JP4057121B2 - Image recognition device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an image recognition apparatus that recognizes an image on a sheet including an image recognition control sheet that controls image recognition.
[0002]
[Prior art]
Conventionally, in the field of image scanners and facsimiles, many proposals have been made on methods for detecting special sheet images having certain characteristics from a plurality of sheet images.
[0003]
Examples of special sheet detection by recognizing marks, symbols and characters printed on a sheet include the following.
In Japanese Patent Application Laid-Open No. 7-210777 (information access device: Fuji Xerox), L-shaped marks are printed at the four corners of a sheet and sheet IDs are printed at specific positions, and a special sheet is detected by detecting this. is doing.
[0004]
Japanese Patent Application Laid-Open No. 7-321977 (facsimile apparatus: Ricoh) discloses an OMR sheet having a start mark, an inclination detection mark, and an end mark (all of which are black rectangles) for the purpose of correcting misalignment.
[0005]
In Japanese Patent Laid-Open No. 7-152858 (optical character reader: Toshiba), partial images (pictures, logos, graphics, etc.) for specifying the type of sheet are printed and recognized.
[0006]
Japanese Laid-Open Patent Publication No. 9-22437 (Barcode Recognition Device: Fujitsu) discloses a method for preventing blurring and blurring at a pixel level in order to improve barcode reading accuracy.
[0007]
Other related technologies include the following.
In Japanese Patent Laid-Open No. 5-314307 (form reading device: Fujitsu), a “hole” is formed at a specific position of a document separation sheet, and this is detected by a sensor.
[0008]
In Japanese Patent Application Laid-Open No. 8-242361 (facsimile apparatus and control method thereof: Canon), the orientation of a document is determined based on the orientation of characters recognized by OCR.
[0009]
Detect special sheets (such as document delimiter sheets) by detecting the characteristics of the sheet itself or identifying image data printed on the sheet (using pattern recognition technology represented by pattern matching and OCR technology). Yes.
[0010]
However, all the conventional technologies that recognize images printed on special sheets and detect that they are special sheets are effective, but they are easily affected by size, resolution, orientation, and noise. In order to do so, complicated processing was required.
[0011]
Further, since the relative position of the mark varies depending on the paper size, position correction processing is required. That is, position correction processing is required as preprocessing for special sheet detection processing.
[0012]
Furthermore, since the number of images forming a mark varies depending on the resolution, a threshold adjustment process is required, and a connected pixel detection and a reduction / enlargement process are required to compare connected pixel shapes.
[0013]
Further, in order to identify and recognize the position of the mark according to the direction, first, a direction determination process is required. That is, the orientation determination process is necessary as a pre-process for the special sheet detection process.
[0014]
Further, since the shape of the mark to be recognized fluctuates due to noise, shape correction processing is required.
Therefore, it does not depend on size and resolution, does not assume a specific orientation, is not easily affected by noise, and is simple processing that is superior in terms of CPU load, memory requirements, processing performance, detection accuracy, etc. Therefore, a special sheet detection method and a special sheet that can be detected by the above are desired.
[0015]
[Problems to be solved by the invention]
As described above, in the method of detecting a special sheet image, the conventional techniques for recognizing an image printed on the special sheet and detecting that it is a special sheet are all effective, but the size, resolution, There is a problem that it is easily affected by the direction and noise, and complicated processing is required to eliminate the influence.
[0016]
Therefore, an object of the present invention is to provide an image recognition apparatus that recognizes an image of an image recognition control sheet that is not easily affected by size, resolution, orientation, and noise and can be detected by simple processing.
[0017]
[Means for Solving the Problems]
The image recognition apparatus according to the present invention is an image recognition apparatus for recognizing an image of an image recognition control sheet, and determines a significant pixel in a pixel column by observing a one-dimensional pixel column parallel to any one boundary line of the image. A first determination unit that determines whether or not a continuous length of significant pixels determined by the determination unit satisfies a first constraint condition; and a first constraint by the first determination unit. A second determination unit that determines whether the number of significant pixel segments that satisfy the first constraint condition satisfies the second constraint condition when the condition is satisfied; When the constraint condition is satisfied, a third determination unit that determines whether the appearance pattern of the significant pixel segment satisfies the third constraint condition, and the third determination unit satisfies the third constraint condition The image is determined to be an image recognition control sheet image. It is composed of a judgment unit.
[0022]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 shows the configuration of an image recognition apparatus 1 according to the present invention. That is, the image recognition apparatus 1 includes a CPU 2 that performs overall control, a memory 3 that stores information, an image memory 4 that stores image data, and a special sheet detection processing unit 5. An image input device 6 for inputting an image such as a color original is connected to the image recognition device 1.
[0023]
FIG. 2 shows the configuration of the special sheet detection processing unit 5. The special sheet detection processing unit 5 includes a black pixel segment searching unit 11, a black pixel segment tracking unit 12, a black pixel segment appearance pattern processing unit 13, a black pixel / pixel column ratio determining unit 14, a black pixel segment number determining unit 15, The pixel segment appearance pattern determination unit 16 and the special sheet detection determination unit 17 are configured.
[0024]
The black pixel segment search unit 11 includes a black pixel detection unit 11a and a black pixel counter 11b.
The black pixel segment tracking unit 12 includes a black pixel detection unit 12a and a black pixel counter 12b.
[0025]
The black pixel segment appearance pattern processing unit 13 includes an appearance pattern recording unit 13a and a black pixel segment counter 13b.
FIG. 3 shows an example of a separator sheet image in which a separator sheet (instruction sheet) as a special sheet according to the image recognition control sheet of the present invention is recognized by the image recognition apparatus 1.
[0026]
This is because a fixed-width lane provided parallel to the four sides of the separator sheet image is equally divided into 16 segments, and five black pixel segments composed of black pixels are divided into segments at both ends of the lane. And each of the third to sixth, seventh to tenth, and eleventh to fourteenth segments is arranged so as to satisfy a certain constraint condition.
[0027]
A fixed-width lane provided in parallel with the sides in the vicinity of the two opposing sides of the separator sheet image is equally divided into 16 segments, and five black pixel segments composed of black pixels are defined as segments at both ends of the lane. The same applies to the ones arranged in the third to sixth, seventh to tenth, and eleventh to fourteenth segments so as to satisfy certain constraint conditions.
[0028]
The separator sheet has “identification patterns” printed on its four sides. In the identification pattern in FIG. 3, the upper side is “10 1000 01000011 01” in the horizontal direction, and a black pixel segment is printed at “1”.
[0029]
This identification pattern is a black pixel segment arrangement pattern indicating a special sheet (separator sheet, instruction sheet), and satisfies the following restrictions (1) to (4).
[0030]
(1) The length of the black pixel segment is 1/16 of the lane length.
(2) The number of black pixel segments is five per lane, and no two are adjacent to each other.
[0031]
(3) The appearance pattern of the black pixel segment is that the lane is equally divided into 16 segments, and the segments at both ends and each of the 3rd to 6th, 7th to 10th, and 11th to 14th segments. .
[0032]
(4) The pattern obtained when the identification pattern is turned upside down or turned left and right (rotated 180 degrees in the paper) is not the main identification pattern.
FIG. 4 shows an example of an identification pattern that satisfies the above-described constraint condition in which a black pixel segment appears. Since the pattern obtained when the identification pattern is reversed upside down or horizontally reversed (rotated 180 degrees in the paper) is not an identification pattern, as shown in FIGS. Is excluded from the identification pattern.
[0033]
As described above, the separator sheet image of the present invention can determine the size and orientation “uniquely” only by recognizing one side of “arbitrary” regardless of the orientation of the separator sheet image. .
[0034]
The two opposing sides of the separator sheet have the same pattern, and the sheet obtained by flipping the separator sheet upside down or horizontally (180 degree rotation) does not have a pattern defined as an identification pattern (mirror image pattern). Is excluded from the identification pattern). That is, in the separator sheet image, the pattern of significant pixel segments appearing in the vicinity of two opposing sides is not point-symmetric with respect to the center of the sheet but is line-symmetric with respect to the center line of the sheet image. Therefore, if the lane of any one side is observed, it can be determined whether the size / orientation is a separator sheet or a separate sheet.
[0035]
FIG. 5 shows an example of a separator sheet on which an identification pattern is printed.
The four images (a), (b), (c), and (d) of FIG. 5 are images (sheets with the arrow direction on the top) read by changing the direction of the same separator sheet, and any of them is arbitrary. The direction at the time of input can be determined by the identification pattern only by observing one side.
[0036]
That is, in FIG. 5A, when reading is performed in a normal direction in which the arrow direction is upward, the upper side identification pattern is “10 1000 0100 0010 01” and the lower side identification pattern is “10 1000 0100 0010 01”. "
[0037]
When this separator sheet is read in the reverse direction (arrow direction is downward), that is, as shown in FIG. 5B, the upper side identification pattern is “10 0100 0010 0001 01” and the lower side identification pattern is “ 10 0100 0010 0001 01 ".
[0038]
When the separator sheet is read as shown in FIG. 5C (the arrow direction is on the left), the upper side identification pattern is “10 1000 0100 00101” and the lower side identification pattern is “10 1000 0100 0010”. 01 ".
[0039]
Further, when this special sheet is read as shown in FIG. 5D (the arrow direction is right), the upper side identification pattern is “10 0100 0010 0001 01” and the lower side identification pattern is “10 0100 0010”. 0001 01 ".
[0040]
As described above, since the identification pattern is different for each side, the direction at the time of input can be determined only by observing one arbitrary side.
Next, the operation of recognizing a sheet by the image recognition apparatus 1 and detecting a separator sheet in such a configuration will be described with reference to the flowchart of FIG.
[0041]
First, the CPU 2 of the image recognition device 1 selects and observes a plurality of sample pixel rows parallel to the lane from the vicinity of the lane of the sheet image data read by the image input device 6 (ST1, 2).
[0042]
The special sheet detection processing unit 5 searches the black pixel segment corresponding to the defined segment length for one sample pixel row selected in an arbitrary lane (ST3) (ST4).
[0043]
The black pixel segment search unit 11 searches for a black pixel run that is not an isolated pixel (ST5). In this case, the black pixel detection unit 11a of the black pixel segment search unit 11 detects black pixels, and the black pixel counter 11b counts black runs corresponding to the defined segment length.
[0044]
If there is no black pixel run in this search, the process proceeds to step ST4. If there is a black pixel run, the black pixel segment tracking unit 12 tracks the black pixel segment (ST6). Here, the black pixel detection unit 12a of the black pixel segment tracking unit 12 detects a black pixel, and the black pixel counter 12b counts the number of black pixels. Note that white pixel noise during black run tracking is ignored within the error range, and black pixel noise during white run tracking is ignored within the error range.
[0045]
Next, it is determined whether or not the segment length constraint condition is satisfied (ST7). In this case, the black pixel / pixel column ratio determination unit 14 invalidates the sample pixel column in which the black pixel number ratio / pixel column from the black pixel segment search unit 11 and the black pixel segment tracking unit 12 does not satisfy the constraint condition in step ST4. Migrate to
[0046]
If not invalidated in step ST7, the appearance pattern recording unit 13a of the black pixel segment appearance pattern processing unit 13 records the appearance pattern of the black pixel segment from the black pixel segment tracking unit 12 (ST8).
[0047]
Here, it is confirmed whether or not the processing has been completed for one sample pixel column (ST9), and if not completed, the process proceeds to step ST4.
When the process is completed for one sample pixel column in step ST9, the black pixel segment appearance pattern determination unit 16 confirms whether the ratio of black pixels in the sample pixel column satisfies the constraint condition (ST10), and the constraint condition is satisfied. If not, the data of this sample pixel column is deleted (ST11).
[0048]
If the constraint condition is satisfied in step ST10, the black pixel segment training determination unit 15 determines whether the black pixel segment training that appears in the black pixel segment training 1 sample pixel row satisfies the constraint condition (ST12). If the constraint condition is not satisfied, the data of the sample pixel column is deleted (ST13).
[0049]
If the constraint condition is satisfied in step ST14 and all the sample pixel columns are completed (ST14), the special sheet detection determination unit 17 compares the appearance pattern of the black pixel segment with the identification pattern list (ST15), It is determined whether or not the majority of the sample pixel columns match (ST16). If they match, it is determined that the sheet is a special sheet (ST17), and if they do not match, it is determined that the sheet is not a special sheet (ST18).
[0050]
FIG. 7 shows an example of a sample pixel column.
Depending on the resolution, this sample pixel column has a “ratio to resolution” such as one per 4 scanning lines at 200 dpi, one per 8 scanning lines at 400 dpi, and one per 12 scanning lines at 600 dpi. Extract.
[0051]
Even if a character or the like exists in a state where it overlaps a part of an arbitrary lane, only the black pixel segment satisfying the constraint condition as the “identification pattern” is searched, so that white pixel noise / white run during black run tracking is searched. The black pixel noise during tracking can be ignored as not being a significant pixel (black pixels during black run tracking / white pixels during white run tracking).
[0052]
In addition, constraints such as the ratio of black pixels to the number of pixels in an arbitrary lane and the ratio of white pixel noise during black run tracking to the segment length can reduce the occurrence of black pixel segment detection errors.
[0053]
Furthermore, after signal level noise removal, symbol level noise removal is realized. By setting a constraint condition on the appearance pattern, it is possible to specify the size / orientation of the sheet at the same time as reducing the sheet detection error.
[0054]
In order to improve the detection accuracy, it is effective to increase the number of sample pixel columns and observe two or more lanes.
In addition, increasing the number of segments set in the lane can define more identification patterns, but the noise removal capability decreases. Therefore, “200 dpi / A5 vertical” is used as the worst condition when using a separator sheet. It is reasonable to set parameters for.
[0055]
In the above embodiment, black pixels are used. However, the present invention can be applied as significant pixels using white pixels or other pixels.
As described above, according to the embodiment of the present invention, a special sheet can be detected by simple processing (only observation of a plurality of one-dimensional pixel columns) regardless of size and resolution.
[0056]
In addition, the special sheet can be detected by a simple process (only observation of a plurality of one-dimensional pixel rows) which is not easily affected by noise.
In addition, the orientation of the special sheet can be determined by simple processing (only observation of a plurality of one-dimensional pixel columns) regardless of the orientation at the time of input.
Also, the type of special sheet can be identified by simple processing (only observation of a plurality of one-dimensional pixel columns).
[0057]
【The invention's effect】
As described above in detail, according to the present invention, it is possible to provide an image recognition apparatus that recognizes an image of an image recognition control sheet that is not easily affected by size, resolution, orientation, and noise and can be detected by simple processing. .
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of an image recognition apparatus according to the present invention.
FIG. 2 shows a configuration of a special sheet detection processing unit.
FIG. 3 is a diagram showing an example of a separator sheet image according to the image recognition control sheet of the present invention.
FIG. 4 is a diagram showing an example of an identification pattern that satisfies a constraint condition.
FIG. 5 is a diagram illustrating an example of a separator sheet on which an identification pattern is printed.
FIG. 6 is a flowchart for explaining an operation of detecting a separator sheet.
FIG. 7 is a diagram illustrating an example of a sample pixel column.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Image recognition apparatus 2 ... CPU
DESCRIPTION OF SYMBOLS 3 ... Memory 4 ... Image memory 5 ... Special sheet detection process part 6 ... Image input device 11 ... Black pixel segment search part 12 ... Black pixel segment tracking part 13 ... Black pixel segment appearance pattern process part 14 ... Black pixel / pixel column ratio Determination unit 15 ... black pixel segment training determination unit 16 ... black pixel segment appearance pattern determination unit 17 ... special sheet detection determination unit

Claims (1)

In the image recognition apparatus for recognizing the image of the image recognition control sheet,
Of the four sides of the image of the image recognition control sheet, two opposite sides are provided symmetrically with respect to the center line of the image of the image recognition control sheet and asymmetric with respect to the center of the image of the image recognition control sheet. Storage means for storing a different one-dimensional pixel array arrangement pattern for each size of the image recognition control sheet;
A determining means for determining the significant pixels in the pixel rows, observing the continuous length of the significant pixels by observing any of the one-dimensional pixel row of the read image recognition control sheet image,
First determination means for determining whether or not a continuous length of significant pixels determined by the determination means satisfies a first constraint condition;
Second determination means for determining whether or not the number of significant pixel segments satisfying the first constraint condition satisfies the second constraint condition when the first determination means satisfies the first constraint condition; ,
Third determination means for determining whether or not the appearance pattern of the significant pixel segment satisfies the third constraint condition when the second determination means satisfies the second restriction condition;
When the third constraint is satisfied by the third determination unit, the image is determined to be an image recognition control sheet image, and the significant pattern observed by the array pattern stored in the storage unit and the determination unit A determination means for determining the size and orientation of the read image recognition control sheet by comparing the continuous length of the significant pixels ,
An image recognition apparatus comprising:

Images