JPH04268685A - Method for discriminating type of slips - Google Patents

Abstract

PURPOSE:To provide the method stably discriminating the type of slips based on ruled line information forming the table of slips. CONSTITUTION:The horizontal and vertical line segments of a table constituting a character reading frame are extracted from inputted slip picture data. Vector patterning is performed by using the direction, length, and position of the line segments extracted for each area so as to comparatively collated with the feature vector of the reference pattern.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for determining the type of a document when reading a plurality of tabular documents.

0002

[Prior Art] Conventionally, when reading characters with an optical character reading device for documents including tables, a reading frame indicating the position of writing characters is printed on the documents with vertical and horizontal ruled lines. ing. An optical character reader detects and cuts out a character pattern at the reading frame position from a form image obtained by optically scanning a form, performs character recognition processing on the cut out character pattern, and recognizes the characters written on the form. is configured to read. At this time,
The position information of the reading frame is set in advance in the optical reading device for each form. If multiple types of documents with different reading frame positions are to be scanned, a barcode or number (document identification mark) to identify the documents is used.
An area is provided on a form, a bar code or number within this area is read to determine the type of form, and character reading processing is performed using reading frame position information corresponding to the type of form.

0003

[Problem to be Solved by the Invention] However, in the conventional method, when performing character reading processing on a mixture of multiple types of forms with different character reading frame structures, it is necessary to provide a method for determining the type of form for each form. It was necessary to provide a barcode or number area on the form. When reading a mixture of multiple types of forms, only the forms that have this area are targeted; if the position of this area is different for each form, or if the barcode or number area is When a document that is not provided is to be read, it is difficult to determine the type.

[0004] The present invention has been made with attention to these points, and even when forms with different areas are mixed together, or even when forms do not have such areas, It is an object of the present invention to provide a method for discriminating types of forms based on ruled line information forming a table of forms so as to stably discriminate the types of forms.

[0005]

[Means for Solving the Problems] The present invention relates to a method for determining the type of documents, and the above object of the present invention is to input image information of documents to be recognized, and to use the input image data horizontally and vertically. Line segments in the direction are extracted, information on the position, length, and direction of each extracted line segment is detected, the input image is divided into a plurality of areas, and the extracted line segments that exist in each divided area are The feature quantity, which is the weight for each length in each direction of the line segment, is determined as the pattern feature quantity of the object to be recognized. The feature amount of the recognized pattern and the feature amount of the stored standard pattern are compared and matched, and based on the comparison result, it is determined which standard pattern the recognized pattern is, and the type of the form is determined. This is achieved by

[0006] Furthermore, inputting image information of documents to be recognized,
Line segments in the horizontal and vertical directions are extracted from this input image data, information on the position, length, and direction of each extracted line segment is detected, and the input image is divided into multiple areas. The feature quantity, which is the weight for each length of each direction of the extracted line segment existing in each area, is determined as the pattern feature quantity of the recognition target, and the feature quantity of the standard pattern, which is the standard of the pattern to be recognized, is determined. are stored in advance, the extracted feature values of the recognized pattern are compared and matched with the feature values of the stored standard pattern, and those that are similar by more than a predetermined value are extracted as candidates in the order of similarity, and the candidates are This is achieved by detecting input image information corresponding to a predetermined specific area of a standard pattern extracted as a standard pattern, and determining the type of form by checking whether it matches the standard pattern in the order of the candidates. be done.

[0007]

[Operation] The present invention is applicable when performing character reading processing on a mixture of multiple types of forms, such as forms that do not have a bar code or number (form identification mark) area, or forms that are provided in different positions. This method relates to a tabular pattern recognition method for distinguishing the type of form, and uses input form image data to identify the horizontal and vertical line segments (ruled lines) that are the constituent elements of the table that make up the character reading frame. It is extracted and divided into a plurality of areas, and the direction, length, and position of the line segment extracted for each divided area are used to create a vector pattern, which is then compared and verified with the feature vector of a standard pattern. In addition, in the invention of claim 2, the direction, length, and position of the line segment extracted for each divided area are used to compare and match the pattern vector converted into a vector pattern and the feature vector of the standard pattern. Candidates are extracted and detailed confirmation is performed in the order of the candidates.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows an example of a form to be recognized by the present invention, and FIG. 2 is a flowchart of the recognition algorithm of the present invention, and the recognition operation will be explained with reference to this flowchart.

First, an image of a form is input using image input means such as an image sensor (step S1), and the size of the form is detected (step S2). The size of the form is detected by detecting the edges of the form using the gray level of the input image. Next, the image data is preprocessed, and this preprocessing is performed by binarizing and smoothing (step S3). As shown in FIG. 3, the smoothing process of the binary image is executed by performing hole filling process or isolated point removal process in the 8-neighborhood area 3 for the pixel of interest 2. In other words, when a black pixel is set to "1" and a white pixel is set to "0", the hole-filling process is performed so that the pixel of interest 2 is set to "0".
0" and all of the 8 neighboring pixels 3 are "1", the value of the pixel of interest 2 is changed from "0" to "1". In addition, the isolated point removal process is performed when the pixel of interest 2 is "1". And when all of the eight neighboring pixels 3 are "0", the value of the pixel of interest 2 is changed from "1" to "0".

Next, for the preprocessed (binarized, smoothed) image, count the number of consecutive black pixels in the horizontal or vertical direction, and calculate the number of consecutive black pixels in the horizontal or vertical direction for each pixel. The value (hereinafter referred to as run length) is determined, and the ruled line is extracted using this run length (step S4). This operation flow is shown in FIG. 4. Explain.

First, to explain the extraction of horizontal ruled lines, horizontally continuous straight lines are extracted from the preprocessed image data to detect the horizontal run length (step S30). Then, among the extracted horizontal run lengths, pixels with extremely short run lengths, for example, less than 5 mm, are treated as noise and invalid data and are deleted.
Pixels having the above run length are treated as valid and the horizontal run length image is binarized (step S31). FIG. 5A shows the state of this image data. Next, connection processing is performed based on this image data. To explain the connection process specifically, the horizontal peripheral distribution feature, which is a horizontal histogram of the image data in FIG. 5(A), is obtained as shown in FIG. 5(B) (step S32), and the peripheral distribution feature is A portion whose height exceeds the threshold value TH1 is set as a ruled line candidate position (step S33). (A) and (B) in Figure 5
), three candidates of ruled line candidate positions X, Y, and Z are extracted. Then, for each ruled line candidate position, a vertical peripheral distribution, which is a vertical histogram within the candidate area, is determined (step S34). FIG. 5C shows how the vertical peripheral distribution for the ruled line candidate position Y in FIG. 5A is obtained. As shown in (D) of FIG.
(pixel), and the portion exceeding the threshold value TH2 is considered valid. The interval between these effective line segments is a predetermined value TH3 (for example, TH
(3=5 pixels) or less, a connection process is performed and the ruled lines of the image input data are corrected by treating them as one connected straight line. FIG. 6A shows how this process is performed for the form shown in FIG. In addition, here, Figure 5
The average value of the vertical peripheral distribution feature values in each horizontal ruled line section as shown in (D) is set as the width of that ruled line. The line width is used as information when cutting out only the inside of the area surrounded by lines when cutting out a character reading area from an image. Vertical ruled lines can also be obtained by similar processing. Steps S20 to S25 in FIG. 4 show extraction of vertical ruled lines. FIG. 6(B) shows how vertical ruled lines are extracted for the form in FIG. 1, and FIG. 6(C) shows how the ruled line extraction is completed.

After the ruled lines are extracted using the run length as described above, vertical and horizontal ruled line information (position (starting and ending point coordinates), length, direction (vertical, horizontal)) is obtained for each ruled line extracted from the image.
(Step S40). In the above steps, the image input data has been corrected, and the ruled lines have been extracted and information about each extracted ruled line has been reliably detected. Hereafter, processing is executed based on this vertical and horizontal ruled line information.

Next, a pattern vector of the form is determined based on the vertical and horizontal ruled line information (step S5 in FIG. 2). The pattern vector will be explained in detail later, but to give an overview here, the pattern vector is as shown in Figure 7, and P[m][n][t][w] (however, , m is the vertical area number (1 to M), n is the horizontal area number (1 to N), and w is the ruled line length level (1 to W
), t is the direction of the ruled line, t=1 is the horizontal direction, t=2 is the vertical direction), and is an M×N×W×2-dimensional vector. In other words, for each M×N area, the degree to which vertical and horizontal ruled lines are included for each ruled line length level (expressed by quantizing in W steps from short to long) ( (hereinafter referred to as weight). FIG. 8 is a flowchart showing the operation to obtain this pattern vector P. First, for each ruled line ki (where i is the ruled line number), the initial point of the ruled line (Xs
i, Ysi), the end point (Xei, Yei), and the length Li, find the weight of the ruled line ki at the level of the length of the ruled line ki in each area in which the ruled line ki is included, and add it sequentially. The pattern vector P is determined by

The operation for obtaining a pattern vector will be explained in detail below with reference to FIG.

First, the input form 4 is divided into M×N areas as shown in FIG. 9 (step S50). Next, the pattern vector is initialized (step S51), and it is determined whether all ruled lines have been processed (step S52). If the ruled line ki has not been processed, the length quantization level of the unprocessed ruled line ki is determined (step S53). The length quantization level represents the length of a ruled line by quantizing it in W steps. The number of quantization levels is W from level 1 to W.
It is expressed in stages, and so that the level of the ruled line connecting both ends of the form is W, horizontal ruled lines are normalized by the form width L0, and vertical ruled lines are normalized by the form height h0. If the length of the ruled line ki is Li, then the length quantization level Wi of the ruled line ki
is determined by the following formula

[0016]

[Equation 1] However, for horizontal ruled lines, a0=L0, for vertical ruled lines, a0=h
It is 0. ceil[X] indicates the smallest integer not less than X. Here, a0/W is a length corresponding to one length quantization level.

Next, in order to detect the area where the ruled line ki exists, the starting point area and end point area of the ruled line are determined (steps S54 and S55). The starting point area and ending point area are determined based on the starting point (Xsi,
Ysi) and the coordinates of the end point (Xei, Yei). First, the vertical position of the starting point area of the ruled line is determined by the following equation.

##EQU00002## where y=(Ysi+Yei)/2 for horizontal ruled lines, y=Ysi for vertical ruled lines, h0 is the height of the form, and M is the number of area divisions in the vertical direction of the form.

Here, h0/M corresponds to the height of one area. Further, the lateral position is determined by the following equation.

[0019]

[Equation 3] However, when it is a horizontal ruled line, x=Xsi, and when it is a vertical ruled line, x=(X
si+Xei)/2, L0 is the width of the form, and N is the number of area divisions in the horizontal direction of the form.

[0020] Here, L0/N corresponds to the width of one area. In the example of ruled lines k1 and k2 shown in FIG. 10, the starting point area for ruled line k1 is (Ms1, Ns1) = (1, 2), and the starting point area for ruled line k2 is (Ms2, Ns2) = (
2,1).

Next, the vertical position of the end point area of the ruled line ki is determined by the following equation.

[0022]

[Equation 4] However, for horizontal ruled lines, y=(Ysi+Yei)/2, and for vertical ruled lines, y=Yei. Also, the horizontal position is determined by the following equation.

[0023]

[Equation 5] However, when it is a horizontal ruled line, x = Xei, and when it is a vertical ruled line, x = (X
si+Xei)/2 In the example of ruled lines k1 and k2 shown in FIG. 10, the end point area of ruled line k1 is (Me1, Ne1)=(
1, 2), the end point area for ruled line k2 is (Me2,
Ne2)=(2,4).

At this stage, the range of elements in which the pattern vector P is updated by the ruled line ki has been determined. That is, the range in which the pattern vector P is updated by the ruled line ki is from the area (Msi, Nsi) to (Mei, Nsi).
This element corresponds to the length level Wi of the ruled line ki in the direction of the ruled line ki (horizontal ruled line or vertical ruled line type) in the area of the section up to ei). For example, in the ruled line k1 in FIG. 10, the starting point and ending point are in the same area (1, 2) and is a horizontal ruled line, so
The updated element of pattern vector P is P[Ms1]
[Ns1][1][W1](=P[1][2][1][
W1]). In the example of ruled line k2 in FIG. 10, the start point area and end point area are different and it spans multiple areas and is a horizontal ruled line, so the updated elements of pattern vector P are P[Ms2][Ns2][1][ W2] to P[Me2][Ne2][1][W2],
That is, P[2][1][1][W2], P[2][2
][1][W2],P[2][3][1][W2],P
There are four items: [2], [4], [1], and [W2]. Note that the weight with which the pattern vector P is updated will be described later.

Next, the weight by which the pattern vector P is updated by the ruled line ki is calculated, and the pattern vector P is updated. When a ruled line falls within one area (ruled line k in Figure 10)
1) and when spanning multiple areas (in the case of ruled line k2 in FIG. 10), the weight update method is different, so first, it is determined whether the starting point area and end point area found for ruled line ki match. (Step S56). If they are in the same area, it is determined that the ruled line ki falls within one area and weight addition processing 1 (step S57) is performed; if they do not match, it is determined that the ruled line ki spans multiple areas and weighted addition is performed. By performing process 2 (step S58), the pattern vector P is updated. Weight addition processing 1 and weight addition processing 2 will be described later. In this way, by updating the pattern vector P using the ruled line ki and sequentially updating the pattern vector P using all the ruled lines (steps S59, S52), the ruled line information of the input form can be converted into a pattern vector.

Next, weight addition processing 1 (step S5 in FIG. 8)
The operation of 7) will be explained. This process updates the pattern vector P based on ruled lines existing in one area, as in the example of ruled line k1 in FIG. 10. 1
The sum of the weights by which the pattern vector P is updated by the ruled lines of the book is always normalized as "1". Due to this,
Since only one element of the pattern vector is updated in weight addition process 1, the weight update amount at this time is a0=1
becomes. Therefore, the starting point area (Msi, Ns
i) and the end point area (Mei, Nei) are equal, and the ruled line k
When it is determined that i exists within one area, the pattern vector P is updated using the following equation 6.

[0027]

[Formula 6] P[Msi][Nsi][t][Wi]=P[
Msi][Nsi][t][Wi]+a0 However, for horizontal ruled lines, t=1; for vertical ruled lines, t=2, a0=1 Next, the operation of weight addition processing 2 will be explained. This process updates the pattern vector P based on a ruled line that spans a plurality of areas, such as the example of ruled line k2 in FIG. As described in the weight addition process 1, the sum of the weights by which the pattern vector P is updated by one ruled line is always standardized as "1". Therefore, in the weight addition process 2, the ratio of the length of a ruled line that spans a plurality of areas included in each area to the entire ruled line is used as the update amount of the weight of each area. Thereby, the total sum of the weights of the pattern vector P updated by one ruled line can all be set to "1". FIG. 11 is a flowchart showing the operation of this weight addition process 2, and will be specifically explained below with reference to FIG.

The ruled lines handled in this weight addition process 2 are as shown in FIG.
It spans multiple areas like the ruled line k2. First, the weight as is determined for the starting point area (step S60). The ratio of the length of the portion included in the starting point area of the ruled line ki to the entire length of the ruled line ki is determined as a weight. Specifically, if the ruled line ki is a horizontal ruled line, it can be determined by Equation 7, and if the ruled line ki is a vertical ruled line, it can be determined by Equation 8.

[0029]

[Math 7]

[0030]

[Equation 8] In both Equations 7 and 8, the first term of the numerator is the boundary coordinate of the starting point area when looking from the starting point to the end point, so the numerator is
is the length of the part included in the starting point area, and by normalizing it by the length Li of the ruled line ki in the denominator, the ratio of the length of the part included in the starting point area to the whole can be found. Starting point weight asi determined by Equations 7 and 8
The starting point area of the pattern vector is updated using Equation 9 (step S61).

[0031]

[Formula 9] P[Msi][Nsi][t][wi]=P[Ms
i][Nsi][t][wi]+asi However, for horizontal ruled lines, t=1; for vertical ruled lines, t=2 Next, the weight aei is determined for the end point area (step S62). Similar to the starting point area weight asi, the ratio of the length of the portion of the ruled line ki included in the end point area to the entire length of the ruled line ki is determined as the weight. Specifically, if the ruled line ki is a horizontal ruled line, it can be found by Equation 10, and if the ruled line ki is a vertical ruled line, it can be found by Equation 11.

[0032]

[Math. 10]

[0033]

[Equation 11] In both Equations 10 and 11, the second term in the numerator is the boundary coordinate of the end point area when looking from the end point to the start point, and therefore the numerator is the length of the portion of the ruled line ki included in the end point area, By normalizing by the length Li of the ruled line ki in the denominator, the ratio of the length of the portion included in the end point area to the entire length is determined. Using the end point weight aei obtained from Equations 10 and 11, the end point area portion of the pattern vector is updated using Equation 12 (step 63).

[0034]

[Formula 12] P[Mei][Nei][t][wi]=P[Me
i] [Nei] [t] [wi] + aei However, for horizontal ruled lines t=1 For vertical ruled lines t=2 Next, the area sandwiched between the start point area and end point area (hereinafter referred to as the intermediate area) Update the weights for . First, it is determined whether the starting point area and the ending point area are adjacent areas (step S64). Whether the starting point area and the ending point area are adjacent areas is determined by whether (Nsi+1) and Nei are equal in the case of horizontal ruled lines, and whether (Msi+1) and Mei are equal in the case of vertical ruled lines. As a result of this determination, if it is determined that the start point area and the end point area are adjacent areas, it is assumed that there is no intermediate area and the weight addition process 2 for the ruled line ki is ended. Furthermore, if it is determined that they are not adjacent, the weight ami is calculated for the intermediate area (step S
65). Starting point area weight asi, ending point area weight aei
Similarly, the ratio of the length of the portion included in one intermediate area of the ruled line ki to the length of the entire ruled line ki is determined as a weight. Specifically, if the ruled line ki is a horizontal ruled line, the formula is 1
If the ruled line ki is a horizontal ruled line in 3, it can be found by Equation 14. Here, the length of the ruled line ki included in one intermediate area is equal to the size of that area, so if you replace it with this,

[0035]

[Math. 13]

[0036]

[Formula 14]. In both Equations 13 and 14, the numerator is the length when crossing one intermediate area, which is equal to the width of the area in the case of a horizontal ruled line, and equal to the height of the area in the case of a vertical ruled line. By normalizing this with the length Li of the ruled line ki, the ratio of the length of the portion included in one intermediate area to the entire length is determined. Using the intermediate area weight ami obtained by Equations 13 and 14, if all intermediate areas of the pattern vector are horizontal ruled lines, use Equation 15 as follows:
In addition, in the case of vertical ruled lines, it is updated according to equation 16 (step S
66). For horizontal ruled lines

[0037]

[Formula 15] P[Msi][n][1][wi]=P[Msi]
[n][1][wi]+ami However, when n is all vertical ruled lines in the interval from (Nsi+1) to (Nei-1)

[0038]

[Formula 16] P[m][Nsi][2][wi]=P[m][N
si][2][wi]+ami However, the pattern vector P is updated in this manner using the ruled line ki throughout the interval from (Msi+1) to (Mei-1). For example, the pattern vector P based on the ruled line k1 shown in FIG. 10 is updated with only one element, P[1][2
][1][w1]=P[1][2][1][w1]+1
becomes. Also, the pattern vector P is updated by ruled line k2 with 4 elements.

[0039]

[Formula 17] P[2][1][1][w2]=P[2][1][
1][w2]+as2 P[2][2][1][w2
]=P[2][2][1][w2]+am2 P[2
][3][1][w2]=P[2][3][1][w2
]+am2 P[2][4][1][w2]=P[2
][4][1][w2]+ae2. By performing the above processing operations (steps S52 to S59) for all horizontal ruled lines and then for all vertical ruled lines, the ruled line information of the input form is converted into a pattern vector, and a pattern vector P is obtained.

Next, based on the pattern vector P of the form to be recognized obtained as described above, the process moves on to identifying the type of the form. To identify the type of form, first, the distance is calculated by comparing the pattern vector P of the form with a predetermined standard pattern vector Sx (where x is the standard pattern number) of the plurality of basic forms to be discriminated. , a candidate is determined (step S6). Distance calculation is performed between pattern vector P and standard pattern S using city block distance. The distance dx is determined by the following formula.

[0041]

##EQU00001## Then, all documents whose distance dx is less than or equal to a preset threshold are determined as document candidates. At this time, the distances are arranged in descending order of distance, and this is used as the priority order for detailed confirmation. In detailed confirmation, each specific area of the form selected as a candidate in the distance judgment is analyzed and verified (
Step S7). At this time, analysis and verification are performed in the order of priority in distance determination, and when a satisfactory result is obtained, the type of input form is determined, and candidates that remain unprocessed are discarded. By doing so, we aim to improve processing efficiency.

[0042] Once the details have been confirmed, the final candidates are determined (step S9), and the type of form has been determined. If detailed confirmation is not possible, it is determined whether or not there is a next candidate (step S10). If there is a next candidate, the process returns to step S7; if there is not, the process is rejected.

[0043] Detailed confirmation will be explained here. Detailed confirmation involves predetermining specific areas with characteristics for each standard form.
An area corresponding to the specific area of the standard form that has been raised as a candidate is cut out from the input form image, and the characters and numbers within that area are read and confirmed, and the grayscale pattern of the cut out area is read to see if they match. This is a process to check whether the For example, in the case of the form shown in FIG. 1, the characters in the specific area 1 for checking details are read and the details are checked to see if it says "transfer slip". When the type of the form is finally determined, the information entry field of the form is cut out according to the writing format of the form, and the contents of that field are read (step S11), and the form type is determined and the reading process is performed. end.

[0044] In the above embodiment, similar form candidates are extracted based on the information on the ruled lines of the input image, and detailed confirmation is performed in a specific area of the standard form selected as the candidate, but this detailed confirmation is omitted. The first candidate may be determined as the first candidate. In particular, detailed confirmation is not necessary for forms that are not provided as specific areas.

[0045]

Effects of the Invention As described above, according to the present invention, a pattern vector is obtained by converting it into a vector pattern based on the information of ruled lines forming a form, and is compared with that of a standard form. It is possible to distinguish the type of form even if it does not have a number or bar code added to it.

Further, candidates are extracted by comparing the pattern vector with that of a standard form, and some of the forms listed as candidates are checked in the order of the candidates, so that the type of form can be reliably determined. Moreover, by doing this, even if the characteristic areas, numbers, and barcode positions differ for each document, it only needs to be done at the final detailed confirmation stage; It is not necessary to check all the specific areas of the form for which the type can be determined, and the speed of the determination process can be increased.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an example of a form targeted by the present invention.

FIG. 2 is a flowchart showing an example of the operation of the present invention.

FIG. 3 is a diagram for explaining smoothing processing.

FIG. 4 is a flowchart illustrating an operation example of ruled line extraction.

FIG. 5 is a diagram for explaining the ruled line extraction operation.

FIG. 6 is a diagram showing a processing result of the form shown in FIG. 1;

FIG. 7 is a diagram for explaining pattern vectors.

FIG. 8 is a flowchart showing an example of operation for obtaining a pattern vector.

FIG. 9 is a diagram showing how an input form is divided into M×N areas.

FIG. 10 is a diagram showing an example of ruled lines on a form.

FIG. 11 is a flowchart showing an example of the operation of weight addition processing.

[Explanation of symbols]

1 Specific area 2 Pixel of interest 3 8 neighboring pixels 4 Input form 5 Starting point area 6 End area

Claims (2)

[Claims] [Claim 1] Image information of documents to be recognized is input, horizontal and vertical line segments are extracted from the input image data, and information on the position, length, and direction of each extracted line segment is detected. Then, the input image is divided into a plurality of areas, and the feature quantity, which is the weight for each direction of the extracted line segment existing in each divided area, is used as the pattern feature quantity of the target to be recognized. and store in advance the feature quantities of a standard pattern that serve as a reference for the pattern to be recognized, and compare and match the extracted feature quantities of the recognized pattern with the stored feature quantities of the standard pattern; A method for determining types of forms, characterized in that the types of forms are determined by determining which standard pattern the recognized pattern is based on the comparison result. [Claim 2] Input image information of documents to be recognized, extract horizontal and vertical line segments from the input image data, and detect information on the position, length, and direction of each extracted line segment. Then, the input image is divided into a plurality of areas, and the feature quantity, which is the weight for each direction of the extracted line segment existing in each divided area, is used as the pattern feature quantity of the target to be recognized. and store in advance the feature quantities of a standard pattern that serve as a reference for the pattern to be recognized, and compare and match the extracted feature quantities of the recognized pattern with the stored feature quantities of the standard pattern; Items that are similar by more than a predetermined value are extracted as candidates in order of similarity, and input image information corresponding to a predetermined specific area of the standard pattern extracted as a candidate is detected to determine whether it matches that of the standard pattern. A method for determining types of forms, characterized in that the types of forms are determined by checking the candidates in the order of the candidates.