JP4629718B2 - Boundary detection method, program, and image processing apparatus - Google Patents

Description

  The present invention relates to an image processing technique, and more particularly to an image processing technique for detecting a position where a feature of a region on an image changes.

  As an apparatus that can read a document (for example, a paper medium), an image reading apparatus that performs optical reading, such as a scanner or a copier, is known. In order to read a document image in such an image reading apparatus, the surface of the document (read target surface) is generally set toward the reading surface of the image reading apparatus, and a document cover covering the document is covered from the back of the document. In the state, a reading is performed.

  When reading an image, light is emitted from the reading surface toward the set document surface, and the reflected light is read by a light reading unit (CCD) configured on the reading surface. The image data of the document is obtained by converting the information on the surface of the document read in this way into an electrical signal.

  When the size of the document is smaller than the reading surface, the back side of the document cover around the document viewed from the reading surface side (the reading surface side of the document cover is the background of the document. When the image is read, this background plate is also read in addition to the surface of the document. Naturally, the acquired image data includes a background plate around the document. It will be a thing.

  The image data acquired in this way is used, for example, to be printed by a printing unit and output to paper, or sent to the image editing unit to cut out a part of the image data or rotate the image data. Correction is performed, and further, it is used for OCR processing of a character area in image data, or is stored in a database as it is, and is used in various fields. However, in such utilization, particularly in the above-described utilization, the reflection of the background plate other than the original in the image data causes various problems.

Although techniques for solving such problems have been disclosed in various fields, none of them is a technique that can be used for general purposes.
Japanese Patent Laid-Open No. 5-110779 JP 7-115514 A

One of the disclosed techniques is a technique developed for subjecting acquired image data to OCR processing or storing it in a database as an electronic file.
In general, as a document to be subjected to OCR processing, a standard document having a fixed printing position such as a form is used. When performing OCR processing of a standardized document image that has been converted to data, the standard document image is appropriately rotated and corrected. If the print position is set in advance from the reference position of the standard document image (for example, the position of the upper left end of the standard document image), the print position is specified from the target standard document image. Character recognition is performed by OCR processing for the specified position, and desired character information can be extracted.

  The edges of a document such as a form generally have white margins. When image data for a document smaller than the reading surface is acquired using a background plate of the same color or a color similar to that, the acquired image data is the document. No white boundary image data is generated in the vicinity of the boundary between the edge and the background plate.

In order to perform the OCR process on the image data, it is necessary to specify in what state (angle or arrangement) the document is reflected in the image data.
However, the edge of the document cannot be detected because it is the same color as the background plate.

  Therefore, as a technique for solving this problem, the white color of the background plate is set to black so that the edge of the image data serving as an index of the state of the original in the image data can be detected. By specifying the angle and arrangement of the original, the original is read.

  However, since the image reading apparatus that must use such a method is used for a special purpose that is not general-purpose, it becomes very expensive and the background plate is black. For this reason, in a thin white original, the black background plate is transparent, and it is difficult to distinguish, for example, characters in the read image data.

As another disclosed technique, there is a technique for outputting paper with the same size as a document applied to a copier or the like.
Unlike the case where the background plate is changed to black as described above, the background plate is configured to have a white color that does not affect the copy quality by arbitrarily rearranging the RGB arrangement of the background plate. Is.

  Naturally, since the recombination information of the RGB arrangement is recognized, it is possible to distinguish from the different arrangement by determining each pixel, and the size of the original can be specified.

However, even in this case, since it can be realized only by a dedicated image reading apparatus having the RGB array information of the background plate in advance, the technique is not suitable for general purpose.
In any of the techniques, the edge of the document is specified from the read image by crafting at the time of reading the image, and there is nothing that can detect the edge of the document only by image processing after the image is read.

  Accordingly, in order to solve the above problems, the present method, program, and image processing apparatus detect a document edge from image data read in an arbitrary combination of a document and a background plate, a boundary detection method, a program, and An object is to provide an image processing apparatus.

  In the disclosed boundary detection method, the frequency component of the image data is extracted for each region having a plurality of pixels in the horizontal and vertical directions of the image data read by the image reading means, and the amount of change in the frequency component of each extracted region is calculated. Find the peak of the change amount of the frequency component, apply linear approximation to the detected peak, provisionally determine a plurality of pixels on the approximate straight line obtained by linear approximation as a boundary, Extract the frequency components of the pixel column including the tentatively determined pixel and a predetermined number of pixel columns before and after the pixel column, determine the amount of change in the frequency components of the plurality of extracted pixel columns, and determine the peak of the frequency component change amount Are detected, linear approximation is performed on the detected peak, and a plurality of pixels on the approximate straight line obtained by the linear approximation are determined as boundaries.

  One of the aspects of this program is that each unit region on the image data is premised on having the computer execute detection of boundaries between regions having different characteristics in any adjacent region on the image data. A function for extracting feature information determined by all the pixels in the unit region, a function for obtaining a difference in the feature information from the adjacent unit region with respect to any unit region, and the difference is at least predetermined. And a function of determining the reference unit region that is equal to or higher than the level as the boundary.

  Another aspect of the present program is based on the premise that the computer executes detection of boundaries between regions having different pixel arrangements in arbitrary adjacent regions on the image data. A function for extracting image frequency information for each unit area, a function for determining representative feature information based on the extracted image frequency information for each unit area, and the adjacent adjacent ones based on the arbitrary unit area The function of obtaining the difference of the representative feature information from the unit area and the function of determining the reference unit area as the boundary at which the difference is at least a predetermined level are realized.

  One aspect of the present image processing apparatus is based on the premise that a boundary between regions having different characteristics is detected in any adjacent region on the image data, and for each unit region on the image data, the unit A feature enhancement unit that extracts feature information determined by all the pixels in the region and a difference between the feature information between the adjacent unit regions with respect to an arbitrary unit region, and the difference is at least a predetermined level or more Boundary determining means for determining the reference unit region as the boundary.

  Another aspect of the present image processing apparatus is based on the premise that a boundary between regions having different pixel arrangements is detected in any adjacent region on the image data, and for each unit region of the image data. Characteristic enhancement means for extracting image frequency information in the unit, and representative feature information based on the extracted image frequency information is determined for each unit region, and the adjacent unit region is determined based on any unit region. Boundary determining means for determining a difference between representative feature information and determining the reference unit region as the boundary, at which the difference is at least a predetermined level or more.

  Further, it may be configured to further include image reading means for optically reading a document and generating the image data of the document.

  In the present method, program, and image processing apparatus, feature information such as an image frequency determined as a whole of the unit pixel is extracted for each unit region on the image data, so that two different surface states are temporarily included in the image data. If paper is reflected, even if the two papers are the same color, the difference in the state can be extracted as different feature information.

  Then, if the amount of change between adjacent regions is obtained using the feature information extracted in this way, the amount of change increases at the boundary between the two papers having different surface states. Can be specified as the boundary between two papers.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
An image processing apparatus according to an embodiment of the present invention includes, for example, a scanner, a copier, and the like. A document having a background image (hereinafter referred to as a background plate image) that is read by an image reading device that reads the document image with the back member facing the back of the document cover) For an image (hereinafter referred to as a framed document image), the boundary between the document image of the framed document image and the background plate image is detected.

The document read by the image reading apparatus has a white margin area around it, and the background plate has the same white color as the margin area.
For the sake of convenience, it is assumed that the original is rectangular paper, and the print area where characters and ruled lines are printed (for example, a form) occupies the area other than the margin area of the original.

FIG. 1 is a block diagram of an image processing apparatus according to an embodiment of the present invention.
The image processing apparatus shown in FIG. 1 includes a print area exclusion unit 100, a feature enhancement unit 102, an edge provisional determination unit 104, and an edge determination unit 106. The image processing apparatus receives a framed document image A as input information and outputs edge information as output information. Information B is output.

  The framed document image A shown in FIG. 1 is obtained by reading the surface of the document in units of pixels by the above-described known image reading device (not shown), and generating R (Red) and G (Green) for each pixel generated by electrical conversion thereof. ) And B (Blue) density information, or image information based on management by a lookup table.

Here, before describing each part, the characteristics of the framed document image will be described with reference to FIG.
2 shows an image plane of a display device (for example, a monitor or the like) not shown in FIG. 1 generated as described above, with the horizontal direction as the X axis and the vertical direction as the Y axis. It is an example when displayed on the (−Y coordinate plane).

  The framed document image A in the figure is one side of the document for easy understanding of the following description (actually not shown in the framed document image, but the left edge of the document image indicated by the broken line in the figure for explanation) ), The right side is the original image on the right side of the broken line, and the left side is the background plate image read from a part of the background plate for the outside of the original.

  Here, an area where characters, ruled lines, symbols, and the like are shown is an original print area 200, a margin portion of the original image surrounding the print area 200 is a margin area 202, and a background board image area is a background board area 204. I will call it.

As shown in the figure, in the portion 206 where the shadow of the edge is reflected in the image, the difference in the gradation value can be clearly detected, so that the position of the edge can be specified.
However, in the portion where the shadow of the edge that cannot be seen from the framed document image is not reflected, the background plate color and the margin color are the same color, as is apparent from the figure, and the border between these colors is adjacent. Since a series of data that does not cause a color difference between the matched pixels is formed, the position of the edge cannot be specified.

  However, the state of the background plate (for example, the surface is rough, the surface is smooth, or the material is distinguished from other materials depending on the type of material) and the margin of the document (for example, the surface is rough) It is generally different from the state, smooth surface, or material state that differentiates from other materials depending on the type of material), so focus on extracting the difference in the state as several types of features A method for specifying the edge region is conceivable.

  The enlarged view 208 shown in the figure is for a rectangular area 210 set to include the boundary between the background board area 204 and the margin area 202 of the framed document image A and the vicinity of the rectangular area. A partially enlarged image is shown when saturation enhancement is performed so as to show the difference in the state.

  By applying saturation enhancement to the framed document image A, the difference in RGB arrangement pattern based on the above-described difference between the blank area 202 and the background board area 204 is emphasized, as shown in FIG. Visually, it becomes possible to discriminate between the boundaries due to the difference of the respective areas.

  Focusing on this feature, each part of the image processing apparatus that detects the edge of the document that could not be discriminated in the framed document image and can distinguish between the document region and the background plate region on the image will be described with reference to FIG. This will be described below with reference to the framed document image A.

  1 scans the gradation value of each pixel of the framed document image A to identify the printing area 200 of FIG. 2 (for example, pixels including 200 or less RGB gradation values). Judged as the print area and removed from the edge detection scanning target).

  In the feature emphasizing unit 102 in FIG. 1, different feature amounts (for example, image frequency (or spectrum) or frequency included in the pixel region) differ from the framed document image A for each designated range (or block). Distribution) is extracted in a predetermined range unit (for example, two-dimensional fast Fourier transform is executed in units of 32 pixels vertical by 4 pixels horizontal to obtain the spectrum of each region, and the high frequency component, low frequency component, and DC component of the obtained spectrum are obtained. And the average value of the frequency distribution of the spectrum is used as the feature amount of the corresponding region).

  The predetermined range is, for example, a range in which the vertical is 1 pixel (pixel) and the horizontal is m (pixel) (where l and m are integers), and the framed document image on the image plane is The unit area of each pixel designated in units of the predetermined range is hereinafter referred to as a pixel area.

In this feature emphasizing unit, it is possible to exclude the print region 200 shown in FIG. 2 specified by the print region excluding unit 100 and extract the feature amount of each pixel region.
The provisional edge determination unit 104 in FIG. 1 provisionally determines a position corresponding to the edge of the document on the framed document image based on the feature amount extracted for each pixel region by the feature enhancement unit 102 (for example, the pixel region The value obtained by adding each feature amount with a certain weight is used as the difference value, the peak at the extreme end of the difference value is used as the edge, and a straight line approximation is performed based on the obtained edge. Tentatively determined as an edge).

  In the edge determination unit 106 in FIG. 1, the feature enhancement unit 102 extracts a feature amount from the position of the edge temporarily determined by the edge temporary determination unit 104 and the vicinity of the edge position on the image plane. Various feature amounts are extracted in units of pixel regions that are narrower than the pixel region, and the edge position of the document on the framed document image is finally determined based on the feature amount (for example, the temporarily determined edge A one-dimensional fast Fourier transform is executed in the unit of 32 pixels vertical by 4 pixels horizontal for the position of and the area in the vicinity thereof, and a wavelet transform is further executed on the feature quantity extracted based on this execution, and the value obtained as a result The position on the approximate line is determined as the edge position by performing a linear approximation on the peak position based on the

Then, edge information B as shown in FIG.
Next, an example of the operation flow of each of the above units of the image processing apparatus according to the embodiment of the present invention will be described below.

  In the image processing scanning method shown in this example, unless otherwise specified, each pixel on the framed document image A shown in FIG. 2 is the same in the horizontal direction in units of 32 pixels vertically from the upper left end of the figure. The pixel row is scanned toward the right end of the figure, and when the processing of the pixel row is completed, the same processing as described above is performed on the same pixel row in the horizontal direction in the lower stage in units of 32 pixels vertically. In the following description, it is assumed that the processing is repeated in units of horizontal lines having a width of 32 pixels in the vertical direction until the lower end of the document image A with a frame is reached.

Further, in order to help understanding of this operation flow, description will be made with reference to actual measurement drawings as appropriate.
FIG. 3 is an example of an operation flow performed in the print area exclusion unit 100 of FIG.
In the operation flow of FIG. 3, first, the position of the start line to be read first on the image plane of the framed document image A of FIG. 2 is set to the position of the upper end of the image A shown in FIG. ).

Then, the pixel located at the left end in FIG. 2 of the set line is set as a reading target (S302).
Here, the RGB gradation values of the set pixels are read, and it is determined whether or not there is a pixel that does not exceed the RGB gradation value 200 that is determined in advance in this example and serves as a reference to be excluded as a print area (S304). .

The RGB gradation value used as the reference for this determination may be appropriately set according to the original.
In step S304, if the RGB gradation value of the read pixel exceeds 200, it is determined that the pixel is not a print area, and the pixel located on the right side of the same line is set as the next read target. (S306).

  In step S304, when the RGB gradation value of the read pixel is 200 or less, it is provisionally set that there is a print area in this area, and the process proceeds to the subsequent noise determination process in step S308.

  In step S308, a determination is made as to whether or not there is a pixel that is temporarily set as a print area by continuously positioning the position on the image plane for the pixel that is temporarily set as having a print area in step S304.

  If it is determined in step S308 that there is no continuous pixel, the process proceeds to step S306, and a pixel located on the right side of the current processing target pixel on the same line is set as a reading target. Then, the process is executed in the same steps as described above.

  This is because a pixel that is not temporarily set as a print area for two consecutive pixels is likely to be noise due to dust or the like, which is not related to the print area. Can be set as appropriate.

If it is determined in step S308 that there is a continuous pixel, the pixel is set as the print area 200 detected first from the left end of the image A in FIG. 2 (S310).
Subsequently, in step S312, it is determined whether or not there is a pixel that has not been processed in this line.

  If it is determined in step S312 that pixels to be processed in this line remain, the process proceeds to step S306, and the remaining pixels are similarly processed in the above-described steps.

If it is determined in step S312 that there is no pixel to be processed in the main line, it is determined whether the main line is the last line moved to the lower end of the image A in FIG. 2 (S314).
If it is determined in step S314 that the main line is not the final line, the line is set at a position immediately below the main line on the image plane (S316). From the left end of the set line, step S302 and subsequent steps are performed. Repeat the process shown.

  When all the scans in the image A in FIG. 2 are completed and the scan up to the final line is completed, it is determined in step S314 that the main line is the final line, and the process ends.

FIG. 4 is a diagram in which the print area specified by the operation flow is compared with the original image.
2A is a view showing the original image A with a frame in FIG. 2 as an original image, and FIG. 2B is a view showing the print area specified by the operation flow in black. is there.

As shown in FIG. 5B, it can be seen that the print area 200 and the portion 206 where the shadow of the edge is reflected are specified as the print area 400 in FIG.
As described above, the framed document image A input to the print area exclusion unit 100 in FIG. 1 is the print area 200 of the framed document image A that is displayed on the image plane in FIG. A portion 206 where the shadow of the edge is reflected is also specified as the print area.

In the subsequent processing, it is possible to exclude the print area 400 of FIG. 4 specified in this way.
FIG. 5 is an example of an operation flow performed in the feature emphasizing unit 102 of FIG.

This operation flow is performed on the image A in which the print area is specified by the print area excluding unit 100 in FIG.
In this operation flow, as shown in FIG. 6, a framed document image A shown on the image plane is shown in a range 600 shown in FIG. 6 (in this example, a block of 32 pixels vertical by 4 pixels horizontal). Each pixel region obtained by dividing the region into units is used as a unit of processing unless otherwise specified.

  In the operation flow of FIG. 5, first, the pixel area 602 designated by the above range 600 is set as the start block to be processed at the upper left end of the framed document image A of FIG. 6 (S500).

  Then, the pixel specified by the print area excluding unit 100 in FIG. 1 as to whether or not the pixel area (the pixel area 602 in this stage and the same applies below) includes the print area 400 in FIG. The determination is made based on the information indicating the print area 400 of FIG. 4 (S502).

  If it is determined in step S502 that the print area 400 of FIG. 4 is included in the pixel area, a pixel area located in the scanning direction of the pixel area in the image A of FIG. 6 is set as a process target area. Correct (S504).

  If it is determined in step S502 that the print area 400 in FIG. 4 is not included in the pixel area, a known two-dimensional fast Fourier transform process (hereinafter abbreviated as 2DFFT) is performed for the pixel area in units of pixel areas. To obtain the spectrum of the pixel region (S506).

  Here, an average value of high-frequency components (in this example, 1 / 2π ≦ ω <3 / 4π, where ω is a variable indicating frequency) of the spectrum obtained in the pixel region is obtained (S508). .

Subsequently, an average value of low-frequency components (0 <ω <1 / 2π in this example) of the spectrum obtained in the pixel region is obtained (S510).
Subsequently, an average value of DC components (in this example, ω = 0) of the spectrum obtained in the pixel region is obtained (S512).

  FIG. 7 shows actual measurement data of the average values of the various components obtained for each pixel area in units of 32 pixels × 4 pixels in the rectangular area 210 (32 pixels × 144 pixels) shown in FIG. In the actual measurement data, the calculation result is shown for each RGB, and the variation of each value for each RGB is easy to understand.

2 corresponds to the X-axis direction of the image A in FIG. 2, and indicates the value of the number of pixels when the number of pixels is counted in the right direction with the left end of the rectangular area 210 in FIG. 2 as the origin of the X coordinate. Has been.
7 corresponds to the position designated by the X coordinate, which is the position of the number of pixels from the left end of the rectangular area 210 in FIG. 2, and direct current components (DC component R, DC component G, And a DC component B), a gradation value, a low frequency component for each RGB (low frequency component R, a low frequency component G, and a low frequency component B), a spectral value, and a high frequency component for each RGB (high frequency component R, high frequency component) The spectral values are shown as G and the high frequency component B).

When the values of these components are indicated by a line graph, graphs as shown in FIGS. 8 to 10 are obtained.
FIG. 8 is a graph with respect to the high-frequency component, showing the change in spectrum for each RGB with the horizontal axis representing the number of pixels (the number of pixels at the left end of each pixel region) and the vertical axis representing the spectral value.

  FIG. 9 is a graph for low-frequency components. Like the graph for high-frequency components, the horizontal axis indicates the number of pixels (the number of pixels at the left end of each pixel region), and the vertical axis indicates the spectrum value. It was.

FIG. 10 is a graph with respect to the direct current component. The horizontal axis indicates the number of pixels (the number of pixels at the left end of each pixel region), and the vertical axis indicates the gradation value.
As expected from the change in the broken line in FIGS. 8 to 10, the most rapid vertical change is observed at the position where the X coordinate is 136, and also in FIGS. 7 to 10, the solid line at the position where the X coordinate is 136. As shown, the presence of an edge of the document is expected for this position.

Now, it returns to description of the operation | movement flow of FIG.
In the above-described step, the average value of the three types of components is obtained. However, here, the half-value width is obtained from the spectrum of the pixel region, and the obtained half-value width is set as the frequency distribution of the pixel region (S514).

  The half-value width is an interval between two periods in the vicinity of the peak period, indicating half the peak value in the frequency distribution diagram obtained when the horizontal axis is the period and the vertical axis is the spectrum intensity. .

  Then, the average value for each component obtained in steps S508 to S512 and the frequency distribution set in step S514 are set as the feature amount of the pixel area (S516).

Here, it is determined whether or not the above-described processing has been executed for all the pixel areas of the framed document image A shown in FIG. 6 (S518).
If it is determined in step S518 that there is a pixel area to be scanned next to the pixel area, the process proceeds to step S504, and the pixel area to be scanned next in the image A in FIG. Re-set and execute the process in the steps described above.

  In step S518, the above-described processing is executed for all the pixel regions of the image A, and when it is determined that there is no pixel region to be scanned next to the pixel region, the processing ends.

In this operation flow, four feature amounts are obtained. However, the present invention is not limited to this, and other feature amounts may be added.
As described above, the feature enhancement unit 102 in FIG. 1 extracts various feature amounts in units of pixel regions of a predetermined size from the framed document image A processed by the print region exclusion unit 100. Can do.

FIG. 11 is an example of an operation flow performed in the temporary edge determination unit 104 in FIG.
This operation flow is executed based on various feature amounts extracted by the feature enhancement unit 102 of FIG.

  In this operation flow, first, the processing target range of the framed document image is determined on the basis of the pixel region specified by the range 600 in FIG. Then, the corresponding processing range is set (S1100).

The determination of the processing range in this example is performed by determining that the range from the left end to the printing region 400 of the image A shown by blackening the printing region in FIG.
When the processing range is determined in this way, the upper end line of the framed document image A shown in FIG. 6 is set as a start line for reading the pixel area (S1102).

Here, the weight determination processing of the various feature amounts, which will be described in detail later, is executed (S1104).
Then, the pixel area located at the left end of the set line is set as the pixel area to be read first in this line (S1106).

  Subsequently, with respect to the pixel area adjacent to the pixel area (in this example, up to two pixel areas connected to the left adjacent to the corresponding pixel area), FIG. The average for each feature amount obtained in the operation flow (from step S508 to step S514) is obtained (S1108).

FIG. 12 illustrates the arrangement relationship of the pixel areas to be averaged in the pixel areas to be processed.
This figure shows a pixel area 604 designated as a processing target pixel area on the framed document image A in FIG. 6 as a processing target. In this case, a pixel adjacent to the left of the pixel area 604 is shown. A total of two pixel areas, the area 1202 and the pixel area 1204 located further to the left, are designated as pixel areas for calculating the average for each feature amount of the adjacent pixel areas with respect to the pixel area 604 designated as the processing target. Is done.

And the average for every feature-value calculated | required in this way is used for the continuing process (step S1110) of FIG.
In step S1110 of FIG. 11, the amount of change for each feature amount is obtained between the various feature amounts of the pixel area to be processed and the average for each feature amount obtained in step S1108.

  The variation amounts of the various feature values obtained in this way are the weights of the various feature values obtained by the weight determination process in step S1104, or the weights statistically obtained in advance (desirably, the high-frequency component is set to 1, The frequency component is 2, the DC component is 1, and the frequency distribution is 1, and the value is set as the feature change amount in the pixel region (S1112).

Here, a process for obtaining the peak of the feature change amount set for each pixel region in step S1112 is performed (S1114).
The processing for obtaining the peak shown here (peak detection processing) will be described in detail later.

Then, the peak determination of the feature change amount of the pixel region based on the peak detection process is performed (S1116).
If it is determined in step S1116 that there is no peak, it is then determined whether the pixel area to be scanned next (pixel area 1206 in FIG. 12) is a print area (S1118). .

  If it is determined in step S1118 that the pixel area to be scanned next is not a print area, the pixel area (the pixel area 1206 in FIG. 12) is only one pixel area from the central pixel area. It is determined whether the pixel region is on the scanning direction side (S1120).

  If it is determined in step S1120 that the pixel area is a pixel area on the scanning direction side from the central pixel area, whether or not the pixel area to be processed is located on the last line. Is determined (S1122).

Returning to the process of step S1116, if each step shown from step S1116 receives a determination opposite to the above, the process is executed as follows.
If it is determined in step S1116 that there is a peak, it is determined that this pixel area to be processed is a pixel area corresponding to an edge, and this pixel area is temporarily determined as the left edge of the original image in this example. (S1124). And it transfers to the process of step S1122 mentioned above.

If it is determined in step S1118 that the pixel area to be processed is a print area, the process proceeds to step S1122.
If it is determined in step S1120 that the pixel region (pixel region 1206 in FIG. 12) is not a pixel region on the scanning direction side from the central pixel region by one pixel region, the process proceeds to step S1126 and the next scanning is performed. A pixel region to be processed is set as a pixel region to be processed, and the processing is repeatedly executed in the order of steps described above from step S1108.

  If it is determined in step S1122 that the pixel area to be processed is not located on the last line, the process proceeds to step S1126, and the processes are executed in the order of steps described above.

If it is determined in step S1122 that the pixel area to be processed is located on the last line, the process proceeds to the subsequent step S1128.
In step S1128, linear approximation on the image plane is performed on all the pixel areas corresponding to the edges detected in step S1124, and the pixels on the obtained approximate straight line are converted into edge portions at the left end of the document image. Are temporarily determined, and this processing is terminated.

  Note that the above operation flow (FIG. 11) may be executed only for the color having the largest characteristic in the spectral values of each RGB color, or for two or more colors arbitrarily combined in RGB. It may be executed.

FIG. 13 is a diagram showing the arrangement of edges provisionally determined by the above processing in light of the framed document image A in FIG.
A straight line 1300 shown with respect to the image A in the figure is the approximate straight line obtained in step S1128, and a pixel corresponding to the approximate straight line 1300 is provisionally determined as the left edge of the document image.

In this figure, the edge of the original image can be set at a position close to the actual boundary between the marginal area 202 and the background plate area 204 indicated by the broken line in FIG.
In this way, the edge provisional determination unit 104 in FIG. 1 can narrow down the edge positions so that the edges can be provisionally determined closer to the actual edge positions.

Here, the weight determination process (step S1104) and the peak detection process (step S1114) included in FIG. 11 will be described in order.
FIG. 14 shows an operation flow of the weight determination process.

  In the weight determination processing, first, each component of the spectrum (high frequency component, low frequency component, and direct current) for each pixel region obtained in step S506 of each step of FIG. 5 showing the operation flow in the feature enhancement unit 102 of FIG. Component) and frequency distribution are normalized by, for example, a simple average with pixel regions located on the left and right sides (S1400).

By this normalization, errors due to minute fluctuations in each component of the spectrum and the frequency distribution obtained in step S506 can be absorbed.
Subsequently, in the processing range determined in step S1100 in FIG. 11 (from the left end of the framed document image to the position where the print area starts), the average value of each component and frequency distribution is obtained, and the obtained average value is used as a reference. Then, a dispersion amount of each component and frequency distribution in the predetermined range is calculated (S1402).

Then, it is determined for each component and frequency distribution whether the variance calculated in this way fluctuates more than a preset threshold value (S1404).
The threshold value is a value set to 20% of the fluctuation average value, for example.

  If it is determined in step S1404 that the amount of dispersion has fluctuated by a predetermined threshold or more, the maximum amount of dispersion is obtained for the component and frequency distribution, and the value is the amount of change in the component and frequency distribution. (S1406).

If it is determined in step S1404 that the amount of dispersion does not fluctuate more than a preset threshold, it is set as no amount of change (S1408).
Finally, the reciprocal of the set ratio of each component and frequency distribution is calculated, the value is determined as the weight of each component and frequency distribution (S1410), and the process ends.

FIG. 15 shows an operation flow of peak detection processing.
In the peak detection process, first, a pixel area to be processed is designated (S1500).
Then, level average values are obtained for all pixel areas on the same line located in the direction opposite to the scanning direction from the designated pixel area (S1502).
FIG. 16 is a level distribution diagram for one line in which the horizontal axis represents the horizontal axis (X axis) of the framed document image A in FIG. 12 and the vertical axis represents the above level.

All the white circles 1600 plotted in the figure indicate representative values in a predetermined pixel area unit.
If the white circle 1602 that protrudes in the figure is the level of the pixel area that is the target of the peak detection process, the levels 1604 of all the pixel areas that are located to the left of the white area 1602 are the objects to be averaged.

Note that the level indicated by the level average value refers to a feature change amount in the edge provisional determination process of FIG.
This peak detection process is also used in the operation flow of the edge determination unit 106 in FIG. 1 to be described below, and the feature change amount is not used in the process of step S1502 at that time. Therefore, in the description of this peak detection process flow Is a level average value for convenience.

When the level average value is obtained in step S1502, the peak determination of the pixel area to be processed is performed (S1504).
This is because a constant α calculated statistically in advance with respect to the above average level value (in the horizontal scanning as shown in this example, α = 2.5 is most desirable, and vertical 4 pixel × When various processes are performed in a mode in which the horizontal 32 pixels is a unit of pixel area and these pixel areas are scanned in the vertical direction, α = 2.0 is most preferable. It is determined whether or not the level of the pixel area (the feature change amount in the provisional edge determination process of FIG. 11) is a vertex exceeding the level average value multiplied by α.

  If it is determined in step S1504 that the target level is not the vertex, the pixel area to be scanned next is reset (S1506), and the process is repeated from step S1502 in the order of the steps.

If it is determined that the target level is the vertex, the peak detection process is terminated.
Next, processing of the edge determination unit 106 in FIG. 1 for narrowing down the exact position of the edge based on the edges temporarily determined by the edge temporary determination unit 104 in FIG. 1 will be described.

  For ease of explanation, it is assumed that the edge positions provisionally determined by the edge provisional determination unit 104 in FIG. 1 are shown in pixels arranged in the same column in units of 32 pixels vertically.

FIG. 17 is an operation flow performed in the edge determination unit 106 of FIG.
In this operation flow, first, pixel rows (32 pixels long × 1 pixel wide) temporarily set as edge positions on the upper end line of the framed document image A shown in FIG. A range of 16 pixels is set as a processing target (S1700).

  Then, the unit of the pixel area in this process is a pixel column having a range of 32 pixels vertically and 1 pixel horizontally, and the one-dimensional fast Fourier transform of the above range is performed in this pixel column unit to obtain the spectrum of each pixel column (S1702). .

FIG. 18 shows the range to be processed with respect to the position line.
FIG. 18A shows one line of the processing target range including the pixel column of the approximate straight line 1300 in FIG.

A rectangular area 1800 in the figure is a processing target range on a line 1802 indicated by a broken line, and although not particularly illustrated, the processing target range is similarly set for other lines.
FIG. 7B shows the positional relationship between the position that is temporarily determined as the position of the edge by enlarging the rectangular area 1800 and the range to be processed.

  From this figure, each pixel row 1806 (vertical 32 pixels × horizontal 1 pixel) arranged 16 pixels forward and backward with respect to the scanning direction indicated by the arrow of the temporary edge 1804 indicated by a line in the center in the figure is the processing target. You can see that it is in range.

  When the spectrum is obtained in step S1702, the spectrum components (high frequency component, low frequency component, and direct current component) and the average value of the frequency distribution are obtained in units of the pixel column 1806 (S1704).

  Then, the component and the frequency distribution are added for each pixel column 1806 with the weight calculated in step S1104 in FIG. 11 or statistically calculated in advance, and the obtained value is used as the feature value of the pixel column (S1706). ).

Here, a known wavelet transform is performed on this feature value for each pixel column (S1708).
The value obtained by the wavelet transform is used for the peak detection processing described with reference to FIGS. 15 and 16 in line units.

The following shows the peak detection process performed in step S1710 based on FIG.
In the peak detection processing flow of step S1710, first, the start pixel column of the pixel column to be processed in the line is designated as the leftmost pixel column in the example of FIG. 18B (S1500).

Subsequently, level average values (in this case, average values of wavelet transform results) are obtained for all pixel columns on the same line located in the direction opposite to the scanning direction from the designated pixel column ( S1502)
Then, based on the average value of the wavelet conversion results, the peak determination of the pixel row to be processed is performed (S1504).

  This is because the average value of the wavelet conversion result is multiplied by a constant α calculated statistically in advance, and the wavelet conversion result of the pixel row to be processed is multiplied by α. It is determined whether or not the vertex exceeds the average value of the results.

  If it is determined in step S1504 that the target wavelet transformation result is not the vertex, the pixel row to be scanned next is reset (S1506), and the processing is repeated from step S1502 in the order of the steps. .

Further, when it is determined that the target wavelet transformation result is the vertex, the peak detection process is terminated.
Now, after the peak detection process in step S1710 is completed, it is determined whether or not the line to be processed is the last line designated as a temporary edge (S1712).

If it is not the last line, the line to be scanned next is set (S1714), and the above-described steps are repeated from step S1702.
If the final line is determined in step S1712, linear approximation on the image plane is performed on all pixel columns in which peaks are detected in step S1710 in the same manner as the processing shown in step S1128 of FIG. Thus, the pixel on the obtained approximate straight line is determined as the edge portion at the left end of the document image (S1716).

Finally, robust correction is performed and this process is terminated (S1718).
The robust correction is performed based on the following flow.
FIG. 19 is a processing flow of robust correction.

As shown in the figure, the slope of the edge is first obtained by obtaining the X-axis difference (Δx) on the image plane between the pixel columns determined as the edge (S1900).
Subsequently, an average value and a variance value are calculated for each of the obtained slopes (S1902).

Here, pixel rows whose difference (Δx) obtained in step S1900 exceeds the average value and the variance value are excluded (S1904).
Further, a pixel column in which the difference (Δx) is in the negative direction is excluded (S1906).

Then, linear approximation by the least square method is performed (S1980).
In subsequent step S1910, the number of times of this process is determined. In this example, if this process has not been completed three times, the process proceeds to step S1912 and the approximate straight line obtained in step S1908 and The distance value with the pixel column used for the calculation is calculated, and when there is a pixel column exceeding a predetermined threshold, the pixel column exceeding the threshold is excluded (S1912).

In the subsequent processing, the threshold value is changed (S1914), and the steps described above from step S1900 are executed in order.
Finally, when it is determined in step S1910 that this process has been executed three times, the pixel on the latest approximate straight line obtained at that time is finally determined as the edge portion on the left end of the document image, and all the processes are performed. finish.

  Note that the above operation flow (FIG. 17) may be executed only for the color having the largest characteristic in the spectral value of each RGB color, or for two or more colors arbitrarily combined in RGB. It may be executed.

As described above, the edge determination unit 106 in FIG. 1 may perform the edge specifying process only for the edge position temporarily obtained by the temporary edge determination unit 104 in FIG.
Since only the vicinity that is expected to be an edge can be processed, even if wavelet transform that is sensitive to noise (other image information) in an image area unrelated to the edge is applied, the edge is included. A more accurate edge position can be detected in the target region where the range of other image regions that are not present is narrow and noise is low.

  In addition, all the processes described above are described only for the left edge of the document for easy understanding of the description. However, the other edges (the upper edge, the right edge, or the lower edge of the document) also have their respective directions. It is possible to determine each edge by performing scanning in the opposite direction from each other, and by integrating these processing results, it becomes possible to determine all the edges surrounding the document, the document size, and the like. .

FIG. 20 shows an example of the overall processing flow in which the four edges (the left edge, the right edge, the upper edge, and the lower edge of the document) are detected and the edges around the document are specified.
Each processing is shown on the left side of the figure, and a processing image of a framed document image (document area is shown by shading) A corresponding to each process is shown on the right side of the figure.

  In this process, it is assumed that the print area included in the framed document image A is excluded by applying the character area exclusion process to the above-mentioned a. First, the vertical edge is detected, and then It was decided to detect the horizontal edge.

  First, in order to detect a vertical edge, a feature amount (for example, an image frequency (or spectrum) or a frequency distribution included in the region) from the framed document image A, in this example, 32 pixels vertical by 4 pixels horizontal. For each pixel area as a unit, extraction is performed in units of 32 pixels in the vertical direction as indicated by a horizontal line in the framed document image of the same process (S2000).

  Subsequently, based on the scanning from the left end side to the right side of the figure for the left edge, a position corresponding to the left end edge is provisionally determined from the extracted feature amount, and the provisional determination position of the left end edge and its vicinity ( The feature amount in the pixel region adjacent to the pixel region is extracted, and the edge position detected based on the feature amount is determined as the position of the left end edge (S2002).

  Subsequently, based on scanning from the right side to the left side of the figure for the right edge, a position corresponding to the right end edge is provisionally determined from the extracted feature value, and the provisional determination position of the right end edge and its vicinity ( The feature amount in the pixel region adjacent to the pixel region is extracted, and the edge position detected based on the feature amount is determined as the position of the right end edge (S2004).

Next, the horizontal edge is detected.
First, in order to detect a horizontal edge, a feature amount (for example, an image frequency (or spectrum) or a frequency distribution included in the region) is detected from the framed document image A. For each pixel area as a unit, extraction is performed in units of lines of 32 pixels in the horizontal direction as indicated by vertical lines in the framed document image of the same processing (S2006).

  Subsequently, based on the scanning from the upper end side to the lower end side in the figure for the upper edge, the position corresponding to the upper edge is provisionally determined from the extracted feature amount, and the upper edge is temporarily determined and its vicinity. The feature amount in the (pixel region adjacent to the pixel region) is extracted, and the edge position detected based on the feature amount is determined as the position of the upper edge (S2008).

  Subsequently, based on the scanning from the lower side to the upper side of the figure for the lower edge, the position corresponding to the lower edge is provisionally determined from the extracted feature value, and the position of the lower edge is temporarily determined and in the vicinity thereof. The feature amount is extracted, and the edge position detected based on the feature amount is determined as the position of the lower edge (S2010).

  In the case where the vertical scanning is performed in order to detect the upper edge or the lower edge, the horizontal scanning described above is used for the pixel setting of the line described above or the pixel setting of the range for specifying the pixel region. It is desirable to reverse the vertical and horizontal pixel settings in

By performing processing from four directions in this way, four edges of the document can be specified.
In the above description, it is assumed that the document has a white margin area around it, and the background plate has the same white color as the margin area, but the present invention is naturally limited to this. In addition, the present invention can be applied to other colors.

  For the sake of convenience, the description has been made assuming that the print area where characters and ruled lines are printed is occupied except for the margin area of the document. However, the above-described print area exclusion process is omitted for a document having no print area. It is also possible.

  Each function included in the various operation flows described above is a program code that can be read by a CPU (Central Processing Unit) in a memory such as a RAM (Random Access Memory) or a ROM (Read Only Memory) configured in the image processing apparatus. The program is executed by a CPU stored in the form of the above and connected to the memory by a bus.

A copy machine is an example of such an image processing apparatus.
FIG. 21 shows a block diagram of the copier.
In the copier 2100 in FIG. 2, the surface of the document C (read target surface) is set on the reading surface 2104 configured in the image reading unit 2102, and the document cover 2106 covering the document is covered from the back of the document. In this state, the image is read.

  The image reading unit 2102 reads an image by irradiating light from the reading surface 2104 side toward the set original C, and the light reading unit 2108 (CCD) configured as the image reading unit outputs one line of reflected light. Each time it is read, the read document surface information is converted into an electrical signal, and the converted image information is sent to the image processing unit 2110 that performs the edge detection process described above.

  As described above, the image processing unit 2110 can detect the boundary between the background image of the same color and the margin of the document (the edge of the document), and sends the edge information obtained by this detection to the document size detection unit 2112. By specifying the size, a copy of the original C with the specified original size can be printed out from the printing unit 2114 to the discharge stand 2116 of the copier.

FIG. 22 is an example of a hardware configuration that uses the image processing technique described above.
In the drawing, a scanner 2200 and an image processing personal computer (PC) 2202 are connected by a cable 2204.

  In such a configuration, the program is stored in a memory (not shown) such as a RAM (Random Access Memory) or a ROM (Read Only Memory) configured in the PC 2202, and the program is stored in a CPU bus-connected to the memory. By executing this, document edge information is extracted from image data read by the scanner 2200 and transmitted to the PC 2202 via the cable 2204.

FIG. 23 is an example of a hardware configuration that uses the image processing technique described above.
In FIG. 20, in addition to the scanner 2200 and the PC 2202 of FIG. 20, an image filing system in which an image data storage unit 2300 that stores image data that has been edited and a filing PC are connected by a cable 2204 is configured.

  In such a configuration, image data read from the scanner 2200 by driving the scanner driver 2304 of the PC 2202 is taken into the PC 2202, and the original edge detection processing 2306 is performed on the fetched image data, and the obtained edge is obtained. The image data storage unit 2300 performs image editing processing 2308 that cuts out the background plate image based on the information to optimize the image size, or performs skew correction by rotation based on the obtained edge information. Store edited image data.

  At this time, the related information of the image data stored in the image data storage unit 2300 is associated with a management table (not shown) of the filing PC 2302 for each image data, and the filing PC 2302 manages the image data.

FIG. 24 is an example of an operation flow in the hardware configuration of FIG.
First, the scanner driver 2304 is driven and the image data read from the scanner 2200 is taken into the PC 2202 (S2400).

Then, the edge detection process described in detail above is performed on the image data (S2402), and the edge information of the document is acquired (S2404).
Here, as editing of the image data, it is determined whether or not the image area is surrounded by the four sides (S2406), and only the image area (original image) surrounded by the four sides is extracted. (S2408).

Then, the image document optimized in this way is recorded in the image data storage unit 2300 (S2410).
FIG. 25 is an example of an operation flow in the hardware configuration of FIG.

First, the scanner driver 2304 is driven and the image data read from the scanner 2200 is taken into the PC 2202 (S2500).
Then, the edge detection process described in detail above is performed on the image data (S2502), and the edge information of the document is acquired (S2504).

  Here, as editing of the image data, for example, the minimum inclination of the acquired edge with respect to the axis (X axis or Y axis) on the image plane is detected (S2506), and the image data is rotated so that the inclination disappears. To correct the skew (S2508).

Then, the image document corrected in this way is recorded in the image data storage unit 2300 (S2510).
It is possible to combine the operation flows of FIG. 24 and FIG. 25. For example, the processing of step S2506 and step S2508 of FIG. 25 is performed between the processing of step S2404 and step S2406 of FIG. be able to.

By performing the skew correction in this way, post-processing can be facilitated.
Also, by cutting off the background image data, it becomes possible to remove excess data, and the amount of data can be reduced, so that data can be stored in a small capacity and the processing speed can be increased in the use of data. .

Next, FIG. 26 is an extended example of processing executed by the PC 2202 having the two configurations described above.
This figure shows processing in a PC 2202 having a Windows (registered trademark) environment. Under this environment, the scanner driver 2304 drives the framed original image A from the scanner 2200, loads it into the PC 2202, and loads the image A into the image A. On the basis of the obtained edge information, the document edge detection processing 2306 is executed to generate a black background plate image E in which the outer background plate image is colored black. The result of executing the character recognition processing 2308 is passed to the application 2310, and the character-recognized information (such as code data) can be extracted from the application 2310.

FIG. 27 is a flow for generating the black background board image E.
First, the scanner driver 2304 is driven and the image data A read from the scanner 2200 is taken into the PC 2202 (S2700).

Then, the edge detection process described in detail above is performed on the image data (S2702), and the edge information of the document is acquired (S2704).
Further, an image area other than the image area surrounded by the four edges is determined for the image data (S2706), and the gradation of the image area other than the image area (original image) surrounded by the four edges is determined. The density is changed to black (S2708).

  Finally, the starting point of the document image (for example, the position of the upper left end of the document image on the image plane) is specified, and for example, the image is printed for the print position from the starting point stored in advance in the storage unit or the like. Character recognition processing such as cutting out and analyzing the image and extracting characters included in the image based on the pattern recognition processing is performed (S2710).

  Note that this processing can be combined with the operation flow of FIG. 25. For example, the processing of step S2506 and step S2508 of FIG. 25 is performed between the processing of step S2704 and step S2706 of FIG. be able to.

  In this way, by coloring the background plate image to black, the image data of the same specifications as the conventional image having the black background plate image taken around the edge of the original image captured from the scanner of the conventional black background plate is obtained. Since the image data can be created, the image data obtained through the image processing apparatus of the present invention can be used by a conventional apparatus (such as an OCR processing apparatus) for editing image data having a black background. .

Each processing described above (of course, including edge detection processing) can also be distributed in the form of a program.
In that case, the program is recorded and distributed on a recording medium such as a floppy (registered trademark) disk, a CD-ROM, or a DVD (for example, a CD-ROM inserted in the CD-ROM insertion unit 2206 in FIG. 22). Alternatively, part or all of the program can be distributed via a transmission medium used in a public network or the like.

  A user who acquires the program distributed in the above form is configured in a data processing device such as a computer, for example, a reading unit that reads information recorded on the recording medium or a communication unit that performs data communication with an external device The above program is read into a memory such as a RAM or a ROM connected to each other via a bus, and the program is executed by a CPU (central processing unit) connected to the memory via a bus. Thus, the image processing can be realized in the user data processing apparatus.

As described above, in the embodiment of the present invention, the processing target is narrowed by first excluding the print area, so that the subsequent processing can be performed at high speed.
Further, since the amount of change in the feature information with the adjacent area based on the print area can be ignored, the boundary between the background plate and the document can be easily specified.

In the embodiment of the present invention, first, for the purpose of slightly widening the unit area of the image, the boundary position between the background plate and the document on the image is provisionally determined based on the feature amount obtained by 2DFFT, Thereafter, only a region including the tentatively determined position and error range is subjected to a one-dimensional fast Fourier transform (1DFFT) with a region smaller than the unit region as a unit, and a wavelet transform is further performed. As a result, processing up to detecting the boundary can be performed at high speed, and the position of the boundary can be detected with high accuracy.
Further, since the image processing does not select a combination of the background plate and the original material, it can be applied to an original image read from a scanner having any background plate.

(Supplementary Note 1) A boundary detection method for detecting a boundary between regions having different characteristics in any adjacent region on image data,
For each unit area on the image data, extract feature information determined by all pixels in the unit area,
Finding the difference of the feature information from the adjacent unit region based on the arbitrary unit region,
Determining the reference unit region as the boundary, at which the difference is at least a predetermined level,
A boundary detection method characterized by that.
(Supplementary Note 2) A boundary detection method for detecting a boundary between regions having different pixel arrangements in any adjacent region on image data,
Image frequency information is extracted for each unit area of the image data,
For each unit region, determine representative feature information based on the extracted image frequency information,
The difference between the representative feature information with the adjacent unit region is obtained with reference to the arbitrary unit region,
Determining the reference unit region as the boundary, at which the difference is at least a predetermined level,
A boundary detection method characterized by that.
(Supplementary Note 3) A boundary detection method for detecting a boundary between regions having different pixel arrangements in any adjacent region on image data,
For each first unit region of the image data, extract the first image frequency information,
For each first unit region, a value obtained by adding various kinds of the extracted first image frequency information with a predetermined weight is used as representative feature information.
In the representative feature information of any of the first unit regions, a first unit region whose amount of change from the representative feature information of the adjacent first unit region is at least a predetermined level is provisionally determined as the boundary,
Second image frequency information is extracted for each second unit region that is narrower than the first unit region in the first unit region that is provisionally determined as the boundary and in the vicinity region of the first unit region,
A second unit region in which a value based on a change amount of the second image frequency information of the adjacent second unit region with respect to the second image frequency information of an arbitrary second unit region is at least a predetermined level or more. As a boundary,
A boundary detection method characterized by that.
(Supplementary Note 4) In an input image data composed of a part of the background plate and the document, the image is input from the front with the background plate arranged on the back side, and a part of the background plate and the edge of the document Is a boundary detection method for detecting a boundary between a part of the background plate and the original of input image data input with the same color image,
Fourier transform is performed for each first unit region of the input image data,
Extracting the first image frequency information obtained by the Fourier transform;
For each first unit region, a value obtained by adding various kinds of the extracted first image frequency information with a predetermined weight is used as representative feature information.
Corresponding to the approximate straight line by applying a linear approximation to the first unit area in which the amount of change from the representative feature information of the adjacent first unit area is at least a predetermined level in any one of the first unit areas Temporarily determine the region as the boundary,
In the region tentatively determined as the boundary and the region near the region, Fourier transform is performed for each second unit region narrower than the first unit region,
Extracting second image frequency information obtained by the Fourier transform;
For each second unit region, a value obtained by adding various kinds of the extracted second image frequency information with a predetermined weight is used as representative feature information.
Wavelet transformation is performed on the representative feature information,
A linear approximation is applied to the second unit region where the value for each second unit region obtained by the wavelet transform is at least a predetermined level,
A position on an approximate line obtained by the straight line approximation is determined as the boundary;
A boundary detection method characterized by that.
(Supplementary Note 5) In the image data in which the four sides of the document are surrounded by the background plate, the detection is performed from four directions on the image data that are parallel or orthogonal to each other and different from each other. 5. The boundary detection method according to 4.
(Supplementary note 6) Supplementary notes 1 to 5 characterized in that, for the image data or the input image data, first, a print area is specified, and the image data or the input image data excluding the print area is a processing target. The boundary detection method according to any one of the above.
(Supplementary note 7) The boundary detection method according to any one of supplementary notes 1 to 6, wherein the feature information or the frequency information is a high-frequency component, a low-frequency component, a DC component, and a frequency distribution.
(Supplementary Note 8) When the determined boundary is a rectangle, the skew correction of the region included in the rectangle is performed based on the boundary.
The boundary detection method according to any one of appendices 1 to 7, wherein:
(Additional remark 9) Based on the determined boundary, leaving one region adjacent to the boundary or the original portion of the image data, and removing the other region or the background plate portion of the image data,
The boundary detection method according to any one of appendices 1 to 8, wherein:
(Supplementary Note 10) Based on the determined boundary, one region adjacent to the boundary or the background plate portion of the image data is colored black,
Recognize characters included in the other area other than the black colored area adjacent to the boundary;
The boundary detection method according to any one of appendices 1 to 8, wherein:
(Supplementary Note 11) A program for causing a computer to execute a process of detecting a boundary between regions having different characteristics in any adjacent region on image data,
A function for extracting feature information determined by all pixels in the unit area for each unit area on the image data;
A function for obtaining a difference in the feature information from the adjacent unit region with reference to the arbitrary unit region;
A function for determining the reference unit region as the boundary, the difference being at least a predetermined level;
A program to realize
(Supplementary Note 12) A program for causing a computer to execute a process of detecting a boundary between regions having different pixel arrays in any adjacent region on image data,
A function of extracting image frequency information for each unit area of the image data;
A function for determining representative feature information based on the extracted image frequency information for each unit region;
A function of obtaining a difference in the representative feature information from the adjacent unit region with respect to any unit region;
A function for determining the reference unit region as the boundary, the difference being at least a predetermined level;
A program to realize
(Supplementary note 13) A program for causing a computer to execute processing for detecting a boundary between regions having different pixel arrangements in any adjacent region on image data,
A function of extracting first image frequency information for each first unit region of the image data;
For each first unit region, a function of adding various values of the extracted first image frequency information with a predetermined weight as representative feature information;
In the representative feature information of any of the first unit regions, a function of temporarily determining, as the boundary, a first unit region whose amount of change from the representative feature information of the adjacent first unit region is at least a predetermined level;
A function of extracting second image frequency information for each second unit region narrower than the first unit region in the first unit region temporarily determined as the boundary and a region near the first unit region;
A second unit region in which a value based on a change amount of the second image frequency information of the adjacent second unit region with respect to the second image frequency information of an arbitrary second unit region is at least a predetermined level or more. The ability to determine the boundary,
A program to realize
(Supplementary Note 14) In an input image data composed of a part of the background plate and the document, the image is input from the front with the background plate arranged on the back side, and a part of the background plate and the edge of the document Is a program for causing a computer to execute a process of detecting a boundary between a part of the background plate and the document of input image data input with the same color image,
A function of performing a Fourier transform for each first unit region of the input image data;
A function of extracting first image frequency information obtained by the Fourier transform;
For each first unit region, a function of adding various values of the extracted first image frequency information with a predetermined weight as representative feature information;
Corresponding to the approximate straight line by applying a linear approximation to the first unit area in which the amount of change from the representative feature information of the adjacent first unit area is at least a predetermined level in any one of the first unit areas A function of temporarily determining a region as the boundary;
In a region provisionally determined as the boundary and a region near the region, a function of performing a Fourier transform for each second unit region narrower than the first unit region;
A function of extracting second image frequency information obtained by the Fourier transform;
For each second unit region, a function of adding various values of the extracted second image frequency information with a predetermined weight as representative feature information;
A function of performing wavelet transformation on the representative feature information;
A function of performing a linear approximation on the second unit region in which the value for each second unit region obtained by the wavelet transform is at least a predetermined level;
A function for determining a position on an approximate line obtained by the straight line approximation as the boundary;
A program to realize
(Supplementary Note 15) In the image data in which the four sides of the document are surrounded by the background plate, the function of performing the detection from four directions on the image data that are parallel or orthogonal to each other and different from each other is further realized. The program according to appendix 14, characterized by:
(Additional remark 16) Additional remarks 11 thru | or 15 which implement | achieve further the function which specifies a printing area | region first and excludes this printing area | region with respect to the said image data or the said input image data, and processes the input image data. The program as described in any one of.
(Supplementary note 17) The program according to any one of supplementary notes 11 to 16, wherein the feature information or the frequency information is a high-frequency component, a low-frequency component, a DC component, and a frequency distribution.
(Additional remark 18) When the determined boundary is a rectangle, the function which performs skew correction of the area | region included in this rectangle based on this boundary is further implement | achieved, The additional description 11 thru | or 17 program.
(Additional remark 19) Based on the determined boundary, the function of removing the other region or the background plate portion of the image data by leaving one region adjacent to the boundary or the original portion of the image data is further realized. The program according to any one of appendices 11 to 18.
(Supplementary Note 20) Based on the determined boundary, a function of coloring one region adjacent to the boundary or a background plate portion of the image data to black,
A function of recognizing characters included in the other region other than the black colored region adjacent to the boundary;
The program according to any one of appendices 11 to 19, further realizing the above.
(Supplementary note 21) An image processing apparatus for detecting a boundary between regions having different characteristics in any adjacent region on image data,
Feature enhancement means for extracting feature information determined by all pixels in the unit region for each unit region on the image data;
Boundary determining means for obtaining a difference in the feature information from the adjacent unit region with respect to any unit region, and determining the reference unit region as the boundary, the difference being at least a predetermined level;
An image processing apparatus comprising:
(Supplementary note 22) An image processing apparatus for detecting a boundary between regions having different pixel arrangements in any adjacent region on image data,
Feature enhancement means for extracting image frequency information for each unit region of the image data;
For each unit region, representative feature information based on the extracted image frequency information is determined, and a difference between the representative feature information with the adjacent unit region is obtained with reference to any unit region, and the difference is at least Boundary determining means for determining the reference unit region as the boundary that is equal to or higher than a predetermined level;
An image processing apparatus comprising:
(Supplementary note 23) An image processing apparatus for detecting a boundary between regions having different pixel arrangements in any adjacent region on image data,
Feature enhancement means for extracting first image frequency information for each first unit region of the image data;
For each of the first unit areas, a value obtained by adding various kinds of the extracted first image frequency information with a predetermined weight is used as representative feature information. Tentative boundary determination means for tentatively determining, as the boundary, a first unit area whose amount of change from the representative feature information of one unit area is at least a predetermined level;
Second image frequency information is extracted for each second unit region that is narrower than the first unit region in the first unit region that is provisionally determined as the boundary and in the vicinity of the first unit region, and any second unit Boundary determination for determining, as the boundary, a second unit area in which a value based on a change amount of the second image frequency information of the area with the second image frequency information of the adjacent second unit area is at least a predetermined level or more Means,
An image processing apparatus comprising:
(Supplementary Note 24) In an input image data composed of a part of the background plate and the document, the image is input from the front with the background plate arranged on the back side. Is an image processing apparatus for detecting a boundary between a part of the background plate and the original of input image data input with the same color image,
Applying a Fourier transform to each first unit region of the input image data, and extracting the first image frequency information obtained by the Fourier transform;
For each first unit area, a value obtained by adding various kinds of the extracted first image frequency information with a predetermined weight is used as representative feature information, and a representative of the adjacent first unit area in any one of the first unit areas Boundary provisional deciding means for tentatively deciding a region corresponding to the approximate straight line as the boundary by performing linear approximation on the first unit region whose amount of change from the feature information is at least a predetermined level;
In the region tentatively determined as the boundary and in the vicinity of the region, Fourier transform is performed for each second unit region narrower than the first unit region, second image frequency information obtained by the Fourier transform is extracted, and the second unit region is extracted. For each two unit regions, a value obtained by adding various kinds of the extracted second image frequency information with a predetermined weight is used as representative feature information, wavelet transform is performed on the representative feature information, and the wavelet transform is obtained. Boundary determination means for performing linear approximation on the second unit region in which the value for each second unit region is at least a predetermined level and determining the position on the approximate line obtained by the linear approximation as the boundary;
An image processing apparatus comprising:
(Supplementary Note 25) In the image data in which the four sides of the document are surrounded by the background plate, each means functions in four directions on the image data that are parallel or orthogonal to each other and different from each other. The image processing apparatus according to Supplementary Note 24, which is a feature.
(Supplementary note 26) The image processing apparatus according to any one of claims 21 to 25, further comprising a print area exclusion unit that excludes a print area of the image data.
(Supplementary note 27) The image processing apparatus according to any one of claims 21 to 26, wherein the feature information or the frequency information is a high-frequency component, a low-frequency component, a DC component, and a frequency distribution. .
(Supplementary note 28) The image processing apparatus according to any one of supplementary notes 21 to 27, further comprising image reading means for optically reading a document to generate the image data of the document.
(Supplementary note 29) The supplementary notes 21 to 28, further comprising skew correction means for performing skew correction on a region included in the rectangle based on the boundary when the determined boundary is a rectangle. The image processing apparatus according to any one of the above.
(Supplementary Note 30) The image processing apparatus further includes an image excluding unit that removes the other region or the background plate portion of the image data while leaving one region adjacent to the boundary or the original portion of the image data based on the determined boundary. 30. The image processing device according to any one of appendices 21 to 29, wherein
(Supplementary Note 31) Based on the determined boundary, image coloring means for coloring one area adjacent to the boundary or a background plate portion of the image data to black,
A character recognition means for recognizing a character included in the other area other than the black colored area adjacent to the boundary;
31. The image processing apparatus according to any one of appendices 21 to 30, wherein the image processing apparatus includes:
(Supplementary Note 32) A background plate is arranged on the back side of the document, image information of the document is read from the front of the document, and a copy of the document is output by specifying the document size based on detection of the boundary between the background plate and the document. Copy machine
A print area exclusion unit for excluding the print area of image information;
Feature enhancement means for extracting feature information determined by all image information in the unit area for each unit area on the image information;
The difference between the feature information with the adjacent unit area is obtained with reference to the arbitrary unit area, and the reference unit area other than the print area where the difference is at least a predetermined level is determined as the boundary. Boundary determination means;
A copier characterized by comprising:
(Supplementary Note 33) A background plate is arranged on the back side of the document, image information of the document is read from the front of the document, and a copy of the document is output by specifying the document size based on detection of the boundary between the background plate and the document. Copy machine
A print area exclusion unit for excluding the print area of image information;
Feature enhancement means for extracting image frequency information for each unit region of the image information;
For each unit region, representative feature information based on the extracted image frequency information is determined, and a difference between the representative feature information with the adjacent unit region is obtained with reference to any unit region, and the difference is at least Boundary determining means for determining, as the boundary, the reference unit area other than the print area that is equal to or higher than a predetermined level;
A copier characterized by comprising:
(Supplementary Note 34) A background plate is arranged on the back side of the document, image information of the document is read from the front of the document, and a copy of the document is output by specifying the document size based on detection of the boundary between the background plate and the document. Copy machine
A print area exclusion unit for excluding the print area of image information;
Feature enhancement means for extracting first image frequency information for each first unit region of the image information;
For each of the first unit areas, a value obtained by adding various kinds of the extracted first image frequency information with a predetermined weight is used as representative feature information. Temporary boundary determination means for temporarily determining, as the boundary, a first unit area other than the print area in which the amount of change from the representative feature information of one unit area is at least a predetermined level;
Second image frequency information is extracted for each second unit region that is narrower than the first unit region in the first unit region that is provisionally determined as the boundary and in the vicinity of the first unit region, and any second unit Boundary determination for determining, as the boundary, a second unit area in which a value based on a change amount of the second image frequency information of the area with the second image frequency information of the adjacent second unit area is at least a predetermined level or more Means,
A copier characterized by comprising:
(Supplementary Note 35) A background plate is placed on the back side of the document, image information of the document is read from the front of the document, and a copy of the document is output by specifying the document size based on detection of the boundary between the background plate and the document. Copy machine
A print area exclusion unit for excluding the print area of image information;
Feature enhancement means for performing Fourier transform for each first unit region of the image information and extracting first image frequency information obtained by the Fourier transform;
For each first unit area, a value obtained by adding various kinds of the extracted first image frequency information with a predetermined weight is used as representative feature information, and a representative of the adjacent first unit area in any one of the first unit areas Boundary provisional deciding means for provisionally deciding a region corresponding to the approximate straight line as the boundary by performing linear approximation on the first unit region other than the print region where the amount of change from the feature information is at least a predetermined level or more ,
In the region tentatively determined as the boundary and in the vicinity of the region, Fourier transform is performed for each second unit region narrower than the first unit region, second image frequency information obtained by the Fourier transform is extracted, and the second unit region is extracted. For each two unit regions, a value obtained by adding various kinds of the extracted second image frequency information with a predetermined weight is used as representative feature information, wavelet transform is performed on the representative feature information, and the wavelet transform is obtained. Boundary determination means for performing linear approximation on the second unit region in which the value for each second unit region is at least a predetermined level and determining the position on the approximate line obtained by the linear approximation as the boundary;
A copier characterized by comprising:
(Supplementary Note 36) In the image data in which the four sides of the document are surrounded by the background plate, each means functions in four directions on the image data that are parallel or orthogonal to each other and different from each other. The copying machine according to Supplementary Note 35, which is characterized.
(Supplementary note 37) The copier according to any one of supplementary notes 32 to 36, wherein a print area of the image information is excluded based on a gradation value.

  As described above, according to the present invention, it is possible to detect the edge of an original from image data read in an arbitrary combination of an original and a background plate, and image data input from an existing image reading apparatus. Even so, the boundary between the document and the background plate can be detected.

  In addition, OCR processing is maintained while maintaining the superiority of white background readers, that is, the characteristics of high contrast, the ability to read thin paper, and the characteristic of not showing through, which cannot be obtained with a black background reader. Can be done.

1 is a block diagram of an image processing apparatus according to an embodiment of the present invention. It is a display example of a document image with a frame. 3 is an example of an operation flow performed in a print area exclusion unit 100. FIG. 6 is a comparison diagram between a print area specified by a print area exclusion unit 100 and an original image. It is an example of the operation | movement flow performed in the feature emphasis part 102. In this example, the framed document image A is divided into unit pixel areas for scanning. This is actual measurement data of the feature amount obtained in the rectangular area 210 of FIG. It is a graph of the high frequency component based on the measurement data of FIG. It is a graph of the low frequency component based on the measurement data of FIG. It is a graph of the DC component based on the actual measurement data of FIG. It is an example of the operation | movement flow performed in the edge temporary determination part 104. FIG. This is an arrangement relationship of pixel areas for which an average of feature amounts is to be obtained. This is an arrangement of edges temporarily determined for the document image A with a frame. It is an operation | movement flow of a weight determination process. It is an operation | movement flow of a peak detection process. It is a level distribution diagram for a predetermined line. It is an operation | movement flow performed in the edge determination part 106. FIG. FIG. 6 is a diagram showing a unit pixel area performed in an edge determination unit 106 for a framed document image A. It is a processing flow of robust correction. It is an example of the whole processing flow which specifies the edge around a document. 1 is a block diagram of a copier. It is an example of the hardware constitutions which use this image processing technique. (Part 1) It is an example of the hardware constitutions which use this image processing technique. (Part 2) It is an example of the operation | movement flow in the hardware constitutions of FIG. (Part 1) It is an example of the operation | movement flow in the hardware constitutions of FIG. (Part 2) It is an application example of this image processing technology. It is a flow for generating a black background board image E.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Print area exclusion part 102 Feature emphasis part 104 Edge temporary determination part 106 Edge determination part A Document image with a frame B Edge information

Claims (6)

Extracting the frequency component of the image data for each region having a plurality of pixels in the horizontal and vertical directions of the image data read by the image reading means;
Obtain the amount of change in the frequency component of each extracted area, detect the peak of the amount of change in the frequency component, perform linear approximation on the detected peak, and use multiple pixels on the approximate line obtained by linear approximation as boundaries Tentatively decided,
Extracting frequency components of a pixel column that includes a pixel that is tentatively determined as a boundary that is narrower than the region and a predetermined number of pixel columns before and after the pixel column
Obtain the amount of change in the frequency component of the extracted plurality of pixel columns, detect the peak of the amount of change in the frequency component, perform linear approximation on the detected peak, and then apply the linear approximation to the pixels on the approximate line obtained by linear approximation Boundary detection method that decides as a boundary. A boundary detection method for reading a document placed on a background plate by an image reading unit and detecting a boundary between the background plate and the document in the read image data,
Apply Fourier transform to the image data of each region, extract the high frequency component, low frequency component and DC component of the image data obtained by the Fourier transform,
Calculates the amount of change in the weighted values of the high-frequency component, low-frequency component, and DC component of the extracted image data for each extracted area, detects the peak of the change, and linearizes the detected peak Apply approximation, provisionally determine the plurality of pixels on the approximate line obtained by linear approximation as boundaries,
A high frequency of a plurality of pixel columns obtained by performing Fourier transform on a pixel column including the plurality of pixels provisionally determined as boundaries and a predetermined number of pixel columns before and after the pixel column and narrower than the region, and obtained by the Fourier transform Wavelet transform is applied to the value obtained by applying a predetermined weight to the component, low frequency component and direct current component values.
2. The boundary according to claim 1, wherein peaks of a plurality of values obtained by the wavelet transform are detected, linear approximation is performed on the detected peaks, and a plurality of pixels on the approximate straight line obtained by the linear approximation is determined as a boundary. Detection method. A process of extracting the frequency component of the image data for each region having a plurality of pixels in the horizontal and vertical directions of the image data read by the image reading means;
Obtain the amount of change in the frequency component of each extracted area, detect the peak of the amount of change in the frequency component, perform linear approximation on the detected peak, and use multiple pixels on the approximate line obtained by linear approximation as boundaries A provisional decision process;
A process of extracting a frequency component of a pixel column that is narrower than the region and includes a pixel provisionally determined as a boundary and a predetermined number of pixel columns before and after the pixel column;
Obtain the amount of change in the frequency component of the extracted plurality of pixel columns, detect the peak of the amount of change in the frequency component, perform linear approximation on the detected peak, and then apply the linear approximation to the pixels on the approximate line obtained by linear approximation A boundary detection program for causing a computer to execute processing for determining a boundary as a boundary . A boundary detection program for reading a document placed on a background plate by an image reading unit and detecting a boundary between the background plate and the document in the read image data,
The extraction process performs a Fourier transform on the image data of each region, and extracts a high frequency component, a low frequency component and a direct current component of the image data obtained by the Fourier transform,
The provisional determination process calculates a change amount of a value obtained by applying a predetermined weight to the values of the high frequency component, the low frequency component, and the direct current component of the image data of each extracted region, detects a peak of the change amount, and detects A linear approximation is performed on the peak, and the plurality of pixels on the approximate straight line obtained by the linear approximation are provisionally determined as boundaries,
In the extraction process, a plurality of pixels obtained by performing Fourier transform on a pixel array including the pixels tentatively determined as a boundary, which is narrower than the area, and a predetermined number of pixel arrays before and after the pixel array, are obtained by the Fourier transform. Wavelet transform is applied to the values obtained by applying a predetermined weight to the values of the high-frequency component, low-frequency component, and DC component of the column,
The determination process detects peaks of a plurality of values obtained by the wavelet transform, performs linear approximation on the detected peaks, and determines a plurality of pixels on an approximate line obtained by linear approximation as boundaries. Item 4. The boundary detection program according to item 3 . First extraction means for extracting a frequency component of the image data for each region having a plurality of pixels in the horizontal and vertical directions of the image data read by the image reading means;
A frequency component change amount of each region extracted by the first extraction means is obtained, a peak of the frequency component change amount is detected, a linear approximation is performed on the detected peak, and an approximate straight line obtained by linear approximation is obtained. Boundary provisional determination means for temporarily determining a plurality of pixels as boundaries,
A second extraction unit that extracts a frequency component of a pixel column that is narrower than the region and includes a pixel temporarily determined as a boundary, and a predetermined number of pixel columns before and after the pixel column;
The amount of change in the frequency component of the plurality of pixel columns extracted by the second extraction means is obtained, the peak of the amount of change in the frequency component is detected, the detected peak is subjected to linear approximation, and obtained by linear approximation. An image processing apparatus comprising: a boundary determining unit that determines a plurality of pixels on the approximate line as a boundary . An image processing apparatus that reads a document placed on a background plate by an image reading unit and detects a boundary between the background plate and the document in the read image data,
The first extraction means performs a Fourier transform on the image data of each region, extracts a high frequency component, a low frequency component and a direct current component of the image data obtained by the Fourier transform,
The temporary boundary determination means calculates a change amount of a value obtained by applying a predetermined weight to the values of the high frequency component, the low frequency component, and the direct current component of the image data of each region extracted by the first extraction means. A peak of the amount is detected, linear approximation is performed on the detected peak, and the plurality of pixels on the approximate straight line obtained by the linear approximation are provisionally determined as boundaries,
The second extraction means performs a Fourier transform on a pixel array that is narrower than the region and includes the pixels that are provisionally determined as boundaries, and a predetermined number of pixel arrays before and after the pixel array, and is obtained by the Fourier transform. Wavelet transform is applied to the values obtained by applying a predetermined weight to the values of the high frequency component, low frequency component and DC component of the plurality of pixel columns,
The boundary determining unit detects peaks of a plurality of values obtained by the wavelet transform, performs a linear approximation on the detected peaks, and determines a plurality of pixels on the approximate straight line obtained by the linear approximation as a boundary. The image processing apparatus according to claim 5 .

Images