデータから特定された認識対象項目のイメージ文字の階調分布を解析して手書きまたは活字の確からしさを数値化した第1スコアを算出する階調分布解析部と、
前記イメージ文字の文字色を解析して手書きまたは活字の確からしさを数値化した第2スコアを算出する文字色解析部と、
前記イメージ文字から抽出した文字ストロークから前記文字ストロークが変化する変曲点の階調と変曲点以外の文字ストロークの階調との比を解析して手書きまたは活字の確からしさを数値化した第3スコアを算出する文字ストローク階調比解析部と、
前記文字ストロークと背景との境界における濃淡変化の強さを表す文字ストロークエッジ強度を解析して手書きまたは活字の確からしさを数値化した第4スコアを算出する文字ストロークエッジ強度解析部と、
前記イメージ文字の位置およびサイズから文字の縦サイズおよび中心位置を解析して手書きまたは活字の確からしさを数値化した第5スコアを算出する文字サイズ/中心位置解析部と、
前記イメージ文字の位置およびサイズから文字サイズおよびピッチを解析して手書きまたは活字の確からしさを数値化した第6スコアを算出する文字サイズ/ピッチ解析部と、
認識対象項目内の各文字の前記第1ないし第6スコアの平均値を算出して手書きおよび活字の混在をチェックする手書き/活字混在チェック部と、
を有し、前記第1ないし第6スコアを重要度に応じて設定されたウエイトを考慮して集計した集計スコアと、前記手書き/活字混在チェック部によるチェック結果とを基に、前記イメージ文字が手書きか活字かを判定する手書き/活字判別部と、
各認識対象項目の文字について、前記手書き/活字判別部による判定結果に応じた文字認識エンジンを用いて文字認識を行う文字認識処理部と、
を備えていることを特徴とする文字認識装置。 - 帳票上の手書きの文字と活字の文字とを判別する処理をコンピュータに実行させる文字認識プログラムであって、
前記コンピュータに、
前記帳票のイメージデータから文字認識の対象となる項目を特定し、
特定された認識対象項目のイメージ文字の位置およびサイズを算出し、
前記イメージ文字の階調分布を解析して手書きまたは活字の確からしさを数値化した第1スコアを算出し、
前記イメージ文字の文字色を解析して手書きまたは活字の確からしさを数値化した第2スコアを算出し、
前記イメージ文字から抽出した文字ストロークから前記文字ストロークが変化する変曲点の階調と変曲点以外の文字ストロークの階調との比を解析して手書きまたは活字の確からしさを数値化した第3スコアを算出し、
前記文字ストロークと背景との境界における濃淡変化の強さを表す文字ストロークエッジ強度を解析して手書きまたは活字の確からしさを数値化した第4スコアを算出し、
前記イメージ文字の位置およびサイズから文字の縦サイズおよび中心位置を解析して手書きまたは活字の確からしさを数値化した第5スコアを算出し、
前記イメージ文字の位置およびサイズから文字サイズおよびピッチを解析して手書きまたは活字の確からしさを数値化した第6スコアを算出し、
前記第1ないし第6スコアを重要度に応じて設定されたウエイトを考慮して集計することにより前記イメージ文字が手書きか活字かを判定し、
認識対象項目内の各文字のスコアの平均値を算出して手書きおよび活字の混在をチェックし、
各認識対象項目の文字を手書きか活字かの判定結果および手書きおよび活字の混在のチェック結果に応じて文字認識を行う、
処理を実行させることを特徴とする文字認識プログラム。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP09852261.8A EP2515257A4 (en) | 2009-12-15 | 2009-12-15 | METHOD, DEVICE AND PROGRAM FOR CHARACTER RECOGNITION |
JP2011545873A JP5363591B2 (ja) | 2009-12-15 | 2009-12-15 | 文字認識方法、文字認識装置および文字認識プログラム |
PCT/JP2009/070886 WO2011074067A1 (ja) | 2009-12-15 | 2009-12-15 | 文字認識方法、文字認識装置および文字認識プログラム |
US13/472,065 US8588520B2 (en) | 2009-12-15 | 2012-05-15 | Character recognition method, character recognition apparatus, and character recognition program |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2009/070886 WO2011074067A1 (ja) | 2009-12-15 | 2009-12-15 | 文字認識方法、文字認識装置および文字認識プログラム |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US13/472,065 Continuation US8588520B2 (en) | 2009-12-15 | 2012-05-15 | Character recognition method, character recognition apparatus, and character recognition program |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2011074067A1 true WO2011074067A1 (ja) | 2011-06-23 |
Family
ID=44166864
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2009/070886 WO2011074067A1 (ja) | 2009-12-15 | 2009-12-15 | 文字認識方法、文字認識装置および文字認識プログラム |
Country Status (4)
Country | Link |
---|---|
US (1) | US8588520B2 (ja) |
EP (1) | EP2515257A4 (ja) |
JP (1) | JP5363591B2 (ja) |
WO (1) | WO2011074067A1 (ja) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2015201038A (ja) * | 2014-04-08 | 2015-11-12 | 富士通株式会社 | 情報処理装置及び情報処理プログラム |
CN112733902A (zh) * | 2020-12-30 | 2021-04-30 | 杭州海康威视数字技术股份有限公司 | 烟盒识别方法 |
JP2022067086A (ja) * | 2020-10-19 | 2022-05-02 | アクセンチュア グローバル ソリューションズ リミテッド | デジタル化された筆記の処理 |
Families Citing this family (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
RU2640322C2 (ru) * | 2014-01-30 | 2017-12-27 | Общество с ограниченной ответственностью "Аби Девелопмент" | Способы и системы эффективного автоматического распознавания символов |
US10725650B2 (en) * | 2014-03-17 | 2020-07-28 | Kabushiki Kaisha Kawai Gakki Seisakusho | Handwritten music sign recognition device and program |
US9405997B1 (en) | 2014-06-17 | 2016-08-02 | Amazon Technologies, Inc. | Optical character recognition |
US9330311B1 (en) * | 2014-06-17 | 2016-05-03 | Amazon Technologies, Inc. | Optical character recognition |
US10489051B2 (en) * | 2014-11-28 | 2019-11-26 | Samsung Electronics Co., Ltd. | Handwriting input apparatus and control method thereof |
JP6142882B2 (ja) * | 2015-01-28 | 2017-06-07 | コニカミノルタ株式会社 | 画像形成装置 |
JP6492894B2 (ja) * | 2015-04-01 | 2019-04-03 | 富士通株式会社 | 認識プログラム、認識方法及び認識装置 |
CN104992188B (zh) * | 2015-07-15 | 2018-04-20 | 南京邮电大学 | 一种基于t混合因子分析的分布式手写数字识别方法 |
CN106096601B (zh) * | 2016-06-06 | 2019-06-18 | 深圳辰通智能股份有限公司 | 一种自动检测票据中字符类型的方法和*** |
CN109243244B (zh) * | 2018-08-13 | 2020-10-30 | 广东小天才科技有限公司 | 一种书写评分方法、装置、设备及存储介质 |
CN109871910B (zh) * | 2019-03-12 | 2021-06-22 | 成都工业学院 | 一种手写字符识别方法及装置 |
JP2021033855A (ja) * | 2019-08-28 | 2021-03-01 | 富士ゼロックス株式会社 | 情報処理装置及び情報処理プログラム |
JP7282314B2 (ja) * | 2019-08-29 | 2023-05-29 | 京セラドキュメントソリューションズ株式会社 | 画像処理装置、画像処理方法及び画像処理プログラム |
CN110705362B (zh) * | 2019-09-06 | 2022-12-09 | 航天新长征大道科技有限公司 | 一种字印分析方法和装置 |
US11488407B1 (en) * | 2021-06-01 | 2022-11-01 | Lead Technologies, Inc. | Method, apparatus, and computer-readable storage medium for recognizing characters in a digital document |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS6414682A (en) * | 1987-07-08 | 1989-01-18 | Toshiba Corp | Device for deciding kind of address |
JP2000181993A (ja) | 1998-12-16 | 2000-06-30 | Fujitsu Ltd | 文字認識方法および装置 |
JP2000331122A (ja) | 1999-05-21 | 2000-11-30 | Fujitsu Ltd | 文字認識方法および装置 |
JP2006092345A (ja) | 2004-09-24 | 2006-04-06 | Fuji Xerox Co Ltd | 文字認識装置、文字認識方法および文字認識プログラム |
JP2008033604A (ja) * | 2006-07-28 | 2008-02-14 | Univ Of Tokyo | 画像処理システム、文字認識システムおよび画像処理プログラム |
JP2009053826A (ja) * | 2007-08-24 | 2009-03-12 | Fuji Xerox Co Ltd | 文書処理装置及び文書処理プログラム |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS63261486A (ja) * | 1987-04-20 | 1988-10-28 | Nec Corp | 書体識別装置 |
US5181255A (en) * | 1990-12-13 | 1993-01-19 | Xerox Corporation | Segmentation of handwriting and machine printed text |
US7072514B1 (en) * | 2003-02-06 | 2006-07-04 | The United States Of America As Represented By The National Security Agency | Method of distinguishing handwritten and machine-printed images |
JP2006092027A (ja) * | 2004-09-21 | 2006-04-06 | Fuji Xerox Co Ltd | 文字認識装置、文字認識方法および文字認識プログラム |
-
2009
- 2009-12-15 JP JP2011545873A patent/JP5363591B2/ja active Active
- 2009-12-15 WO PCT/JP2009/070886 patent/WO2011074067A1/ja active Application Filing
- 2009-12-15 EP EP09852261.8A patent/EP2515257A4/en not_active Withdrawn
-
2012
- 2012-05-15 US US13/472,065 patent/US8588520B2/en not_active Expired - Fee Related
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS6414682A (en) * | 1987-07-08 | 1989-01-18 | Toshiba Corp | Device for deciding kind of address |
JP2000181993A (ja) | 1998-12-16 | 2000-06-30 | Fujitsu Ltd | 文字認識方法および装置 |
JP2000331122A (ja) | 1999-05-21 | 2000-11-30 | Fujitsu Ltd | 文字認識方法および装置 |
JP2006092345A (ja) | 2004-09-24 | 2006-04-06 | Fuji Xerox Co Ltd | 文字認識装置、文字認識方法および文字認識プログラム |
JP2008033604A (ja) * | 2006-07-28 | 2008-02-14 | Univ Of Tokyo | 画像処理システム、文字認識システムおよび画像処理プログラム |
JP2009053826A (ja) * | 2007-08-24 | 2009-03-12 | Fuji Xerox Co Ltd | 文書処理装置及び文書処理プログラム |
Non-Patent Citations (1)
Title |
---|
See also references of EP2515257A4 |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2015201038A (ja) * | 2014-04-08 | 2015-11-12 | 富士通株式会社 | 情報処理装置及び情報処理プログラム |
US9921742B2 (en) | 2014-04-08 | 2018-03-20 | Fujitsu Limited | Information processing apparatus and recording medium recording information processing program |
JP2022067086A (ja) * | 2020-10-19 | 2022-05-02 | アクセンチュア グローバル ソリューションズ リミテッド | デジタル化された筆記の処理 |
JP7364639B2 (ja) | 2020-10-19 | 2023-10-18 | アクセンチュア グローバル ソリューションズ リミテッド | デジタル化された筆記の処理 |
CN112733902A (zh) * | 2020-12-30 | 2021-04-30 | 杭州海康威视数字技术股份有限公司 | 烟盒识别方法 |
CN112733902B (zh) * | 2020-12-30 | 2023-09-05 | 杭州海康威视数字技术股份有限公司 | 烟盒识别方法 |
Also Published As
Publication number | Publication date |
---|---|
EP2515257A4 (en) | 2016-12-07 |
EP2515257A1 (en) | 2012-10-24 |
US8588520B2 (en) | 2013-11-19 |
JPWO2011074067A1 (ja) | 2013-04-25 |
JP5363591B2 (ja) | 2013-12-11 |
US20120308127A1 (en) | 2012-12-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5363591B2 (ja) | 文字認識方法、文字認識装置および文字認識プログラム | |
JP3833153B2 (ja) | 画像処理方法及び装置 | |
US8494273B2 (en) | Adaptive optical character recognition on a document with distorted characters | |
JP5379085B2 (ja) | スキャンされた文書画像内の前景画素群の連結グループをマーキング種類に基づき分類する方法及びシステム | |
US20210064859A1 (en) | Image processing system, image processing method, and storage medium | |
US7653244B2 (en) | Intelligent importation of information from foreign applications user interface | |
US20030063802A1 (en) | Image processing method, apparatus and system | |
CN111027297A (zh) | 一种对图像型pdf财务数据关键表格信息的处理方法 | |
CN109784342B (zh) | 一种基于深度学习模型的ocr识别方法及终端 | |
US20070253040A1 (en) | Color scanning to enhance bitonal image | |
US9646230B1 (en) | Image segmentation in optical character recognition using neural networks | |
US11983910B2 (en) | Image processing system, image processing method, and storage medium each for obtaining pixels of object using neural network | |
KR20150137752A (ko) | 문자 인식 방법 및 그 장치 | |
JP2002352191A (ja) | 手書き判別能力を持つプリンティング制御インターフェースシステムとその方法 | |
US8194941B2 (en) | Character noise eliminating apparatus, character noise eliminating method, and character noise eliminating program | |
US20080175483A1 (en) | Pattern separating extraction program, pattern separating extraction device, and pattern separating extraction method | |
Al Abodi et al. | An effective approach to offline Arabic handwriting recognition | |
CN111738252A (zh) | 图像中的文本行检测方法、装置及计算机*** | |
CN111047618B (zh) | 基于多尺度的无参考屏幕内容图像质量评估方法 | |
US8687239B2 (en) | Relevance based print integrity verification | |
Pati et al. | Handwriting Quality Assessment using Structural Features and Support Vector Machines | |
US8589416B2 (en) | System and method of performing data processing on similar forms | |
US8488886B2 (en) | Font matching | |
US20240176951A1 (en) | Electronic document validation | |
US11948342B2 (en) | Image processing apparatus, image processing method, and non-transitory storage medium for determining extraction target pixel |
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: 09852261 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2011545873 Country of ref document: JP Kind code of ref document: A |
|
REEP | Request for entry into the european phase |
Ref document number: 2009852261 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2009852261 Country of ref document: EP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |