JP3710164B2 - Image processing apparatus and method - Google Patents

Description

[0001]
[Industrial application fields]
The present invention relates to an image processing apparatus and method, and more particularly to an image processing apparatus and method for determining a character type in an input document image and passing it to a lower-level process.
[0002]
[Prior art]
There is an urgent need for computerization of information that facilitates information management and retrieval in a modern society where information is overflowing. In order to digitize information, OCR (optical character recognition) that converts a character, which is an image read by an input device such as a scanner, into a character code is indispensable, and its accuracy is steadily improving.
[0003]
OCR is different from Japanese (Kanji, Hiragana, Katakana) and English and others (alphabet), so it is difficult to recognize alphabets (especially lowercase letters) with Japanese OCR. Use the same algorithm or switch dictionaries. Therefore, a technique for distinguishing between alphabet and Japanese is required.
[0004]
Conventionally, there was no method for discriminating between alphabet and Japanese, so the user operated the operation panel, keyboard, pointing device, or the like to give an instruction to distinguish them.
[0005]
[Problems to be solved by the invention]
However, this requires a great deal of time and trouble for the user.
[0006]
When reading a plurality of pieces of data, an ADF (Auto Document Feeder) reduces the document redirection work to the user, but there is a possibility that English pages and Japanese documents are mixed in the plurality of pages. If it waits for a user's instruction each time one sheet is read, the advantage of ADF is spoiled. In addition, a large amount of memory is required in order to receive a user instruction after reading all pages.
[0007]
[Means for Solving the Problems]
as well as
[Action]
The present invention has been made in view of such problems, and image processing that can improve processing accuracy in lower-level processing, for example, character recognition processing, by efficiently determining the character type in an input document image. An apparatus and method are to be provided.
[0008]
  In order to solve this problem, for example, an image processing apparatus of the present invention has the following configuration. That is,
  EnteredManuscriptimageAn image processing apparatus that determines the type of characters in the image and passes it to a lower-level process.
  In the input document imageText imageincludingcharacterlineAreaDiscriminating means for discriminating;
  DeterminedcharacterThe row areacharacterRow direction4 aligned verticallyDividing means for dividing into areas,
  In the dividing meansCounting means for counting significant dots in each divided area;
  SaidCounted by counting meansWhen the number of significant dots in each region is B1, B2, B3, and B4 in order from the upper region, the first dot distribution R1 is calculated based on the ratio of (B2 + B3) to (B1 + B4), and (B3 + B4) Based on the ratio to (B1 + B2), the second dot distribution R2Calculating means for calculating
  SaidCalculatedWhen it is determined that the first dot distribution R1 is greater than or equal to the first threshold value T1 or the calculated second dot distribution R2 is greater than or equal to the second threshold value T2, the character line area is an alphabetic character line. Judge that there is,
  When it is determined that the first dot distribution R1 is smaller than the first threshold value T1 and the second dot distribution R2 is smaller than the second threshold value T2, the character line area is a Chinese character area character line.Determination means for determining.
[0009]
Further, according to a preferred embodiment of the present invention, the sub-process is a character recognition process, and a recognition algorithm used when performing a recognition process according to a result determined by the determination means, or a recognition algorithm Recognition dictionary, orrecognitionIt is desirable to have a dictionary selected. As a result, since the recognition algorithm or the recognition dictionary used in the recognition process can be selected in advance, the recognition speed and the recognition accuracy can be increased.
[0011]
The discriminating unit preferably discriminates a row based on a dot distribution projected in the arrangement direction of the character string pattern. Thereby, it becomes possible to discriminate the line with high accuracy.
[0012]
Further, it is desirable to further include noise removing means for removing noise in the inputted document image. As a result, the influence of noise can be eliminated, so that the row can be discriminated more accurately.
[0013]
Furthermore, it is desirable to further comprise means for rotating the input image by a predetermined angle to determine the direction as an erect image. This makes it possible to make a determination according to the input direction of the document image.
[0014]
Furthermore, it is desirable to provide means for correcting the tilt when the input image is tilted. As a result, it is possible to process a document image input with an inclination more accurately.
[0015]
【Example】
Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings.
[0016]
FIG. 12 shows a block configuration of the character recognition apparatus in the embodiment. In the figure, 1 is a CPU that controls the entire apparatus, 2 is a ROM that stores a boot program and the like, and 3 is a RAM that stores a program executed by the CPU 1 and an operating system (OS). Reference numeral 4 denotes an external storage device such as a hard disk device, in which an OS, a program relating to character recognition processing, and a recognition dictionary (for Japanese characters and for English characters) are stored. Reference numeral 5 denotes an input device including a keyboard and a pointing device, and reference numeral 6 denotes an image scanner that reads a document image. An image memory 7 temporarily stores the read image, and a display device 8 displays various messages and recognized information.
[0017]
Before describing the processing of the embodiment in the above configuration, first, Japanese and English character strings will be considered.
[0018]
Characters in Japanese are generally written to their full height, whereas in English there are l and y, etc., so there is a line segment (ie, a dot) in the middle of the height divided into several parts. ) Concentrate. Therefore, it is possible to determine whether the character string to be recognized is Japanese or English based on the degree of concentration.
[0019]
An example of Japanese and English is shown in FIG. FIG. 7 (a) is an example of Japanese, and FIG. 7 (b) is an example of English, but black pixels are concentrated in areas (2) and (3) when English (b) is divided into four equal parts. On the other hand, Japanese is relatively distributed as shown in FIG. FIG. 5C shows the English when there is no character such as y protruding below. In this case, black pixels are concentrated in the areas (3) and (4).
[0020]
Based on the above principle, the operation process of the first embodiment will be described with reference to the flowchart of FIG. A program based on the flowchart is stored in the external storage device 4 and is executed by loading it onto the RAM 3.
[0021]
In this embodiment, the division number n when dividing one row into small areas is set to 4, and the calculation of the ratio of the black pixel distribution is determined as follows for the black pixel distributions R1 and R2.
[0022]
R1 = (B2 + B3) / (B1 + B4)
R2 = (B3 + B4) / (B1 + B2)
B1 to B4 indicate the number of black pixels in each region. Therefore, the black pixel distribution R1 is a value indicating how many black pixels the regions (2) and (3) have with respect to the regions (1) and (4), and R2 is the region (3), (4) means a value indicating how many black pixels the regions (1) and (2) have. Note that the number of black pixels occupied by a specific area is not limited to the above, and may be calculated using other scales. Now, a bitmap image input by an image input device such as a scanner is divided into blocks in step S101.
[0023]
An example of a binary bitmap image with one pixel and one bit is shown in FIG. In this embodiment, a simple binary image is preferable. The bit map image is formed in the form shown in FIG. 2B in the region separation process in step S101.
[0024]
A flowchart of an example of the region separation processing is shown in FIG. In step S301, a window of m × m pixel size is defined in the bitmap image, and if there is a predetermined number of black pixels (in this embodiment, “1”) in this window, the corresponding window is set to black and the resolution is set. Significantly reduce and concatenate character parts. FIG. 4 shows how the resolution is converted in step S301. Next, when the contour line is traced in step S302, it is possible to distinguish the elongated pattern peculiar to the character from other figures. In the last step S303, the regions can be separated as shown in FIG. 2B by combining adjacent text parts in the same group connection process.
[0025]
In the embodiment, in order to define each block, a data structure for block definition is determined. FIG. 5A shows the structure of the element “type” (short type) that defines the type of the block, and elements “startx” and “starty” that define the upper left corner position x-coordinate and y-coordinate of the block (each Short type), elements “width” and “height” (each short type) that define the width and height of the block, and an element “next_address” for storing the address of the next block.
[0026]
Here, the element “type” is assigned any numerical value from 0 to 2, and “0” indicates that the corresponding block is “title”, “1” indicates “text”, and “2” indicates “ Others (graphics, photos, etc.) ”
[0027]
Note that this is a method for determining the type of each block, but in the embodiment, when processing is performed with a reduced resolution as shown in FIG. 4 (at this time, the character patterns are connected to each other to form a lump of black pixel areas). The outline of the area is tracked, and it is determined whether or not it is an elongated series. If the area is elongated (the aspect ratio satisfies a predetermined condition), it is determined that the corresponding area is a character string area. Here, the character string can be divided into two types, title (headline) and body text. The former (title) generally has a large character size. Therefore, when the length of the cross section substantially perpendicular to the longitudinal direction of the elongated black pixel region determined to be a character region is greater than or equal to a predetermined length, it is determined as a title. In general, since the title is often above the body, it may be determined whether it is a title or text according to the position of the title. However, since the block determination itself is not a main part of the present invention, further explanation is omitted.
[0028]
FIG. 5B shows an example of data represented by the above structure. In the element next_address of each block, the address of the next block is stored. Finally, by substituting NULL, it is clearly indicated that there is no subsequent data.
[0029]
Returning to the description of FIG. When block data is extracted in step S101, the process proceeds to step S102 and loops until there is no unprocessed block (until the next_address of the structure becomes NULL). If there is no unprocessed block, the process is terminated. If there is an unprocessed block, the process proceeds to step S103.
[0030]
In step S103, if the block to be processed currently contains a character (element type is 0 or 1), the process proceeds to step S104, and if not, the process returns to 102.
[0031]
In step S104, the data in the block is shaded on the Y axis to extract the Y axis (row) where characters exist. 6 is performed on the text 2 in FIG. 2B as an example of the shading on the Y axis. A flowchart is shown in FIG. 9 in order to more specifically explain the bevel on the Y axis.
[0032]
Prior to the following description, the variables line_h [] and line_sy [] are both array variables secured in the RAM, line_h [] stores the height information of the character string for one line, and line_sy [] The relative y coordinate of the upper left corner of each row from the upper left corner coordinate of the block of interest is stored. A variable i represents a relative x coordinate in the block, and a variable j represents the same y coordinate. Further, flag is a variable indicating whether or not the cutting process for one line is being performed, and kuro stores information indicating whether or not there is a black pixel in one focused dot line. Is a variable. Nline is a variable that counts the number of character string lines in the block of interest.
[0033]
In step S901, the variables nline, flag, and j are each cleared to “0”. Then, the variable i is cleared to “0” in step S902 and the variable kuro is cleared in step S903.
[0034]
Next, the process proceeds to step S904, and it is determined whether or not the variable j exceeds the height of the target block (obtained by BLK [K] .height if the target block is k-th). If so, the processing for all the lines in the block of interest has been completed, and this processing is terminated.
[0035]
Therefore, the description will be continued here assuming that the value of the variable j is less than the height of the target block.
[0036]
In this case, the process proceeds to step S905, in which pixel data at a position where the x coordinate of the image data stored in the image memory 7 is startx + i and the y coordinate is starty + j is read, and it is determined whether or not it is a black pixel. Here, it can be understood that startx and starty are element names of the structure of the target block.
[0037]
If it is determined that the pixel at the position indicated by the variables i and j is not a black pixel, the process advances to step S906 to compare the variable i at that time with the width width of the block of interest. As a result, if it is determined that i <width, the process proceeds to step S907, and the variable i is incremented by “1”.
[0038]
Thus, when the variable i is sequentially incremented in the j-th line of the block of interest and it is determined that the pixel of interest is a black pixel, the process proceeds to step S908, and the variable kuro is substituted with “1”. It is indicated that at least a black pixel exists in the target line (line indicated by the variable j).
[0039]
Thereafter, the process proceeds to step S909 to determine whether or not the flag is “0”. A flag of “0” means that the previous line is a blank part, and a black pixel is not generated until the line indicated by the variable j. Therefore, in this case, the process proceeds to step S910, and the occurrence of the character string pattern is detected. Therefore, the variable j at that time is substituted into the array variable line_sy [nline]. Since the character pattern exists in a plurality of dot lines, “1” is assigned to flag so that the processing in step S910 is not performed on the next line.
[0040]
Further, when the presence of a black pixel is detected in the line indicated by the variable j, it is not necessary to detect the state of the pixel located on the right side of the line, so the process of step S907 is not performed. Assuming that the processing of the attention line has been completed, the process proceeds to step S912.
[0041]
In step S912, it is determined whether kuro = 0 and flag = 1.
[0042]
That is, it is determined whether or not a blank line is detected while lines with black pixels are being continuously detected. To explain in a more easy-to-understand manner, it is determined whether a character string pattern area for one line has been determined.
[0043]
If not, the process proceeds to step S913, the variable j is incremented by “1”, and the processes in and after step S902 are repeated.
[0044]
When it is determined that the character string pattern for one line has been detected in this way, the process proceeds to step S914, and the array variable line_h [nline] is set to a value obtained by subtracting line_sy [nline] from the variable j at that time. To do. As described above, since line_sy [nline] stores the y coordinate of the upper left corner of the character string pattern, line_h [nline] stores the number of lines of continuous black pixels, that is, the character string pattern. Height information will be stored.
[0045]
In step S915, the variable flag is cleared to “0” and nline is incremented by “1” in preparation for detection of the next character string pattern (character string row).
[0046]
As a result, the variable j eventually exceeds the height height of the block of interest, and this processing is terminated at that time. At this time, the number of character string lines detected in the block of interest is stored in nline, and the y coordinate value of the upper left corner of each character string pattern is stored in line_sy [0] to line_sy [nline], and line_h [0] to line [ nline] stores the height information of each character string pattern.
[0047]
The number of lines, the start point of each line, and the height can be extracted from the character image of each block by the processing of FIG.
[0048]
Returning to the description of FIG. In step S105, a loop is rotated for processing the extracted row data. If all the lines have not been processed, the process returns to step S102, otherwise the process proceeds to step S106.
[0049]
In step S106, as shown in FIG. 11, the rows extracted in step S104 are viewed in raster order, the number of black pixels in area (1) is counted to obtain B1, and then the number of black pixels in area (2) is determined. Count B2 is obtained, the number of black pixels in area (3) is counted to obtain B3, and finally the number of black pixels in area (4) is counted to obtain B4 (refer to FIG. 7 for the definition of areas 1 to 4).
[0050]
Here, each area means an area obtained by dividing the height indicated by the line height information line_h [] into four as described above.
[0051]
A flowchart of an example of the process is shown in FIG. As shown in FIG. 7, the upper left of the line image extracted in the line extraction process of step S104 is the coordinate (0, 0), the width of the line image is W, and the pixel height is H. Therefore, the lower right coordinate is (W-1, H-1). The following variables are also secured in the RAM 3.
[0052]
In step S1001, the black pixel counters counter (0) to counter (3) are cleared to “0”. Here, counter [0] is B1, coefficient [1] is B2, counter [2] is B3, and counter [3] is a variable for coefficient B4.
[0053]
In step S1002, the ordinate counter j for processing a two-dimensional image is reset. In step S1003, it is checked whether or not the ordinate counter j exceeds the line image height H. If it exceeds, the process proceeds to end and ends the process. If not, the process proceeds to step S1004.
[0054]
In step S1004, the abscissa counter i for processing a two-dimensional image is reset. In step S1005, it is checked whether or not the abscissa counter i exceeds the row image width W. If it exceeds, the process proceeds to step S1009, and if not, the process proceeds to step S1006. In step S1006, it is checked whether pixel [i] [j] is black. If black, the process proceeds to step S1007, and if white, the process proceeds to step S1008.
[0055]
In step S1007, counter [j / 4] is incremented because pixel [i] [j] is black.
[0056]
However, “j / 4” is valid only for the integer part. Therefore, j / 4 is one of four values 0, 1, 2, and 3, and any one of counters [0] to [3] is incremented.
[0057]
That is, the black pixel in the small region (1) is added to counter [0], the black pixel in the small region 2 is added to counter [1], and the black pixel in the small region 3 is added to counter [2]. Then, the black pixels in the small area 4 are added to counter [3].
[0058]
In step S1008, the abscissa counter i is incremented, and the process returns to step S1005. In step S1009, the ordinate counter j is incremented, and the process returns to step S1003.
[0059]
As a result of the above processing, black pixels in each small area are counted as B1 = counter [0], B2 = counter [1], B3 = counter [2], and B4 = counter [3].
[0060]
Returning to the description of FIG. In step S107, whether (B2 + B3) / (B1 + B4) is greater than or equal to a preset threshold value T1, or (B3 + B4)/ (B1 + B2)Is greater than or equal to a preset threshold value T2. If the determination of any of these is affirmed, the target line of the target block is determined to be an English character string, so the attribute for that line is set to “English” in step S108. If both determinations are negative, the attention line is determined to be in Japanese, and an attribute to that effect is given.
[0061]
As described above, according to the present embodiment, it is possible to determine whether English or Japanese is used for each line, so that it is possible to select an appropriate recognition dictionary in the subsequent character recognition processing, thereby improving the recognition rate. It becomes possible. In some cases, the recognition algorithm itself may be switched.
[0062]
<Description of the second embodiment>
In the first embodiment, the alphabet / Japanese discrimination is performed for each line of the character image.
[0063]
For example, if the first line of a character image is extracted, the black pixel distribution of the first line is calculated, and alphabet Japanese is determined, the result may be used as the determination result for the entire image.
[0064]
Alternatively, the character image is cut by shading on the Y axis, the character is cut by shading on the X axis, and m characters are arbitrarily selected, and the average of the black eye distribution of the m characters is calculated to calculate the entire character image. There is also a method to determine Japanese alphabet.
[0065]
By these methods, calculation time can be shortened.
[0066]
Further, in the first embodiment described above, the line extraction means is a shadow on the Y axis with respect to the character image. However, the present invention is not limited to this, and for example, it may be performed by area separation in step S101 in FIG. Specifically, after the resolution conversion in step S301 in FIG. 3 and the contour line tracking in step S302, the object extracted by the contour line tracking is determined as a character, but the same group combination in step S303 is determined. Instead, do region separation as rows. The result of executing FIG. 2A by this processing is as shown in FIG. 8, and no shading on the Y axis is necessary. However, since the accuracy of the extracted row is low due to the effect of resolution conversion, it is necessary to tune the value of the threshold value T in consideration of this. Since this threshold value T depends on the document image, it is adjusted from the input device 5 as appropriate.
[0067]
Also, if there is an isolated pixel that is originally one pixel in the blank area as noise in the input document image, the above processing may not be performed normally. Therefore, it is desirable to provide a process for determining an isolated pixel and removing it. The isolated pixel may be determined by determining whether there is another black pixel within a predetermined distance around the black pixel. Alternatively, a method of winding an image and deleting it by pattern matching may be used.
[0068]
In the first embodiment, the image subjected to the region separation process in step S101 is a 1-pixel 1-bit image, but is not limited thereto, and may be, for example, a 1-pixel 8-bit multi-value image. In that case, there is a method of extracting a high frequency component by applying a differential filter and dividing it into a character part and a photograph part. In order to carry out this alphabet Japanese discrimination, the multi-valued image of 8 bits per pixel may be binarized with a certain threshold value.
[0069]
Further, by correcting the inclination of the character image before performing the alphabet / Japanese automatic discrimination method, a better result can be obtained for an image in which the document is inclined. The tilt correction can be realized, for example, by obtaining the tilt of the extracted line and performing coordinate conversion, and thus description thereof is omitted.
[0070]
Further, by detecting the direction of the character image before applying this alphabet / Japanese automatic discrimination method, a result independent of the direction of the original can be obtained. For example, the character image direction is detected by performing line extraction and character extraction on the character image extracted by area division, and performing OCR by rotating the extracted several images by 0 °, 90 °, 180 °, and 270 °. Since OCR certainty can be obtained, the direction of the character image can be detected and the original image can be obtained.
[0071]
Further, although the number of black pixels in each small area is counted after extracting the row image, it is not limited to this. For example, the number of black pixels in the small area may be counted at the same time when the row image is extracted by performing shading in the Y-axis direction. In that case, since the height of the row image is still unclear, the number of black pixels in the row can be held and the black pixels in the small region can be obtained after the height is known.
[0072]
<Description of the third embodiment>
In the above embodiment, the line of the character string is divided into four equal parts, and whether the corresponding line is Japanese or English is determined by the ratio of the number of black pixels in the line. In the third embodiment, An example of determining the character type will be described by paying attention to the fact that Japanese character patterns generally have more dots per unit area than English.
[0073]
The apparatus configuration is the same as that of the first embodiment described above, and the processing content will be described here.
[0074]
The processing procedure in the third embodiment will be described with reference to the flowchart of FIG.
[0075]
In the flowchart of FIG. 13, steps S3101 to S3105 are the same as steps S101 to S105 of FIG. That is, it is assumed that the processes related to block formation and detection of the number of lines of text or headings in each block are the same.
[0076]
In steps S <b> 3101 to S <b> 3105, the document image is input and divided into blocks, and the target block is a title or text, and the line is extracted. If it is determined that there is an unprocessed line, the process proceeds to step S3106.
[0077]
In step S3106, the black pixels in the row are viewed as shown in FIG. 14, and the character width is detected and the number of black pixels is counted.
[0078]
FIG. 15 shows a flowchart of the processing content of step S3106, which will be described below.
[0079]
As shown in FIG. 14, the upper left position of the row image extracted by the row extraction process in step S3104 (see FIG. 9) is the coordinate (0, 0), the pixel width of the row image is IW, and the pixel height is Let H be. Therefore, the lower right coordinate is (IW-1, H-1). The details of this process will become clear from the following description. In short, the coordinates of the start position (line head) and end position (line end) of the position where the actual character string pattern of the line of interest exists, A process of counting the number of dots between them is performed.
[0080]
When the start position of the black pixel is detected in step S4001, the variable startx (x coordinate) that stores and holds the position is updated, and the variable flag_blk that is controlled so as not to update the variable startx in the subsequent processing is set to “ Clear 0 ”. In step S4002, the variable contour for counting the number of black pixels is cleared to “0”. In step S4003, the variable i for specifying the x-coordinate position of the target pixel position is cleared to “0”. In the next step S4004, the contents of flag_line are substituted into flag_line2, and then flag_line is cleared to “0”. In the first stage, it is assumed that “0” is stored in flag_line. Accordingly, flag_line2 is cleared to “0” in the first stage.
[0081]
Further, flag_line stores information indicating whether or not there is a black pixel in the vertical 1-dot column indicated by the current variable i, and flag_line2 stores information indicating whether or not there is a black pixel in the immediately preceding vertical 1-dot column. . Therefore, if the character pattern is interrupted while black pixels are being continuously detected, flag_line2 = 1 (there is a black pixel in the immediately preceding vertical one dot row) and flag_line = 0 (notice It is only necessary to determine that there is no black pixel in the vertical one dot row.
[0082]
In step S4005, the variable j that stores and holds the y coordinate of the pixel position to be extracted is cleared to “0”.
[0083]
In step S4006, the pixel at the position indicated by the variables i and j in the target row is read from the image memory 7 to determine whether it is a black pixel. If it is a white pixel, the process jumps to step S4012 to increment the variable j, that is, the y coordinate by one, and in step S4005 until it is determined in step S4013 that the variable j is equal to or lower than the height of the target line. The following process is repeated.
[0084]
Thus, if no black pixel is found for one vertical dot row indicated by the variable j, the determination in step S4013 is “NO”, so the flow proceeds to step S4014, flag_line2 is “1”, and It is determined whether or not flag_line is “0”. That is, it is determined whether the end of the character pattern has been found. If not, the variable i is incremented to prepare for the next column position. In step S4017, the variable i is compared with the line width. If it is determined that the process for the line width has not been completed, the process returns to step S4004 and the above process is repeated.
[0085]
In this process, when a black pixel is detected for the first time, it is determined in step S4007 that flag_blk is “0”. Therefore, the process proceeds to step S4008, and the value of the variable i at that time is set to the variable startx. substitute. As a result, the x coordinate of the first dot position of the character string pattern is stored in startx.
[0086]
Next, “1” is substituted for the variable flag_blk so that the process of substituting for startx is not performed thereafter.
[0087]
Next, since one black pixel is detected, the variable counter is incremented by “1”, and since a black pixel is detected in the vertical column of interest, “1” is assigned to flag_line. Then, the process proceeds to step S4012.
[0088]
As a result, when the first black pixel in the target row is detected, the variable i which is the x coordinate value at that time is substituted for the variable startx, and this process is not performed thereafter. Each time a black pixel is detected, the variable counter is incremented by “1”.
[0089]
Now, in the process of detecting the black pixel of one character, if the process moves between the adjacent character patterns, there is naturally a blank portion between them, so that no black pixel is detected.
[0090]
At this time, the determination in step S4014 is “Yes”. That is, since the value of the variable i at that time is the end position of the character pattern of interest, the value is substituted into endx, assuming that the position is the temporary line end position. Accordingly, the content of endx is updated by the number of characters included in the target line (see FIG. 14), but the x coordinate value at the end of the character at the end position of the target line is finally stored in endx. Become.
[0091]
As a result, the start position of the initial character pattern of the target line in the target block is stored in startx, the end of the end-of-line character pattern is stored in endx, and the number of black pixels between them is stored in counter.
[0092]
Returning to the description of FIG.
[0093]
In step S3107, the number of white pixels (WN) is set.
WN = row height (H) × row width (W) −number of black pixels (BN)
To obtain the black pixel ratio R.
R = number of black pixels (BN) / number of white pixels (WN)
Ask for.
[0094]
Here, the row height is as described in the first embodiment, and the row width is a value obtained by endx-startx.
[0095]
It is checked whether or not R is larger than the threshold value T determined in advance in step S4108 (whether there are many black pixels). If R> T, the process proceeds to step S3109. If R <= T, the process proceeds to step S3110. move on. In step S3109, it is determined that the character image is Japanese, and the process returns to step S3105. In step S3110, it is determined that the character image is an alphabet, and the process returns to step S3105.
[0096]
As described above, according to the third embodiment, the character pattern has a large number of dots when it is Japanese, and conversely, it has a small number of dots when it is English. The character type of each line can be determined. Therefore, when moving to the character recognition process, it is possible to determine the recognition dictionary in advance or to determine a priority dictionary, and to increase the character recognition rate.
[0097]
<Description of the fourth embodiment>
In the third embodiment, alphabetic Japanese discrimination is performed for each line of the character image, but this is not limited. For example, if the first line of a character image is extracted, the black pixel distribution of the first line is calculated, and alphabet Japanese is determined, the result may be used as the determination result for the entire image.
[0098]
Alternatively, a character image is line-cut by shading to the X axis, and after character cutting by shading to the Y-axis, m characters are arbitrarily selected, an average of R of m characters is calculated, and the entire character image is calculated. There is also a method to perform alphabet Japanese discrimination. By these methods, calculation time can be shortened.
[0099]
Further, the character type may be determined by combining the first embodiment and the third embodiment described above. If it does in this way, it will become possible to improve the precision about character type judgment.
[0100]
In addition, there are various fonts and styles for characters. Therefore, if Japanese is a narrower font and English is a thicker font, there is a risk of misclassification. Therefore, if the thin line processing of characters is performed before the processing as in the first embodiment, the risk of such erroneous determination is eliminated. However, a strict thinning process involves a very complicated process. However, in the third embodiment, since the character types are simply made unambiguous, simple thinning processing may be performed.
[0101]
Hereinafter, simple thin line processing will be described.
[0102]
Simple thin line processing is possible by erasing black pixels by pattern matching. Focusing on two images, for example, in units of 3 × 3 pixels, if the predetermined erase pattern matches the image pattern in the window, the target pixel (center pixel in the window) is forcibly converted from black to white. The thin line process is performed by repeating the process. An example of the erase pattern is as shown in FIG.
[0103]
Further, in the third embodiment, the line extracting means is a shadow on the Y axis with respect to the character image. However, the present invention is not limited to this. For example, the line extracting means may be performed by area separation in step S3101 in FIG.
[0104]
Specifically, as in the first embodiment, after performing resolution conversion and contour tracking, an elongated object extracted by contour tracking is determined as a character. Instead, do region separation as rows. The result of executing FIG. 2 (a) by this processing is as shown in FIG. However, since the accuracy of the extracted row is low due to the influence of resolution conversion, the value of the threshold value T must be tuned in consideration thereof.
[0105]
Further, in the third embodiment described above, if there is noise in a 1-pixel 1-bit image to be subjected to this processing, the alphabet-Japanese discrimination ability cannot be utilized. Therefore, it is better to remove isolated dots that are clearly recognized as noise. As a method of removing the isolated dots, there is a method of winding an image and deleting it by pattern matching.
[0106]
In the third embodiment described above, the image subjected to the region separation process in step S3101 in FIG. 13 is a 1-pixel 1-bit image. However, the image is not limited thereto, and may be, for example, a 1-pixel 8-bit multi-value image. In that case, there is a method of extracting a high frequency component by applying a differential filter and dividing it into a character part and a photograph part. In order to carry out this alphabet Japanese discrimination, it is necessary to binarize the multi-valued image of 8 bits per pixel with a constant threshold value.
[0107]
Further, by correcting the inclination of the character image before performing the alphabet automatic Japanese discrimination method, a better result can be obtained for an image in which the document is inclined. The inclination correction can be realized by, for example, obtaining the inclination of the extracted line and performing coordinate conversion.
[0108]
In addition, by detecting the direction of the character image before applying the automatic alphabet / Japanese discrimination method, a result independent of the direction of the document can be obtained. For example, the character image direction is detected by performing line extraction and character extraction on the character image extracted by area division, and performing OCR by rotating the extracted several images by 0 °, 90 °, 180 °, and 270 °. Since the certainty of OCR can be obtained, the direction of the character image is detected and the original image is used.
[0109]
As described above, according to the first to fourth embodiments, it is possible to determine the character type according to the distribution or density of Japanese and English character patterns. Therefore, it is possible to determine a dictionary to be used when recognizing as a pre-process for character recognition, or to determine the priority order thereof, so that the character recognition rate can be improved.
[0110]
In the first to fourth embodiments, both Japanese and English have been described. However, the same applies when German or French is used instead of English, and a kanji range (for example, instead of Japanese) Therefore, the present invention is not limited to the above example.
[0111]
In the embodiment, the image input from the image scanner 6 is processed. However, the image may be sent via a communication line, for example, on a predetermined storage medium (for example, a floppy disk). Since the same processing can be performed even if the image is read out from the stored image, it is not limited by the above embodiment.
[0112]
Furthermore, the present invention may be applied to a system composed of a plurality of devices or an apparatus composed of a single device. Needless to say, the present invention can also be applied to a case where the present invention is achieved by supplying a program to a system or apparatus.
[0113]
【The invention's effect】
As described above, according to the present invention, it is possible to increase the processing accuracy in lower-order processing, for example, character recognition processing, by efficiently determining the character type in the input document image.
[0114]
Further, according to a preferred embodiment of the present invention, the sub-process is a character recognition process, and a recognition algorithm used when performing a recognition process according to a result determined by the determination means, or a recognition algorithm Since the recognition dictionary or the recognition dictionary is selected, the recognition speed and the recognition accuracy can be increased.
[0115]
In addition, the dot existence ratio with respect to a substantially central area of each divided area is calculated, and when the calculated ratio is greater than or equal to a predetermined value, the determination unit determines that the attention line is an alphabetic character, In the following cases, since it is determined that the character is a kanji-speaking character, it becomes possible to accurately determine a kanji-speaking character and an alphabetic character.
[0116]
In addition, since the line is determined based on the dot distribution projected in the arrangement direction of the character string pattern, the line can be determined with high accuracy.
[0117]
Furthermore, since noise removal means for removing noise in the input document image can be provided, the influence of noise can be eliminated, so that the line can be determined with higher accuracy.
[0118]
Further, by providing means for rotating the input image by a predetermined angle to determine the direction as an erect image, it is possible to make a determination according to the input direction of the document image.
[0119]
In addition, since the tilt is corrected when the input image is tilted, it is possible to process a document image input tilted with higher accuracy.
[0120]
[Brief description of the drawings]
FIG. 1 is a flowchart illustrating a processing procedure in an embodiment.
FIG. 2 is a diagram illustrating a relationship between a document image and a block dividing process.
FIG. 3 is a flowchart illustrating an example of block division processing according to the embodiment.
4 is a diagram illustrating a state in which the resolution conversion process in FIG. 3 has been performed.
FIG. 5 is a diagram illustrating a structure of a block variable structure and an example of data thereof in the embodiment.
FIG. 6 is a diagram showing a concept of row discrimination.
FIG. 7 is a diagram for explaining an example of Japanese and English character strings and the principle of character identification;
FIG. 8 is a diagram illustrating an example of an attribute of each block after the block dividing process according to the embodiment.
FIG. 9 is a flowchart of row identification processing according to an embodiment.
FIG. 10 is a flowchart illustrating details of pixel counting processing according to the embodiment.
FIG. 11 is a diagram illustrating scanning contents of pixel counting processing according to the embodiment.
FIG. 12 is a diagram illustrating a block configuration of a character recognition device according to an embodiment.
FIG. 13 is a flowchart showing an operation processing procedure of the third embodiment.
FIG. 14 is a diagram showing an outline of a processing process in a third embodiment.
FIG. 15 is a flowchart showing black pixel distribution calculation processing in the third embodiment;
FIG. 16 is a diagram illustrating an example of a pattern used in thinning processing.
[Explanation of symbols]
1 CPU
2 ROM
3 RAM
4 External storage device
5 input devices
6 Image scanner
7 Image memory
8 display devices

Claims (10)

An image processing apparatus that determines the type of characters in an input document image and passes it to a lower-level process.
A discriminating means for discriminating a region of the character line including the character image in the input document image;
Dividing means for dividing the determined character line area into four areas arranged in a direction perpendicular to the direction of the character line;
Counting means for counting significant dots in each region divided by the dividing means ;
When the significant number of dots in each area counted by the counting means is B1, B2, B3, and B4 in order from the upper area , the first dot distribution R1 is calculated based on the ratio of (B2 + B3) to (B1 + B4). Calculating means for calculating a second dot distribution R2 based on a ratio of (B3 + B4) and (B1 + B2) ;
When it is determined that the calculated first dot distribution R1 is greater than or equal to the first threshold T1, or the calculated second dot distribution R2 is greater than or equal to the second threshold T2, the character line area is Judge that it is an alphabetic character line,
When it is determined that the first dot distribution R1 is smaller than the first threshold value T1 and the second dot distribution R2 is smaller than the second threshold value T2, the character line area is a Chinese character area character line. An image processing apparatus comprising: determination means for determining. The lower-level processing is character recognition processing, and is a recognition used when character recognition processing is performed on the character line area , depending on whether the determination unit determines that the character line area is an alphabetic character line or a kanji character line. The image processing apparatus according to claim 1, wherein a dictionary is selected. The lower-level processing is character recognition processing, and is a recognition used when character recognition processing is performed on the character line area , depending on whether the determination unit determines that the character line area is an alphabetic character line or a kanji character line. The image processing apparatus according to claim 1, wherein an algorithm is selected.   The image processing apparatus according to claim 1, wherein the determination unit determines a line based on a dot distribution projected in a character line direction.   Before the character line area is determined by the determining means, a noise removal process for removing noise in the input document image, a tilt correction process for correcting a tilt of the input document image, and in the input document The image processing apparatus according to claim 1, wherein at least one of the processes for determining a direction in which the character is upright is performed. An image processing method for determining a character type in an input document image and passing it to a lower-level process,
A discriminating step for discriminating a region of a character line including a character image in the input document image;
A dividing step of dividing the determined character line area into four areas arranged in a direction perpendicular to the direction of the character line;
A counting step of counting significant dots in each region divided in the dividing step ;
When the significant number of dots in each area counted in the counting step is B1, B2, B3, and B4 in order from the upper area , the first dot distribution R1 is set based on the ratio of (B2 + B3) to (B1 + B4). A calculation step of calculating and calculating a second dot distribution R2 based on a ratio of (B3 + B4) and (B1 + B2) ;
When it is determined that the calculated first dot distribution R1 is greater than or equal to the first threshold T1, or the calculated second dot distribution R2 is greater than or equal to the second threshold T2, the character line area is Judge that it is an alphabetic character line,
When it is determined that the first dot distribution R1 is smaller than the first threshold value T1 and the second dot distribution R2 is smaller than the second threshold value T2, the character line area is a Chinese character area character line. An image processing method comprising: a determination step of determining. The lower-level process is a character recognition process, and the recognition used when the character line area is character-recognized according to whether it is determined as an alphabet character line or a kanji-range character line in the determination step. 7. The image processing method according to claim 6 , wherein a dictionary is selected. The lower-level process is a character recognition process, and the recognition used when the character line area is character-recognized according to whether it is determined as an alphabet character line or a kanji-range character line in the determination step. The image processing method according to claim 6 , wherein an algorithm is selected. The image processing method according to claim 6 , wherein in the determination step, a line is determined based on a dot distribution projected in a character line direction. Before determining the character line area in the determining step, a noise removal process for removing noise in the input document image, a tilt correction process for correcting the tilt of the input document image, and in the input document The image processing method according to claim 6 , wherein at least one of the processes for determining the direction in which the character is erect is performed.

Images