JP2011216070A - Method for restoring hidden character and image processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for restoring hidden characters and an image processor which accurately restores hidden characters.SOLUTION: In the method for restoring hidden characters, the image processor calculates an image block size in accordance with widths of strokes of target characters including hidden characters to be restored, and marks an original area in a target area of an input image including the target characters. The image processor generates reference image blocks having pixels of the original area as centers respectively, and extracts a boundary of a foreground of the input image to be complemented. The image processor calculates, for each pixel on the boundary of the foreground, complementation priorities in accordance with complexities of image blocks having the pixels on the boundary as centers respectively, and uses the most similar image block among the reference image blocks to complement an image block wherein the pixel having the highest priority is taken as the center. If the foreground has not been completely complemented, the image processor extracts the boundary of the foreground in the input image to be complemented.

Description

  The present invention generally relates to character recognition, and more particularly to a method for restoring a concealed character and an image processing apparatus.

  2. Description of the Related Art Conventionally, an optical character recognition (OCR) system, for example, recognizes a character of an imaged object and outputs the recognized character. In recent years, with the spread of digital cameras and the increase in functions, OCR has been used more frequently. For example, the camera is a portable “scanner” that captures text from hardcopy documents, books, newspapers, slogans, posters, and other media.

  In such text recognition, the OCR system may recognize characters by mistake. For example, a technique relating to an OCR system that reduces erroneous recognition when a ruled line and a character stroke in a text are in contact with or intersecting with each other is disclosed. In addition, a technique related to an OCR system that distinguishes between characters in a text and a writing portion when the text to be recognized is written is disclosed.

JP 2002-366900 A JP 2010-55353 A JP 2010-55354 A

  However, the above-described conventional technique has a problem that the concealed character cannot be accurately restored.

  In general, the OCR system makes it difficult to recognize a visible watermark character by hiding the visible watermark character in the document image in the foreground content. FIG. 22 is a diagram illustrating an example of a watermark character concealed in the foreground. As shown in FIG. 22, the characters in the watermark are blocked by the foreground. For this reason, the OCR system cannot accurately restore characters in the watermark. Further, since the OCR system restores the characters in the watermark and recognizes the restored characters, it cannot directly recognize the characters “related party confidential” in the watermark.

  In addition, when text is captured by a digital camera, the text will always be obstructed by characters in the text due to non-uniform light rays, unwanted or existing obstructions, and processing effects inside the camera. . Such blockage makes it difficult to recognize characters in the text by the OCR system.

  The disclosed technology has been made in view of the above, and an object of the present invention is to provide a method for restoring a concealed character and an image processing apparatus capable of accurately restoring the concealed character.

  In the first proposal, a method for restoring a concealed character includes a step of calculating an image block size according to a stroke width of a target character including a concealed character to be restored, and an input image including the target character. Marking the original area in the target area; generating each reference image block centered on each of the pixels in the original area; extracting the foreground boundary of the input image to be complemented; and foreground For each pixel at the boundary, the step of calculating the complement priority according to the complexity of the image block centered on the pixel, and the highest priority using the most similar image block in the reference image block If the image block centered on a pixel having a pixel and the foreground completion are not all completed, And performing a process of extracting the boundary of the scene.

  Preferably, the reliability coefficient calculated by measuring the proportion occupied by the background pixels in the image block, the reliability coefficient calculated by measuring the proportion occupied by the target character pixel that is the complement target inside the image block, and the inside of the image block Complementary priority is calculated from the complexity based on the number of different gradient vectors.

  In a specific embodiment, the image block size is calculated from the stroke width of the target character.

  According to one preferred embodiment of the first scheme, the step of marking the original region in the target region of the input image containing the target character includes the pixel in the target character and all adjacent pixels in the image block. Generating a first mask, generating a second mask including the entire target area except foreground pixels and all adjacent pixels in the image block, and overlapping areas of the first mask and the second mask Extracting as a source region.

  The hidden character restoration method according to one of the preferred embodiments of the first scheme further comprises the steps of compressing the input image and / or selecting the reference image block.

  In the second proposal, the hidden character restoration method includes a step of extracting a contour pixel of a target character, a step of capturing a line segment of the target character based on the extracted contour pixel, and a captured line segment. Selecting a parallel line segment that is assumed to be a line segment to be paired with one stroke, and connecting collinear segments as a parallel line pair based on the selected parallel line segment, and And supplementing the region defined by each of the parallel line pairs using the gray value of the target character.

A method for restoring a concealed character according to another preferred embodiment of the second scheme includes the step of estimating a stroke width of a target character based on a captured line segment, and a character on a projection axis of the target character. Estimating the width and the character height. Here, in the Hough transform region, the selected parallel line segment satisfies the following condition.
1) For the line segment 1 (θ1, ρ1), the line segment 2 (θ2, ρ2) and the stroke width SW, θ1 = θ2 and a * SW <| ρ1-ρ2 | <b * SW are satisfied (here , Θ represents the rotation angle of the line segment, ρ represents the distance to the zero point of the line segment, and a * SW and b * SW are threshold values).
2) For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, while for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is the character height. Smaller than that.
3) Pixels between two line segments are located inside the stroke, not the background.
And the said collinear segment satisfy | fills the following conditions.
1) For line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ1 = θ2 and | ρ1−ρ2 | are satisfied to be smaller than the threshold values.
2) For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, while for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is the character height. Smaller than that.
3) It is satisfied that the pixel between the two line segments is located inside the stroke, not the background.

  In the third proposal, a method for restoring a concealed character includes a step of dividing a text area in an input image into a target character that is a recognition target, a background and a foreground, and a concealment that is a recognition target in the text area by the method described above. Restoring the restored character and recognizing the restored character.

  The hidden character restoration method according to one of the preferred embodiments of the third proposal further includes a step of extracting a target area including a target character to be recognized from the text area.

In one specific embodiment of the third plan, the step of extracting the target area including the target character to be recognized from the text area is performed by extracting a contour pixel of the target character, horizontal direction θ and vertical direction A step of performing Hough conversion on the contour pixel at (90-θ) (where θ indicates the rotation angle of the target character), and Hough conversion is being performed along the horizontal direction θ and the vertical direction (90-θ). Determining the minimum and maximum distance values ρ _h1 , ρ _h2 , ρ _v1 and ρ _v2 , four lines (ρ _h1 , 90−θ), (ρ _h2 , 90−θ), (ρ _v1 , θ), It includes a step of calculating four corners using (ρ _v2 , θ) and a step of cutting out a region covering the four corners as a target region.

  In the fourth plan, the image processing apparatus includes: an image block size calculating unit that calculates an image block size based on a stroke width of a target character including a hidden character to be restored; and an input image including the target character. An original area marking portion for marking the original area in the target area. Further, the image processing apparatus includes a reference image block generation unit that generates each reference image block centered on each pixel in the original region, a boundary extraction unit that extracts a foreground boundary of the input image to be complemented, Have The image processing apparatus includes a complementary priority calculating unit that calculates a complementary priority for each pixel at the boundary of the foreground according to the complexity of the image block centered on the pixel. In addition, the image processing apparatus includes a foreground complementing unit that complements an image block centered on a pixel having the highest priority by using the most similar image block in the reference image block. The image processing apparatus includes a determination unit that executes the process of the boundary extraction unit when all of the foreground interpolation is not completed.

  The complementation priority calculating unit measures the reliability coefficient calculated by measuring the ratio of the background pixels in the image block, the reliability coefficient calculated by measuring the ratio of the target character pixel to be complemented in the image block, Complementary priority is calculated from the complexity based on the number of different gradient vectors inside the image block.

  In the image processing apparatus according to one of the fourth proposed embodiments, the image block size is calculated from the stroke width of the target character.

  In the image processing apparatus according to the fourth plan, the original region marking unit generates a first mask including the pixel in the target character and all adjacent pixels in the image block. In addition, the original region marking unit generates a second mask that includes the entire target region other than the foreground pixels and all adjacent pixels in the image block. Then, the original area marking part extracts an overlapping area between the first mask and the second mask as an original area.

  An image processing apparatus according to one of the preferred embodiments of the fourth proposal compresses an input image and selects a reference image block.

  In the fifth plan, the image processing apparatus includes a contour pixel extracting unit that extracts a contour pixel of a target character, and a line segment capturing unit that captures a line segment of the target character based on the extracted contour pixel. Further, the image processing apparatus includes a parallel line selection unit that selects a parallel line segment that is assumed to be a line segment that forms a pair of strokes based on the captured line segment, and the selected parallel line segment. And a collinear line connecting part for connecting the collinear line segments as parallel line pairs. The image processing apparatus includes a region complementing unit that supplements a region limited by each parallel line pair using the gray value of the target character.

An image processing apparatus according to one of the embodiments of the fifth plan includes a stroke width estimation unit that estimates a stroke width of a target character based on a captured line segment, and a character width and a character height on a projection axis of the target character. And a size range estimation unit for estimating the size. Here, in the Hough transform region, the selected parallel line segment satisfies the following condition.
1) As for line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ1 = θ2 and a * SW <| ρ1-ρ2 | <b * SW must be satisfied (here Where θ represents the rotation angle of the line segment, ρ represents the distance to the zero point of the line segment, and a * SW and b * SW are threshold values).
2) For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, while for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is the character. Smaller than height.
3) It is satisfied that the pixel between the two line segments is located inside the stroke, not the background.
And the said collinear segment satisfy | fills the following conditions.
1) For line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, it is satisfied that θ1 = θ2 and | ρ1−ρ2 | are smaller than the threshold values.
2) For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, and for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is the character. Smaller than height.
3) It is satisfied that the pixel between the two line segments is located inside the stroke, not the background.

  In a sixth plan, the image processing apparatus includes a target character that is a recognition target in a text area in an input image, a dividing unit that divides the text region into a background and a foreground, a hidden character restoration device as described above, and a restored character. And a character recognition unit for recognizing.

  The image processing apparatus according to one of the embodiments of the sixth plan further includes a target area extracting unit that extracts a target area including a target character to be recognized from the text area.

In one specific example of the sixth plan, the target area extraction unit extracts the contour pixel of the target character, and performs Hough transform on the contour pixel in the horizontal direction θ and the vertical direction (90−θ). Where θ represents the rotation angle of the target character and the minimum and maximum distance values ρ _h1 , ρ _h2 , ρ _v1 and ρ _v2 during Hough transform along the horizontal direction θ and the vertical direction (90-θ). And calculate four angles using four lines (ρ _h1 , 90−θ), (ρ _h2 , 90−θ), (ρ _v1 , θ), (ρ _v2 , θ), and A region covering the four corners is cut out as a target region.

  Further, in the seventh plan, the storage medium includes a program code that can be read by the device, and when the program code is executed in the information processing apparatus, a method for restoring characters in which the program code is hidden in the information processing apparatus is executed. Let

  In the eighth plan, the program product includes an instruction that can be executed by the device. When the instruction is executed in the information processing apparatus, the instruction is executed by the information processing apparatus.

  According to one aspect of the concealed character restoration method and the image processing apparatus disclosed in the present application, there is an effect that the concealed character can be accurately restored.

FIG. 1 is a flowchart of a process procedure of a hidden character restoration method according to the first embodiment. FIG. 2A is a diagram illustrating an example of a concealed image input first. FIG. 2B is a diagram illustrating a restoration result when a 7 × 7 size image block is selected. FIG. 2C is a diagram illustrating a restoration result when a 17 × 17 size image block is selected. FIG. 3 is a diagram illustrating an example of the marked original area. FIG. 4A is a diagram illustrating an example of an image block having the lowest complexity. FIG. 4B is a diagram illustrating an example of an image block with low complexity. FIG. 4C is a diagram illustrating an example of an image block with high complexity. FIG. 4D is a diagram illustrating an example of an image block having the highest complexity. FIG. 5 is a flowchart of a process procedure of the hidden character restoration method according to the second embodiment. FIG. 6 is a diagram illustrating an example of a hidden character to be restored. FIG. 7 is a diagram illustrating an example in which the outline pixel of the target character in FIG. 6 is extracted. FIG. 8 is a diagram showing an example of a character line segment captured with respect to the contour pixel of the target character in FIG. FIG. 9 is a diagram illustrating an example of estimating the target character size in the second embodiment. FIG. 10 is a diagram illustrating an example of determining pixel attribution between two line segments in the second embodiment. FIG. 11 is a diagram showing an example of parallel line segments selected from the captured character line segments shown in FIG. FIG. 12 is a diagram illustrating an example of a result of executing the collinear line connecting process on the parallel line segments illustrated in FIG. 11. FIG. 13 is a diagram illustrating an example of complementing a region between parallel line segments of a target character in the second embodiment. FIG. 14A is a diagram schematically illustrating an example of a target character before completion. FIG. 14B is a diagram schematically illustrating an example of the target character after completion. FIG. 15 is a flowchart of the process procedure of the hidden character restoration method according to the third embodiment. FIG. 16A is a diagram illustrating an example of an input image including a concealed character that is a recognition target in the third embodiment. FIG. 16B is a diagram illustrating an example of a result obtained by performing an image text region dividing step on the input image illustrated in FIG. 16A. FIG. 16C is a diagram illustrating an example of a result of executing a process of extracting a target area with respect to the image text area illustrated in FIG. 16B. FIG. 16D is a diagram illustrating an example of a result of executing the process of restoring the concealed character with respect to the target area illustrated in FIG. 16C. FIG. 16E is a diagram illustrating an example of a result of executing the process of recognizing a hidden character after restoration illustrated in FIG. 16D. FIG. 17 is a diagram schematically illustrating an example of a target character extraction method according to the third embodiment. FIG. 18 is a block diagram illustrating the configuration of the image processing apparatus according to the fourth embodiment. FIG. 19 is a block diagram illustrating the configuration of the image processing apparatus according to the fifth embodiment. FIG. 20 is a block diagram illustrating the configuration of the image processing apparatus according to the sixth embodiment. FIG. 21 is a block diagram illustrating a configuration of an information processing apparatus that executes a method for restoring a concealed character according to the disclosed technology. FIG. 22 is a diagram illustrating an example of a watermark character concealed in the foreground.

  Hereinafter, embodiments of the disclosed technology will be described with reference to the drawings. For the sake of brevity and clarity, not all features of the actual embodiment are described in the specification, but the developer can achieve specific objectives in any practical application. As such, many decisions will be made depending on the embodiment. For example, it may be adapted to constraints related to the system and business, and these constraints may be changed according to the embodiment. In addition to these, the development work is very complicated and time-consuming, but for those skilled in the art who have an idea from the disclosed technology, these work are merely duties associated with practical use.

  Here, in order to avoid obscuring the disclosed technology, only the structure of the apparatus and / or processing steps directly related to the disclosed technology are shown in the drawings, and others not directly related to the disclosed technology are omitted. . In addition, elements and features described in one drawing or one embodiment of the disclosed technology may be combined with elements and features shown in one or more other drawings or embodiments.

  Referring to the following description of embodiments of the disclosed technology based on the drawings, the objects, features, and advantages of the disclosed technology can be more easily understood. Elements and features in the drawings are only intended to illustrate the mechanism of the disclosed technology. In the drawings, the same or similar technical features or elements are denoted by the same or similar reference numerals.

  Hereinafter, a hidden character restoration method according to the disclosed technology will be described in detail based on the drawings. For example, the image processing apparatus recognizes the watermark shown in FIG. 22 by separating and restoring it.

  FIG. 1 is a flowchart of a process procedure of a hidden character restoration method according to the first embodiment. As shown in FIG. 1, in the method for restoring a concealed character according to the first embodiment, the image processing apparatus performs an input image compression step S210, an image block size calculation step S220, an original region marking step S230, and a reference image block generation. Step S240 is executed. Subsequently, the image processing apparatus executes a reference image block selection step S250, a boundary extraction step S260, a complementation priority calculation step S270, a foreground complementation step S280, and a determination step S290.

  In order to increase the processing speed, the image processing apparatus compresses the input image in step S210. As a method for compressing an input image, any known technique in the art may be used as long as the size of the image after compression is appropriate and the stroke width is sufficient. This input image compression step S210 is an optional step. That is, the image processing apparatus may start processing from the image block size calculating step S220 without compressing the input image.

  In the image block size calculation step S220, the image processing apparatus calculates the image block size according to the stroke width of the target character including the concealed character to be restored (that is, the concealed character to be recognized). To do. Experiments have shown that the effect on the final restoration result differs depending on the image block size. 2A to 2C show examples of the influence on the restoration quality of concealed characters due to the image block size. FIG. 2A is a diagram illustrating an example of a concealed image that is input first, FIG. 2B is a diagram illustrating a restoration result when a 7 × 7 size image block is selected, and FIG. 2C is a diagram illustrating 17 × It is a figure which shows the decompression | restoration result when a 17 size image block is selected. These indicate that when different image block sizes are selected for text images having the same size, the restoration effect of the image processing apparatus is different.

  Preferably, according to a specific example of the disclosed technique, the image block size S is calculated by Expression (5).

  Here, SW indicates the width of the stroke of the target character. When the image block size calculated according to the first embodiment is selected, the image processing apparatus accurately restores the concealed characters.

  After calculating the image block size, the image processing apparatus marks the original area in the target area of the input image including the target character in the original area marking step S230. Specifically, according to the first embodiment, the image processing apparatus first generates a first mask including the pixel in the target character and all adjacent pixels in the image block of the pixel. Next, the image processing apparatus generates a second mask that includes the entire target area other than the foreground pixels and all adjacent pixels in the image block of the foreground pixels. Then, the image processing apparatus extracts an overlapping area between the first mask and the second mask as an original area. For example, in the original area marking step S230, the image processing apparatus marks the original area as shown in FIG. FIG. 3 is a diagram illustrating an example of the marked original area.

  Next, in the reference image block generation step S240, the image processing apparatus generates a reference image block for the pixels in the original area marked in step S230. Here, the image processing apparatus sets each pixel in the original region as the center of one reference image block. In other words, the image processing device generates each reference image block centered on each of the pixels in the original region. The image processing apparatus selects the size of the image block size calculated in step S220 as the image block size of the reference image block.

  If the number of generated reference image blocks is large and there are many similar image blocks, the image processing apparatus selects the reference image blocks generated in step S240 in the reference image block selection step S250. Then, the image processing apparatus merges the reference image blocks having a high degree of similarity into one reference image block. As a result, the image processing apparatus saves storage space and improves the search speed at the time of foreground interpolation described later. Note that the image processing apparatus may execute the boundary extraction step S260 without executing the reference image block selection step S250.

  Next, in the boundary extraction step S260, the image processing apparatus extracts the foreground boundary of the input image to be complemented. Subsequently, in the complementation priority calculation step S270, the image processing apparatus calculates the complementation priority for each pixel at the foreground boundary according to the complexity of the image block centered on the pixel.

  Next, in the foreground interpolation step S280, the image processing apparatus uses the most similar image block in the reference image block to complement the image block centered on the pixel having the highest priority. In other words, the image processing device searches for the most similar image block from the reference image block for the pixel having the highest priority. Then, the image processing apparatus copies the searched image block to an image block centered on the pixel having the highest priority, and updates the foreground boundary after complementation. When searching for a reference image block, the image processing apparatus searches for a reference image block most similar to the image block to be complemented by any known technique, for example, the Euclidean distance.

  After that, in the determination step S290, when all foreground interpolation has not been completed, the image processing apparatus executes foreground boundary extraction processing in the input image to be complemented. In other words, the image processing apparatus determines whether or not the completion of the entire foreground has been completed, and if it is determined that there is still a foreground boundary that has not been completed, the process returns to the boundary extraction step S260, The subsequent processing is executed sequentially.

  On the other hand, if the image processing apparatus determines that the completion of the entire foreground has been completed in determination step S290, the image processing apparatus ends the process.

  According to the first embodiment, the image processing apparatus specifies the order of complementation mainly based on the complexity of the configuration of the image block. This is because the reconstruction error at the time of complementation affects from the previously complemented region to the next region to be complemented. Therefore, the image processing apparatus first reconstructs image blocks that are unlikely to cause reconstruction errors, and then complements the image blocks that need to be reconstructed in the order of complexity. Thereby, the image processing apparatus minimizes the error of the last reconstructed character. An image block having a simpler configuration than an image block having a complicated configuration can easily search for an accurate similar reference image block from the reference image block. 4A to 4D show image blocks having different complexity levels. As a result, the complexity of the image block gradually increases from FIG. 4A to FIG. 4D, and the simpler the image block, the closer to the basic stroke elements. On the other hand, the matching feature is embodied in an image block having a complicated configuration.

  As mentioned above, reconstructing an image block with a simple configuration usually introduces small errors. Therefore, in a specific example according to the disclosed technology, a factor for measuring the complexity of the configuration is used to reflect the result in the complement priority. Here, the complexity C (p) is defined by the types of different edge directions of the image block.

Also, as shown in the following equation (4), the complexity C (p) is calculated based on the number N _gv of different gradient vectors inside the image block. In order to reduce the computational complexity, the complementary priority P (p) is calculated by applying the simple formula shown in the following formula (1) to the first embodiment. The terms R (p) and W (p) in the equation (1) constitute a reliability coefficient and are calculated by the following equations (2) and (3), respectively. Here, the reliability coefficient R (p) is calculated by measuring the ratio of the background pixels in the image block. Further, W (p) is calculated by measuring the ratio of the target character pixel that is the complement target within the image block. Then, by setting the reliability coefficient of the image block in which the target character pixel to be complemented does not exist to 0 inside, the processing speed can be improved without affecting the result of the reconstruction.

Here, calculation expression of the complementary priority indicated by the above formula (1) is only one example, in practical use, it is possible to weight the complexity C _p use different weighting factors depending on the situation The effect suitable for the situation can be obtained.

  According to the hidden character restoration method in the first embodiment, by specifying the complement order according to the complexity of the configuration of the image block, it is possible to avoid a remarkable reconstruction error due to an inaccurate complement order. Further, according to the method for restoring a concealed character in the first embodiment, the size of the image block size is further specified based on the stroke width of the concealed character to be restored. Therefore, an optimal reconstruction result can be acquired by performing complementation for different fonts using a matching image block of an optimal size.

  In the hidden character restoration method according to the first embodiment, the image processing apparatus does not need to execute the input image compression step S210 and the reference image block selection step S250, depending on the system performance and the situation. These steps may be selectively executed.

  Next, processing of the hidden character restoration method according to the second embodiment will be described with reference to FIGS. FIG. 5 is a flowchart illustrating a processing procedure of a hidden character restoration method according to the second embodiment, and FIG. 6 is a diagram illustrating an example of a hidden character to be restored.

  Here, the processing procedure of the concealed character processing shown in FIG. 5 will be described in detail by taking the concealed character to be restored shown in FIG. 6 as an example.

  As shown in FIG. 5, in the concealed character restoration method according to the second embodiment, the image processing apparatus performs a character outline pixel extraction step S610, a character line segment capturing step S620, a character stroke width estimation step S630, a character size. The range estimation step S640 is executed. Subsequently, the image processing apparatus executes a parallel line segment selecting step S650, a collinear line connecting step S660, and a region complementing step S670.

  In character outline pixel extraction step S610, the image processing apparatus extracts outline pixels of the target character. FIG. 7 is a diagram illustrating an example in which the outline pixel of the target character in FIG. 6 is extracted.

  Next, in the character line segment capturing step S620, the image processing apparatus captures the line segment of the target character based on the extracted contour pixels. For example, the image processing apparatus captures a line segment in the horizontal direction θ and the vertical direction (90−θ) of the target character by using the extracted outline pixel according to the result of the Hough conversion. The image processing apparatus estimates the stroke width of the target character based on the captured line segment in the character stroke width estimation step S630. In other words, the image processing apparatus roughly estimates the stroke width of the target character using the average value of the nearest distances between two adjacent line segments. FIG. 8 is a diagram showing an example of a character line segment captured with respect to the contour pixel of the target character in FIG. In addition, the part in the square frame in a figure is a part which is a decompression | restoration object, ie, an interesting line segment.

  Next, in the character size range estimation step S640, the image processing apparatus estimates the range of the target character size. That is, the image processing apparatus estimates the character width and character height on the projection axis of the target character. FIG. 9 is a diagram illustrating an example of estimating the target character size in the second embodiment. In the figure, “w” is the character width projected on the horizontal axis of the character, and “h” is the character height projected on the vertical axis of the character.

  Here, in the hidden character restoration method according to the second embodiment, the image processing apparatus may not execute the character stroke width estimation step S630 and the character size range estimation step S640. For example, after executing the character line segment capturing step S620, the image processing apparatus may execute the next parallel line segment selecting step S650 without performing these two steps.

  In the parallel line segment selection step S650, the image processing apparatus selects a parallel line segment based on the captured line segment. Here, the selected parallel line segment can be estimated as a line segment to be paired with one stroke. In other words, the image processing apparatus selects a parallel line segment that is presumed to be a line segment that forms a pair of one stroke based on the captured line segment.

In the second embodiment, if the image processing apparatus has already acquired the stroke width and the character height, the image processing apparatus selects parallel line segments according to the following three conditions. In the Hough transform region, for line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ indicates the rotation angle of the line segment, and ρ indicates the distance to the zero point of the line segment. .
Condition 1) θ1 = θ2 and a * SW <| ρ1-ρ2 | <b * SW are satisfied, where a * SW and b * SW are threshold values, for example, 1.5 * SW and 0, respectively. .5 * SW.
Condition 2) For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, and for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is It is smaller than the character height.
Condition 3) A pixel between two line segments is located inside the stroke, not the background.
FIG. 10 is a diagram illustrating an example of determining pixel attribution between two line segments in the second embodiment. As shown in FIG. 10, pixel attribution is determined by examining whether the selected pixel is located on a connecting line between each point in two parallel line segments. FIG. 11 is a diagram showing an example of parallel line segments selected from the captured character line segments shown in FIG.

  Subsequently, in the collinear line connecting step S660, the collinear line is connected as a parallel line pair based on the selected parallel line segment.

Similarly, according to a specific embodiment of the disclosed technology, after acquiring the stroke width and the character height, the collinear line segment can be selected according to the following three conditions. In the Hough transform region, for line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ indicates the rotation angle of the line segment, and ρ indicates the distance to the zero point of the line segment. .
Condition 1) Satisfy that θ1 = θ2 and | ρ1−ρ2 | are smaller than a threshold, for example, 1.
Condition 2) For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, and for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is It is smaller than the character height.
Condition 3) A pixel between two line segments is located inside the stroke, not the background.
FIG. 12 is a diagram illustrating an example of a result of executing the collinear line connecting process on the parallel line segments illustrated in FIG. 11.

  Finally, in the region complementation step S670, the region limited by each of the parallel line pairs is complemented using the gray value of the target character. FIG. 13 is a diagram illustrating an example of complementing a region between parallel line segments of a target character in the second embodiment.

  The image processing apparatus restores a partial stroke of the concealed target character after completing the processing of the area complementation step S670. Therefore, the image processing apparatus improves the accuracy of recognition for the convenience of subsequent character recognition processing. As shown in FIG. 14, FIG. 14A is a diagram schematically illustrating an example of a target character before complementing, and FIG. 14B is a diagram schematically illustrating an example of a target character after complementing.

  As described above, the hidden character restoration method according to the first and second embodiments of the disclosed technology has been described in detail with reference to the drawings. The image processing apparatus may use a combination of the two hidden character restoration methods. As a result, the image processing apparatus restores the concealed character more accurately. Then, the image processing apparatus more accurately executes an efficient recognition process for concealed characters described below.

  FIG. 15 is a flowchart of the process procedure of the hidden character restoration method according to the third embodiment. As shown in FIG. 15, in the method for restoring a concealed character according to the third embodiment, the image processing apparatus performs an image text region segmentation step S1610, a target region extraction step S1620, a character stroke restoration step S1630, and a character recognition step. S1640 is executed.

  In the image text area division step S1610, the image processing apparatus performs detection, division, and quantization on the input image, and divides the text area in the input image into target characters, backgrounds, and foregrounds to be recognized. In this step, the image processing apparatus first detects a text area in the input image. The image processing device detects a text region based on different features of the text character, such as vertical edge gray value, texture or edge direction change. The research on the detection of the text area has already been deeply advanced, and since this is not the gist of the disclosed technique, a detailed description is omitted here. Next, the image processing apparatus generates a desired image by executing necessary specific image adjustment processing. For example, the image processing apparatus executes an image oblique correction process, a noise removal process, an image enlargement / reduction process, and the like. After executing this step, the image processing apparatus divides the text area into target characters, foreground and background, and assigns different gray values. FIG. 16A is a diagram illustrating an example of an input image including a concealed character that is a recognition target in the third embodiment. FIG. 16B is a diagram illustrating an example of a result obtained by performing an image text region dividing step on the input image illustrated in FIG. 16A.

  In target area extraction step S1620, the image processing apparatus extracts a target area including a target character to be recognized from the text area.

FIG. 17 is a diagram schematically illustrating an example of a target character extraction method according to the third embodiment. In the third embodiment, the image processing apparatus extracts a target area including a target character to be recognized from a text area according to the following steps shown in FIG. That is, the image processing apparatus extracts the contour pixel of the target character, and performs Hough conversion on the contour pixel in the horizontal direction θ and the vertical direction (90−θ). Here, θ represents the rotation angle of the target character. Find the minimum and maximum distance values ρ _h1 , ρ _h2 , ρ _v1 and ρ _v2 during the Hough transform along the horizontal direction θ and the vertical direction (90−θ), and four lines (ρ _h1 , 90−θ), Four angles are calculated using (ρ _h2 , 90−θ), (ρ _v1 , θ), (ρ _v2 , θ). Further, a region covering the calculated four corners is cut out as a target region. FIG. 16C is a diagram illustrating an example of a result of executing a process of extracting a target area with respect to the image text area illustrated in FIG. 16B.

  Here, the target area extraction step S1620 is not essential for the disclosed technology. That is, the image processing apparatus may execute the character and stroke restoration step S1630 without executing the step after the image text region dividing step S1610 ends.

  In the character and stroke restoration step S1630, the image processing apparatus restores the concealed character that is the recognition target in the text area by the method described in the first and second embodiments. FIG. 16D is a diagram illustrating an example of a result of executing the process of restoring the concealed character with respect to the target area illustrated in FIG. 16C.

  Finally, in the character recognition step S1640, the image processing apparatus recognizes the character restored in the character stroke restoration step S1630. FIG. 16E is a diagram illustrating an example of a result of executing the process of recognizing a hidden character after restoration illustrated in FIG. 16D.

  The process of the hidden character restoration method according to the first to third embodiments has been described above with reference to FIGS. The operation of the image processing apparatus according to the disclosed technology will be described below with reference to FIGS.

  FIG. 18 is a block diagram illustrating a configuration of an image processing apparatus 1900 according to the fourth embodiment. As illustrated in FIG. 18, the image processing apparatus 1900 according to the fourth embodiment includes a compression unit 1910, an image block size calculation unit 1920, an original region marking unit 1930, a reference image block generation unit 1940, a selection unit 1950, and a boundary extraction unit 1960. , A complementary priority calculating unit 1970, a foreground complementing unit 1980, and a determining unit 1990. The compression unit 1910 is arranged to compress the input image. The image block size calculation unit 1920 is arranged to calculate the image block size according to the stroke width of the target character including the hidden character to be restored. The original area marking unit 1930 is arranged to mark the original area in the target area in the input image including the target character. The reference image block generation unit 1940 is arranged to generate a reference image block for the pixels in the original region. Here, each pixel in the original area is set as the center of one reference image block. The selection unit 1950 is arranged to select the selection image block. The boundary extraction unit 1960 is arranged so as to extract the foreground boundary in the input image to be complemented. The complementary priority calculating unit 1970 is arranged so as to calculate the complementary priority for each pixel at the foreground boundary according to the complexity of the image block centered on the pixel. The foreground complementing unit 1980 is arranged to complement the image block centered on the pixel having the highest priority using the most similar image block in the reference image block. The determination unit 1990 is arranged so as to repeatedly execute the processes of the boundary extraction unit 1960, the complementation priority calculation unit 1970, and the foreground complementation unit 1980 when the foreground complementation is not completed.

  According to the image processing apparatus 1900 according to the fourth embodiment, the image block size calculation unit 1920 calculates the image block size by the following equation (5). Here, SW indicates the width of the stroke of the target character.

  According to the image processing apparatus 1900 according to the fourth embodiment, the original area marking unit 1930 generates a first mask that includes pixels in the target character and all adjacent pixels in the image block. In addition, the original area marking unit 1930 generates a second mask that includes the entire target area except foreground pixels and all adjacent pixels in the image block. Then, the original area marking unit 1930 extracts an overlapping area between the first mask and the second mask as an original area.

  Similarly, the complementary priority order calculating unit 1970 calculates the complementary priority order P (p) by the following formula.

  The processing in the compression unit 1910 included in the image processing apparatus 1900 according to the fourth embodiment is the same as the input image compression step S210 included in the hidden character restoration method according to the first embodiment. The processing in the image block size calculation unit 1920 is the same as that in the image block size calculation step S220, and the processing in the original region marking unit 1930 is the same as in the original region marking step S230. The process in the reference image block generation unit 1940 is the same as the process in the reference image block generation step S240, and the process in the selection unit 1950 is the same as the process in the reference image block selection step S250. Further, the process in the boundary extraction unit 1960 is the same as the process in the boundary extraction step S260, and the process in the complement priority calculation unit 1970 is the same as the process in the complement priority calculation step S270. Further, the processing in the foreground complementing unit 1980 is the same as the processing in the foreground complementing step S280. Therefore, for the sake of brevity, these detailed descriptions are omitted.

  Note that the image processing apparatus 1900 according to the fourth embodiment may be configured to be selectively installed without the compression unit 1910 and the selection unit 1950 in practical use.

  FIG. 19 is a block diagram illustrating the configuration of the image processing apparatus 2000 according to the fifth embodiment. As illustrated in FIG. 19, the image processing apparatus 2000 according to the fifth embodiment includes an outline pixel extraction unit 2010, a line segment capture unit 2020, a stroke width estimation unit 2030, a size range estimation unit 2040, a parallel line segment selection unit 2050, A collinear line connecting unit 2060 and a region complementing unit 2070 are provided. The contour pixel extraction unit 2010 is arranged to extract the contour pixels of the target character. The line segment capturing unit 2020 is arranged to capture the line segment of the target character based on the extracted contour pixels. The stroke width estimation unit 2030 is arranged to estimate the stroke width of the target character based on the captured line segment. The size range estimation unit 2040 is arranged to estimate the character width and character height on the projection axis of the target character. The parallel line segment selection unit 2050 is arranged to select a parallel line segment that is assumed to be a pair of strokes based on the captured line segment. The collinear line connecting unit 2060 is arranged to connect the collinear line as a parallel line pair based on the selected parallel line. The area complementing unit 2070 is arranged to complement the area limited by each parallel line pair using the gray value of the target character.

Similarly, the selected parallel line segment must satisfy the following condition in the Hough transform region.
Condition 1) Line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW satisfy θ1 = θ2 and a * SW <| ρ1-ρ2 | <b * SW. , Θ represents the rotation angle of the line segment, ρ represents the distance to the zero point of the line segment, and a * SW and b * SW are threshold values.
Condition 2) For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, while for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is It is smaller than the character height.
Condition 3) A pixel between two line segments is located inside the stroke, not the background.

The collinear segment must satisfy the following conditions.
Condition 1) For line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ1 = θ2 and | ρ1−ρ2 | are satisfied to be smaller than the threshold value.
Condition 2) For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, and for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is It is smaller than the character height.
Condition 3) A pixel between two line segments is located inside the stroke, not the background.

  The processing in the contour pixel extraction unit 2010 included in the image processing apparatus 2000 according to the fifth embodiment is the same as the processing in the character contour pixel extraction step S610 in the hidden character restoration method according to the second embodiment. The processing in the line segment capturing unit 2020 is the same as the processing in the character line segment capturing step S620. The process in the stroke width estimation unit 2030 is the same as the process in the character stroke width estimation step S630. The processing in the size range estimation unit 2040 is the same as the processing in the character size range estimation step S640. The process in the parallel line segment selection unit 2050 is the same as the process in the parallel line segment selection step S650. The process in the collinear line connecting unit 2060 is the same as the process in the collinear line connecting step S660. The process in the area complementing unit 2070 is the same as the process in the area complementing step S670. Therefore, for the sake of brevity, detailed description of these components is omitted here.

  Similarly, the image processing apparatus 2000 according to the fifth embodiment does not need to install the stroke width estimation unit 2030 and the size range estimation unit 2040, and selectively installs them according to the system design requirements and usage conditions. It is something that can be done.

  Similarly to the hidden character restoration method, the image processing apparatuses according to the fifth and sixth embodiments may be used in combination. As a result, the image processing apparatus exhibits a better restoration effect, and further accurately executes the concealed character efficient recognition process described below.

  FIG. 20 is a block diagram illustrating the configuration of the image processing apparatus 2100 according to the sixth embodiment. As illustrated in FIG. 20, the image processing apparatus 2100 according to the sixth embodiment includes a dividing unit 2110, a target area extracting unit 2120, a hidden character restoration device 2130, and a character recognition unit 2140. The dividing unit 2110 is arranged to divide a text area in the input image into a target character, a background, and a foreground that are recognition targets. The target area extraction unit 2120 is arranged to extract a target area including a target character that is a recognition target from the text area. The concealed character restoration device 2130 is arranged to restore the concealed character, and is, for example, the image processing device 1900 according to the fourth embodiment or the image processing device 2000 according to the fifth embodiment. Character recognition unit 2140 is arranged to recognize the restored character.

Similarly, in a specific embodiment, the target area extraction unit 2120 extracts the contour pixel of the target character, and performs Hough conversion on the contour pixel in the horizontal direction θ and the vertical direction (90−θ). Here, θ represents the rotation angle of the target character. Find the minimum and maximum distance values ρ _h1 , ρ _h2 , ρ _v1 and ρ _v2 during the Hough transform along the horizontal direction θ and the vertical direction (90−θ), and four lines (ρ _h1 , 90−θ), Four angles are calculated using (ρ _h2 , 90−θ), (ρ _v1 , θ), (ρ _v2 , θ). Further, a region covering the calculated four corners is cut out as a target region.

  Processing in the dividing unit 2110 included in the image processing apparatus 2100 according to the sixth embodiment is the same as the processing in the image text region dividing step S1610 in the hidden character restoration method described with reference to FIGS. . Further, the process in the target area extraction unit 2120 is the same as the process in the target area extraction step S1620. The processing in the concealed character restoration device 2130 is the same as the processing in the character and stroke restoration step S1630. That is, the hidden character restoration apparatus 2130 executes the restoration method for the hidden characters according to the first or second embodiment. The processing in character recognition unit 2140 is the same as the processing in character recognition step S1640. Therefore, for the sake of brevity, detailed description of these components is omitted here.

  Here, the image processing apparatus 2100 according to the sixth embodiment may not be provided with the target area extraction unit 2120, and may be selectively installed according to the system design requirements and usage conditions.

  Each component module and each component in the above-described image processing apparatus may be arranged by software, firmware, hardware, or a combination thereof. In addition, the specific method or system which can be used for arrangement | positioning is a well-known technique, and is not demonstrated here. When realized by software or firmware, a program that configures the software is installed from a storage medium or a network to a computer having a dedicated hardware configuration (for example, the information processing apparatus 1100 illustrated in FIG. 21). When various programs are installed, the information processing apparatus 1100 executes various functions and the like.

  FIG. 21 is a block diagram illustrating a configuration of an information processing apparatus 1100 that executes a method for restoring a concealed character according to the disclosed technology. In FIG. 21, the central processing unit (CPU) 1101 executes various processes according to a program stored in a read-only memory (ROM) 1102 or a program uploaded from the storage unit 1108 to the random access memory (RAM) 1103. To do. The RAM 1103 may store data necessary when the CPU 1101 executes various processes. The CPU 1101, ROM 1102, and RAM 1103 are connected to each other via a bus 1104. Similarly, the input / output interface 1105 is connected to the bus 1104.

  The input unit 1106, the output unit 1107, the storage unit 1108, and the communication unit 1109 are connected to the input / output interface 1105. Here, the input unit 1106 includes a keyboard, a mouse, and the like. The output unit 1107 includes a display such as a cathode ray tube (CRT) or a liquid crystal display (LCD) and a speaker. The storage unit 1108 includes a hard disk and the like. The communication unit 1109 includes a network interface card such as a LAN card, a modem, and the like. The communication unit 1109 executes communication processing via a network such as the Internet, for example. The input / output interface 1105 may also be connected to the drive 1110. The drive 1110 may be attached with a removable medium 1111 such as a magnetic disk, an optical disk, a magneto-optical disk, and a semiconductor memory. Then, a computer program read from the removable medium 1111 may be installed in the storage unit 1108.

  When a series of processing is executed by software, a program constituting the software is installed from a network such as the Internet or a storage medium such as a removable medium 1111.

  Such a storage medium is such that the program is stored in the medium as shown in FIG. 21, and is limited to a removable medium 1111 that provides the program to the user separately from the device. is not. For example, the storage medium may be a ROM 1102, a hard disk included in the storage unit 1108, or the like. Programs are stored in these storage media, and are provided to the user together with a device including them. The removable medium 1111 is, for example, a magnetic disk such as a floppy (registered trademark), an optical disk such as a compact disk read-only memory (CD-ROM) or a digital versatile disk (DVD), a mini disk (registered trademark), or the like. Magneto-optical disk and semiconductor memory.

  The disclosed technology further provides a program product in which an instruction code readable by a device is stored. When the instruction code is read and executed by the device, the method of the embodiment according to the disclosed technology can be executed.

  The disclosed technology also includes a storage medium on which a program product in which an instruction code readable by a device is stored. Note that the storage medium includes, but is not limited to, a floppy (registered trademark) disk, an optical disk, a magneto-optical disk, a memory card, a memory stick, and the like.

  In the description of the above examples, an example of the embodiment has been described. Also, the features shown in the description of the embodiments may be used in one or more other embodiments in the same or similar form, combined with features in other embodiments, or features in other embodiments. Or may be replaced.

  Here, when the term “including / having” is used in the text, it means the presence of a feature, element, step or component, and one or more other features, elements, steps or components. It does not exclude the presence or addition.

  The method of the disclosed technology is not limited to being executed in the time order described in the specification, and may be executed in parallel or independently according to other time orders. Therefore, the execution order of the method described in this specification does not limit the technical scope of the disclosed technology.

  Although the disclosed technology has been described above by describing the embodiments of the disclosed technology, all of the embodiments are merely examples and are not limiting. A person skilled in the art may design various modifications and improvements to the disclosed technology or similar devices based on the appended claims and the technical idea of the disclosed technology. These modifications, improvements or similar devices are naturally considered to be within the protection scope of the disclosed technology.

  The following supplementary notes are further disclosed with respect to the embodiments including the above examples.

(Appendix 1) An image processing apparatus
Calculating an image block size according to a stroke width of a target character including a hidden character to be restored; and
Marking the original region in the target region of the input image containing the target character;
Generating each reference image block centered on each of the pixels in the original region;
Extracting the foreground boundary of the input image to be complemented;
For each pixel at the foreground boundary, calculating a complementation priority according to the complexity of the image block centered on the pixel;
Complementing the image block centered on the pixel with the highest priority using the most similar image block in the reference image block;
A concealed character restoration method comprising: performing foreground boundary extraction processing in the input image to be complemented when all of the foreground completion has not been completed.

(Appendix 2) The image processing apparatus is
A confidence coefficient calculated by measuring the percentage of background pixels in the image block, a confidence coefficient calculated by measuring the percentage of target character pixels to be complemented in the image block, and different gradient vectors in the image block The method for restoring a concealed character according to supplementary note 1, wherein the complementary priority is calculated from the complexity based on the number of characters.

(Supplementary Note 3) The image processing apparatus is
The hidden character restoration method according to attachment 1, wherein the image block size is calculated from a stroke width of a target character.

(Supplementary Note 4) The step of marking the original area in the target area of the input image including the target character includes:
Generating a first mask including pixels in the target character and all adjacent pixels in an image block of the pixels;
Generating a second mask that includes the entire target region other than the foreground pixels and all adjacent pixels in the image block of the foreground pixels;
The method for restoring a concealed character according to appendix 1, further comprising: extracting an overlapping area between the first mask and the second mask as an original area.

(Supplementary Note 5) The image processing apparatus is
The hidden character restoration method according to claim 1, further comprising the step of compressing the input image and / or selecting the reference image block.

(Appendix 6) The image processing apparatus
Extracting contour pixels of the target character;
Capturing a line segment of a target character based on the extracted contour pixels;
Selecting a parallel line segment that is presumed to be a line segment of a stroke based on the captured line segment; and
Connecting collinear segments as parallel line pairs based on the selected parallel segments;
The method for restoring a concealed character according to any one of supplementary notes 1 to 5, further comprising: complementing a region limited by each parallel line pair using the gray value of the target character.

(Supplementary note 7) The image processing apparatus is
Estimating a stroke width of the target character based on the captured line segment;
Further including estimating a character width and a character height on the projection axis of the target character,
In the Hough transform region, the selected parallel line segment is
For line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ1 = θ2 and a * SW <| ρ1-ρ2 | <b * SW, where θ is the rotation of the line segment Indicates the angle, ρ indicates the distance to the zero point of the line segment, a * SW and b * SW are threshold values,
For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, while for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is larger than the character height. Small,
Satisfy that the pixel between the two line segments is located inside the stroke, not the background,
The collinear line is
For line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ1 = θ2 and | ρ1-ρ2 |
For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, while for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is larger than the character height. Small,
The method for restoring a concealed character according to appendix 6, wherein the pixel between two line segments is located inside the stroke, not the background.

(Appendix 8) The image processing apparatus is
Dividing the text region in the input image into target characters, background and foreground to be recognized;
The method for reconstructing a concealed character according to any one of appendix 1 to appendix 7, further comprising the step of recognizing the restored character.

(Supplementary note 9) The image processing apparatus
9. The method for restoring a concealed character according to appendix 8, further comprising: extracting a target region including a target character that is a recognition target from the text region.

(Additional remark 10) The step which extracts the target area | region containing the target character which is recognition object from the said text area | region is the following.
Extracting contour pixels of the target character;
θ indicates the rotation angle of the target character, and Hough conversion is performed on the contour pixel in the horizontal direction θ and the vertical direction (90−θ);
Obtaining minimum and maximum distance values ρ _h1 , ρ _h2 , ρ _v1 and ρ _v2 during Hough transform along the horizontal direction θ and the vertical direction (90−θ);
Calculating four angles using four lines (ρ _h1 , 90−θ), (ρ _h2 , 90−θ), (ρ _v1 , θ), (ρ _v2 , θ);
The method for restoring a concealed character according to appendix 9, comprising: cutting out an area covering the four corners as a target area.

(Supplementary Note 11) An image block size calculating unit arranged to calculate an image block size based on a stroke width of a target character including a hidden character to be restored;
An original area marking portion arranged to mark the original area in the target area of the input image including the target character;
A reference image block generator arranged to generate each reference image block centered on each of the pixels in the original region;
A boundary extraction unit arranged to extract the boundary of the foreground of the input image to be complemented;
A complementary priority calculating unit arranged to calculate a complementary priority according to the complexity of the image block centered on the pixel for each pixel at the boundary of the foreground;
A foreground complementing unit arranged to complement the image block centered on the pixel having the highest priority using the most similar image block in the reference image block;
An image processing apparatus comprising: a determination unit arranged to execute the processing of the boundary extraction unit when all of the foreground complement has not been completed.

(Additional remark 12) The said complementation priority calculation part measures and calculates the reliability coefficient calculated by measuring the ratio for which the background pixel in an image block occupies, and the ratio for which the target character pixel which is the complement object in an image block occupies The image processing apparatus according to appendix 11, wherein the complementary priority is calculated from the reliability coefficient obtained and the complexity based on the number of different gradient vectors inside the image block.

(Additional remark 13) The said image block size calculation part is an image processing apparatus of Additional remark 11 which calculates the said image block size from the width | variety of the stroke of a target character.

(Supplementary Note 14)
Generating a first mask including a pixel in the target character and all adjacent pixels in an image block of the pixel;
Generating a second mask that includes the entire target region except foreground pixels and all adjacent pixels in the image block of the foreground pixels;
Extracting an overlapping region of the first mask and the second mask as an original region;
The image processing apparatus according to appendix 11.

(Supplementary note 15) a compression unit arranged to compress an input image;
The image processing apparatus according to appendix 11, further comprising: a sorting unit arranged to sort the reference image block.

(Supplementary Note 16) An outline pixel extraction unit arranged to extract outline pixels of the target character;
A line segment capture unit arranged to capture a line segment of a target character based on the extracted contour pixels;
A parallel line selection unit arranged to select a parallel line segment that is presumed to be a line segment to be paired with one stroke based on the captured line segment;
A collinear line connecting portion arranged to connect collinear line segments as parallel line pairs based on the selected parallel line segments;
The image processing according to any one of Supplementary Note 11 to Supplementary Note 15, further comprising: a region complementing unit arranged to supplement a region limited by each of the parallel line pairs using the gray value of the target character. apparatus.

(Supplementary Note 17) A stroke width estimation unit arranged to estimate a stroke width of a target character based on the captured line segment;
A size range estimator arranged to estimate a character width and a character height on the projection axis of the target character,
In the Hough transform region, the selected parallel line segment is
For line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ1 = θ2 and a * SW <| ρ1-ρ2 | <b * SW, where θ is the rotation of the line segment Indicates the angle, ρ indicates the distance to the zero point of the line segment, a * SW and b * SW are threshold values,
For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, while for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is larger than the character height. Small,
Satisfy that the pixel between the two line segments is located inside the stroke, not the background,
The collinear line is
For line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ1 = θ2 and | ρ1-ρ2 |
For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, while for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is larger than the character height. Small,
The image processing apparatus according to appendix 16, wherein the pixel between the two line segments satisfies that the pixel is located inside the stroke, not the background.

(Supplementary Note 18) A dividing unit arranged to divide a text area in an input image into a target character to be recognized, a background, and a foreground;
The image processing apparatus according to any one of supplementary notes 11 to 17, further comprising: a character recognition unit arranged to recognize the restored character.

(Supplementary note 19) The concealed image processing device according to supplementary note 18, further comprising a target area extracting unit arranged to extract a target area including a target character to be recognized from the text area.

(Supplementary note 20) The target area extraction unit
Extracting outline pixels of the target character;
θ represents the rotation angle of the target character, and Hough transformation is performed on the contour pixel in the horizontal direction θ and the vertical direction (90−θ).
Find the minimum and maximum distance values ρ _h1 , ρ _h2 , ρ _v1 and ρ _v2 during the Hough transform along the horizontal direction θ and the vertical direction (90−θ),
Use the four lines (ρ _h1 , 90−θ), (ρ _h2 , 90−θ), (ρ _v1 , θ), (ρ _v2 , θ) to calculate the four angles,
Cutting out the area covering the four corners as a target area,
The concealed image processing apparatus according to appendix 19.

1100 Information processing apparatus 1101 Central processing prime minister (CPU)
1102 Read only memory (ROM)
1103 Random Access Memory (RAM)
DESCRIPTION OF SYMBOLS 1104 Bus 1105 Input / output interface 1106 Input part 1107 Output part 1108 Storage part 1109 Communication part 1110 Drive 1111 Removable medium 1900 Image processing apparatus 1910 Compression part 1920 Image block size calculation part 1930 Original area mark part 1940 Reference image block generation part 1950 Selection unit 1960 Boundary extraction unit 1970 Complementary priority calculation unit 1980 Foreground complementation unit 1990 Determination unit 2000 Image processing device 2010 Outline pixel extraction unit 2020 Line segment capture unit 2030 Stroke width estimation unit 2040 Size range estimation unit 2050 Parallel line segment selection unit 2060 Collinear line connecting unit 2070 Region complementing unit 2100 Image processing device 2110 Dividing unit 2120 Target region extracting unit 2130 Concealed character restoration device 2140 Character recognition unit

Claims (9)

The image processing device
Calculating an image block size according to a stroke width of a target character including a hidden character to be restored; and
Marking the original region in the target region of the input image containing the target character;
Generating each reference image block centered on each of the pixels in the original region;
Extracting the foreground boundary of the input image to be complemented;
For each pixel at the foreground boundary, calculating a complementation priority according to the complexity of the image block centered on the pixel;
Complementing the image block centered on the pixel with the highest priority using the most similar image block in the reference image block;
A concealed character restoration method comprising: performing foreground boundary extraction processing in the input image to be complemented when all of the foreground completion has not been completed. The image processing device
A confidence coefficient calculated by measuring the percentage of background pixels in the image block, a confidence coefficient calculated by measuring the percentage of target character pixels to be complemented in the image block, and different gradient vectors in the image block The concealed character restoration method according to claim 1, wherein the complement priority is calculated from a complexity based on the number of characters. The image processing device
The hidden character restoration method according to claim 1, wherein the image block size is calculated from a stroke width of a target character. Marking the original region in the target region of the input image containing the target character,
Generating a first mask including pixels in the target character and all adjacent pixels in an image block of the pixels;
Generating a second mask that includes the entire target region other than the foreground pixels and all adjacent pixels in the image block of the foreground pixels;
The method for restoring a concealed character according to claim 1, further comprising: extracting an overlapping area of the first mask and the second mask as an original area. The image processing device
The method of claim 1, further comprising: compressing an input image and / or selecting a reference image block. The image processing device
Extracting contour pixels of the target character;
Capturing a line segment of a target character based on the extracted contour pixels;
Selecting a parallel line segment that is presumed to be a line segment of a stroke based on the captured line segment; and
Connecting collinear segments as parallel line pairs based on the selected parallel segments;
The method for restoring a concealed character according to any one of claims 1 to 5, further comprising: complementing a region defined by each parallel line pair using the gray value of the target character. . The image processing device
Estimating a stroke width of the target character based on the captured line segment;
Further including estimating a character width and a character height on the projection axis of the target character,
In the Hough transform region, the selected parallel line segment is
For line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ1 = θ2 and a * SW <| ρ1-ρ2 | <b * SW, where θ is the rotation of the line segment Indicates the angle, ρ indicates the distance to the zero point of the line segment, a * SW and b * SW are threshold values,
For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, while for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is larger than the character height. Small,
Satisfy that the pixel between the two line segments is located inside the stroke, not the background,
The collinear line is
For line segment 1 (θ1, ρ1), line segment 2 (θ2, ρ2) and stroke width SW, θ1 = θ2 and | ρ1-ρ2 |
For the horizontal line segment, the distance between the projected lines along the horizontal direction θ is smaller than the character width, while for the vertical line segment, the distance between the projected lines along the vertical direction (90−θ) is larger than the character height. Small,
The method for restoring a concealed character according to claim 6, wherein the pixel between the two line segments is located inside the stroke, not the background. The image processing device
Dividing the text region in the input image into target characters, background and foreground to be recognized;
The method for recognizing a concealed character according to claim 1, further comprising: recognizing the reconstructed character. An image block size calculator arranged to calculate the image block size based on the stroke width of the target character including the hidden character to be restored;
An original area marking portion arranged to mark the original area in the target area of the input image including the target character;
A reference image block generator arranged to generate each reference image block centered on each of the pixels in the original region;
A boundary extraction unit arranged to extract the boundary of the foreground of the input image to be complemented;
A complementary priority calculating unit arranged to calculate a complementary priority according to the complexity of the image block centered on the pixel for each pixel at the boundary of the foreground;
A foreground complementing unit arranged to complement the image block centered on the pixel having the highest priority using the most similar image block in the reference image block;
An image processing apparatus comprising: a determination unit arranged to execute the processing of the boundary extraction unit when all of the foreground complement has not been completed.

Images