JP7508212B2 - Image processing device, image processing method, and program - Google Patents

Description

The present invention relates to a technology for identifying inverted areas from a document image.

When performing OCR processing on document images obtained using a scanner or camera, a technique is known that identifies areas where the background and text are inverted (inverted areas), such as white-out characters, and eliminates the inverted state, thereby improving OCR accuracy. Patent Document 1 discloses a method of performing edge extraction on an input image obtained by scanning a document, etc., identifying inverted areas by analyzing the edge extraction results, and eliminating the inverted state in image parts where the inverted areas exist.

JP 2010-186246 A

However, with the technology of Patent Document 1, the processing load of edge extraction increases depending on the content of the document image. For example, in the case of a document image with a large number of characters, it is necessary to perform edge extraction for all characters and to perform inversion area determination processing for each extracted edge information, which results in excessive processing costs.

The image processing device according to the present disclosure comprises a processing means for performing morphological processing, including at least a contraction process for erasing normal characters, on a document image including an inverted region including characters in which the background and foreground are inverted, and normal characters in an area other than the inverted region, to generate an enhanced image in which connected pixel blocks that may constitute the inverted region remain ; an analysis means for analyzing a partial image of the document image corresponding to a candidate region of the inverted region identified based on information of the connected pixel blocks remaining in the enhanced image; and an identification means for identifying the inverted region in the document image including the characters in which the background and foreground are inverted, based on the analysis result by the analysis means, wherein, when there are multiple candidate regions, the identification means integrates the candidate regions based on their positional relationship, and identifies the inverted region based on the analysis result of the partial image of the document image corresponding to the integrated candidate region .

The technology disclosed herein makes it possible to accurately identify inverted regions while keeping processing costs down, even when a document image contains a large amount of content.

A diagram showing the configuration of an information processing system. FIG. 1 is a functional block diagram showing a software configuration related to identification of a reversed region included in a document image according to a first embodiment; 1 is a flowchart showing an outline of a process in an image processing apparatus according to a first embodiment; FIG. 1A is a diagram showing an example of a binary image; FIG. 1B is a diagram showing an example of a result of non-inverting an inverted region in the binary image; FIG. 1C is a diagram showing an example of an enhanced image; and FIG. 1D is a diagram showing an example of a case where the outer frame of the inverted region is left. 1 is a flowchart showing details of a process for determining an inverted region according to the first embodiment; FIG. 4A is a diagram showing an example of a binary image, FIG. 4B is a diagram showing an example of an enhanced image, and FIG. 4C is a diagram showing an example of a result of non-inverting an inverted region in the binary image. 10 is a flowchart showing details of an inversion region determination process according to a second embodiment. 1A is a diagram showing an example of a binary image; FIG. 1B is a diagram showing an example of an enhanced image; FIG. 1C is a diagram showing an example of a provisional integrated region; and FIG. 1D is a diagram showing an example of a result of non-inverting an inverted region in the binary image. 11 is a flowchart showing details of a candidate region re-evaluation process according to a modification of the second embodiment. FIG. 2A is a diagram showing an example of a partial binary image, and FIG. 2B is a schematic diagram of a reduced binary image. The result of projecting the partial binary image in the x-axis direction

Below, the embodiments for carrying out the present invention will be explained with reference to the drawings. Note that the following embodiments do not limit the invention as claimed, and not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[Embodiment 1]
<Hardware Configuration>
Fig. 1 is a diagram showing the configuration of an information processing system according to this embodiment. The information processing system is made up of a scanner device 100 and an image processing device 110. Below, the hardware configurations of the scanner device 110 and the image processing device 110 will be briefly described with reference to Fig. 1.

The scanner device 100 has a control unit 101, an image reading unit 102, and a communication unit 103. The control unit 101 is composed of a CPU, RAM, ROM, etc., and the CPU loads a specific program stored in the ROM into the RAM and executes it to control the scanner device 100 as a whole. The image reading unit 101 optically reads a document placed on a platen (not shown) and generates a scanned image. The communication unit 103 is a communication interface that exchanges image data with an external device (here, the image processing device 110) via a network.

The image processing device 110 has a control unit 111, a mass storage unit 112, a display unit 113, an input unit 114, and a communication unit 115. The control unit 111 is composed of a CPU, a RAM, a ROM, etc., and the CPU loads a predetermined program stored in the ROM into the RAM and executes it, thereby controlling the image processing device 110 in an overall manner. The mass storage unit 112 is, for example, a HDD or SSD, and stores various data and various programs. The display unit 113 is, for example, a liquid crystal display, and displays various information to the user. The input unit 114 is, for example, a keyboard or a mouse, and accepts various operations by the user. The display unit 113 and the input unit 114 may be integrated, such as a touch panel. The communication unit 115 is a communication interface that exchanges image data with an external device (here, the scanner device 100) via a network.

In this embodiment, the image reading unit 101 of the scanner device 100 scans a paper document such as a form to generate a scanned image. The generated scanned image data is transmitted to the image processing device 110 by the communication unit 103. In the image processing device 110, the communication unit 115 receives the scanned image data and stores it in the mass storage unit 112. Note that the configuration of the information processing system shown in FIG. 1 is one example, and for example, the scanner device 100 and the image processing device 110 may be integrated, or the image processing device 110 may be a cloud-type server computer connected via the Internet.

Figure 2 is a functional block diagram showing the software configuration related to the identification of inverted areas contained in a document image of the image processing device 110 according to this embodiment. Note that in this specification, the term "inverted area" is used to mean an area in a document image where the foreground, such as text, is inverted, with the foreground portion, such as a colored background, removed.

The image processing device 110 has a binarization unit 200 and an inversion region processing unit 210, which further includes a morphology processing unit 211, a connected pixel block extraction unit 212, an inversion region determination unit 213, and a pixel inversion unit 214. The functions of these units are realized by the CPU executing a program stored in the ROM in the control unit 111. Note that the image processing device 110 may further have functions other than those shown in FIG. 2, such as an OCR function.

Below, the function of each unit will be explained with reference to the flowchart shown in FIG. 3. Prior to the series of processes in the image processing device 110 shown in the flowchart in FIG. 3, the user optically reads a paper document using the scanner device 100 to generate a scanned image. Then, when the user sends the scanned image data to the image processing device 110 and issues a predetermined start instruction in the image processing device 110, the series of processes shown in the flowchart in FIG. 3 begins. Here, the explanation will be given assuming that a scanned image of a receipt has been input as the image to be processed. Note that in the following explanation, the symbol "S" means a step.

In S301, the binarization unit 200 performs binarization processing on the input scanned image to generate a binary image to be processed by the inverted area processing unit 210. The binarization processing is a process of quantizing image data consisting of multiple gradation values, for example, 256 gradations (8 bits), into binary image data of "0" and "1". Here, a known binarization method such as discriminant analysis can be used. FIG. 4(a) shows the result of the binarization processing on the scanned image of a receipt. The binary image data obtained by the binarization processing is stored in the RAM in the control unit 111. Note that the binarization processing is not essential, and it is sufficient to obtain an image in which the foreground and background parts of the input scanned image are distinguished.

Steps S302 to S305 correspond to the process of identifying the inverted area in the binary image generated in S301 and non-inverting the identified inverted area. This series of processes results in a binary image in which the character string "Amount" expressed in white characters within inverted area 400 as shown in FIG. 4(b) is converted to normal characters. This will be explained in detail below.

First, in S302, the morphology processing unit 211 performs morphology processing on the binary image generated in S301 to delete pixels in the image area that is assumed not to be an inverted area from the binary image. Specifically, a contraction process is first performed on the binary image, and then an expansion process is performed on the image after the contraction process, and this is repeated a predetermined number of times. The contraction process performed first has the effect of eliminating thin lines made up of black pixels in the binary image. Therefore, black pixels as the foreground that make up normal characters, lines, etc. in the image to be processed will disappear by the reduction process. On the other hand, since the inverted area is composed of lines thicker than characters and lines, some of the black pixels that make up the background in the inverted character area will also disappear by the contraction process, but they are larger in area than the black pixels that are the foreground of normal characters, so they are more likely to remain. In other words, by performing the contraction process, only the black pixels that make up normal characters, lines, etc. are deleted, and the black pixels that make up the inverted area remain. Then, by performing the same scale of expansion process on the image after the contraction process, the black pixels that were not lost by the contraction process are restored to their state before the contraction. The purpose of the morphology process is to emphasize image regions that are likely to be inverted regions in the image to be processed. Therefore, although contraction processing is necessary to remove normal characters and lines, it is not necessary to perform expansion processing the same number of times. By performing the above-mentioned morphology process on a binary image, a binary image is obtained in which only black pixel regions that may be inverted regions remain. Hereinafter, an image obtained by morphology processing, in which image regions that are candidates for inverted regions are emphasized, will be referred to as an "enhanced image". Figure 4(c) shows an enhanced image as a result of morphology processing on the binary image shown in Figure 4(a). It can be seen that the black pixels that make up the normal characters and lines have disappeared, and only the black pixels in the inverted region 400 including the white characters for "amount" and the black pixels that make up the horizontally long rectangle at the bottom of the receipt remain.

In S303, the connected pixel block extraction unit 212 extracts connected pixel blocks (hereinafter, referred to as "CC") from the enhanced image obtained in S302. CC stands for "Connected Component," and here means a block of black pixels connected in the enhanced image. The black pixels constituting the same black pixel block among the black pixel blocks extracted from the enhanced image are given the same label, and are identified as a region that is a candidate for the inverted region (hereinafter, referred to as a "candidate region"). From the enhanced image shown in FIG. 4(c) described above, a total of two black pixel blocks are extracted: a black pixel block 401 corresponding to the inverted region 400 and a black pixel block 402 corresponding to the strip-shaped object at the bottom. Different labels are given to each of the black pixel blocks, and they are identified as candidate regions. Information on the identified candidate regions is stored in the RAM in the control unit 111.

In S304, the inverted region determination unit 213 acquires a partial binary image corresponding to the candidate region identified in S302, performs a predetermined analysis process on the partial binary image, and determines an inverted region in the binary image based on the obtained analysis result. In the case of the highlighted image shown in FIG. 4(c) described above, of the black pixel blocks 401 and 402 serving as candidate regions, the black pixel block 401 is determined to be the inverted region. Details of the inverted region determination process will be described later.

In S305, the pixel inversion unit 214 performs a process of inverting pixels in the binary image obtained in S301 that correspond to the inversion area determined in S304. Specifically, for each pixel that constitutes the inversion area, a process of changing the pixel value to a value different from the current value (i.e., changing "1" to "0" and "0" to "1") is performed, so that if it is a white pixel, it is changed to a black pixel, and if it is a white pixel, it is changed to a black pixel. This pixel inversion process changes the pixel attribute only for the pixels that constitute the inversion area in the binary image. In other words, for each pixel that constitutes a normal character or ruled line in the binary image, black pixels remain as black pixels and white pixels remain as white pixels, and only for the part of the white-filled character, black pixels are changed to white pixels and white pixels are changed to black pixels. In this way, a binary image in which only the inversion area is not inverted, as shown in FIG. 4B described above, is obtained. This process inverts the character string in the inversion area, and a binary image that does not include so-called white-filled characters can be obtained. By performing OCR processing on this binary image, the OCR accuracy can be improved.

In addition, in this embodiment, all pixels in the image area determined to be an inversion area are inverted, but some pixels may be left as they are without being inverted. For example, by not inverting the pixels on the outer frame of the inversion area, it is possible to leave the outer frame of the inversion area (delimiter information equivalent to the ruled lines) as shown in FIG. 4(d). In this case, it is sufficient to leave only the pixels of a portion of the area determined to be an inversion area that is a specified width from the outer edge. Additional processing of this level does not significantly increase the processing load. By leaving the delimiter information equivalent to the ruled lines in this way, it becomes possible to use it as delimiter information for the character string that is the character recognition result during OCR processing.

(Details of character area determination process)
Next, the details of the inverted region determination process (S304) described above will be described with reference to the flowchart of FIG.

In S501, information on all candidate regions identified in S303 and the binary image data obtained in S301 are read from the RAM. In the following S502, a binary image (hereinafter referred to as a "partial binary image") of a portion of all candidate regions read in S501 that corresponds to a candidate region of interest is generated. Specifically, an image area of a portion that corresponds to the candidate region of interest is cut out from the binary image obtained in S501.

Next, in S503, the black pixel density in the partial binary image obtained in S502 is calculated; specifically, the proportion of black pixels among all pixels constituting the partial binary image. Then, in S504, the contours of the white pixel parts in the partial binary image obtained in S502 are extracted. The contours extracted here are the straight line parts of the components of characters, etc., and for example, several tens of contours are extracted per character.

Then, in S505, it is determined whether the target candidate region is an inverted region based on the black pixel density calculated in S503 and the number of contour lines extracted in S504. Specifically, a predetermined threshold is defined in advance for each of the black pixel density and the number of contour lines, and if both are within the range of the predetermined threshold, the target candidate region is deemed to contain characters and is determined to be an inverted region. On the other hand, if either or both of the black pixel density and the number of contour lines exceed the range of the predetermined threshold, the target candidate region is deemed to contain no characters and is determined to be not an inverted region. In this case, the threshold for the black pixel density is determined by taking into consideration that the closer the black pixel density is to 100%, the fewer the number of white pixels that form characters in the determined character region, and that the more complex the character is, the lower the black pixel density is. In other words, a value such as 75% may be set based on the black pixel density expected for the most complex character that can be used as a white-out character. The threshold for the number of contour lines may be set by taking into consideration the average number of contour lines per character and the expected number of characters in the inverted region. By adding this number of contour lines to the judgment index, the judgment accuracy can be improved. If it is determined that the target candidate region is an inverted region, the process proceeds to S506. On the other hand, if it is determined that the target candidate region is not an inverted region, the process proceeds to S507.

In S506, the target candidate region is set as an inverted region. Specifically, information such as a flag indicating that the target candidate region is an inverted region is generated and stored in the RAM in the control unit 111.

In S507, it is determined whether processing has been completed for all of the candidate areas read out in S501. If there are any unprocessed candidate areas, the process returns to S502, where the next candidate area of interest is selected and processing continues. On the other hand, if processing has been completed for all candidate areas, the process exits.

The above is the content of the character region determination process. Note that in the black pixel density calculation process in S503 and the contour extraction process in S504, a partial binary image cut out from the binary image is used, and an enhanced image obtained by applying morphology processing is not used. The reason is that the emphasized image may cause the character lines of the white character part to be blurred due to the influence of the contraction process or expansion process, and the character-likeness determination based on the black pixel density and the number of contour lines may not be performed as expected. By using a binary image, it is possible to perform a determination with higher accuracy. In addition, the character-likeness determination index is not limited to the black pixel density and the number of contour lines. For example, connected pixel clusters (CC) may be extracted from white pixels in the partial binary image, and the character-likeness may be determined based on the size of the extracted connected pixel clusters (the larger the CC, the more white pixels that form a character) and the spacing between the connected pixel clusters (the more complex the character, the narrower the spacing).

As described above, according to this embodiment, pixels that are unlikely to form an inverted region are excluded from the target pixels for pixel inversion processing at an early stage. This reduces the processing cost required for searching for an inverted region. In addition, since the determination of whether or not an inverted region is present is performed on a partial binary image after excluding information such as characters and lines that form tables from within the document image, it is possible to realize processing that is less dependent on the number of characters or the number of table cells.

[Embodiment 2]
In the first embodiment, a candidate region is specified from the enhanced image, and an inverted region is determined using a binary image (partial binary image) corresponding to the candidate region. However, the candidate region that is the premise of the partial binary image may not be appropriate. For example, when the outline of the white pixels forming the character is close to the edge of the inverted region, a thin black pixel line is formed between the outer edge of the inverted region and the white pixels forming the character, and the black pixels in the intervening portion may be lost by morphology processing. FIG. 6(a) shows a binary image obtained by performing a binarization process on a scanned image of an estimate having two inverted regions 600a and 600b, and FIG. 6(b) shows an enhanced image obtained by performing a morphology process on the binary image. In the enhanced image shown in FIG. 6(b), the inverted region 600b is separated into six small regions 611 to 616. When the extraction process (S303) of connected pixel blocks is applied to this enhanced image, each of the black pixel blocks corresponding to the small regions 611 to 616 is identified as a candidate region. As a result, the expected inversion region determination result cannot be obtained. Therefore, an aspect capable of dealing with the above-mentioned problem will be described as embodiment 2. Note that a description of the contents common to embodiment 1, such as the basic configuration of the information processing system, will be omitted, and the following description will focus on the inversion region determination process, which is the difference.

(Details of the inversion area determination process)
7 is a flowchart showing details of the inversion region determination process according to this embodiment. The following will be described with reference to the flowchart in FIG.

S701 is the same as S501 in the flow of FIG. 5 in embodiment 1, and the information of all candidate regions generated in S303 and the binary image data obtained in S301 are read from RAM.

In the next step S702, all the candidate areas read out in S701 are re-evaluated. In this embodiment, the re-evaluation involves integrating the read candidate areas based on their positional relationships. More specifically, for each candidate area, it is determined whether the distance between the candidate area and the adjacent candidate areas is within a predetermined distance, and the candidate areas that are within the predetermined distance are integrated to generate a single larger candidate area. In the example of FIG. 6B described above, the six candidate areas corresponding to the six small areas 611 to 616 are integrated into one, and the area indicated by the dashed frame is newly identified as the candidate area. Then, the information on the candidate area that was integrated is deleted (erased from RAM) since it is no longer necessary. Note that, for the relatively large area 610 corresponding to the inverted area 600a, there is no other candidate area whose inter-area distance satisfies the threshold, so it is maintained as it is without being integrated with the other candidate areas.

The processes in steps S703 to S708 correspond to steps S502 to S507 in the flow in FIG. 5 of embodiment 1, and as there are no particular differences, a description of them will be omitted.

The above is the content of the inverted region determination process according to this embodiment. In this embodiment, even in cases where what should originally be extracted as one black pixel block is extracted as separate black pixel blocks, by performing the candidate region integration process, it is possible to treat them as one black pixel block (candidate region) that is a candidate for the inverted region. In the example of FIG. 6(b) above, the candidate region shown by the dashed rectangle is generated by the integration process, and each process from S703 onwards is applied to the integrated candidate region. As a result, the image region corresponding to the entire inverted region 600b is set as the inverted region in S707, so that it can be appropriately uninverted in S305. FIG. 6(c) shows the result of applying the process of this embodiment to the binary image of the estimate. It can be seen that not only the inverted region 600a but also the inverted region 600b is appropriately uninverted.

<Modification>
When performing the reevaluation as described above, there are cases where the candidate regions are excessively integrated. FIG. 8(a) shows a binary image obtained by performing a binarization process on a scanned image of an estimate having four inverted regions 800a to 800d. FIG. 8(b) shows an enhanced image obtained by performing a morphology process on the binary image. In the enhanced image shown in FIG. 8(b), the inverted region 800b is separated into six small regions 811 to 816, similar to the inverted region 600b in FIG. 6(b) described above, and a total of nine candidate regions 810 to 818 exist in the entire enhanced image. If the integration process in the reevaluation described above is applied to these nine candidate regions as is, the candidate region 817 and the candidate region 818 are close to each other, and therefore the two candidate regions are integrated. In this case, the candidate region 817 and the candidate region 817 are close to each other, but they do not constitute a single inverted region. As a result, in the new candidate area obtained by merging the two candidate areas, even though there are areas that are not actually inverted areas, they are all uninverted, and even pixels that do not constitute the inverted area are inverted. Therefore, a modified example of a mode in which it is determined whether or not to merge prior to the merging process will be described.

Figure 9 is a flowchart showing the details of the candidate area reevaluation process according to this modified example. The following will be explained with reference to the flowchart in Figure 9.

In S901, among all the candidate areas read out in S701, candidate areas that may be integrated are identified, and information representing a group of the identified candidate areas (hereinafter referred to as "provisional integration information") is generated. When generating this provisional integration information, the method described in S702 above is applied. Specifically, among each candidate area, a candidate area that satisfies the condition that the distance from other candidate areas is within a predetermined threshold is identified, and information that can identify each candidate area that may be integrated is generated, which indicates an area in which the identified candidate areas are provisionally integrated into one (hereinafter referred to as "provisional integration area"). Figure 8 (c) shows a provisional integration area 819 obtained by combining candidate areas 811 to 816 into one out of the nine candidate areas 810 to 818 shown in Figure 8 (b), and a provisional integration area 820 obtained by combining candidate areas 817 and 818 into one. Then, provisional integration information corresponding to these two provisional integration areas 819 and 820, respectively, is generated. The subsequent steps S902 to S905 are executed for each piece of provisional integrated information obtained in S901.

First, in S902, for each piece of provisional integration information generated in S901, an area to be evaluated to determine whether integration is possible (hereinafter referred to as an "evaluation area") is identified. Specifically, using information on multiple candidate areas included in the provisional integration information, a portion of the provisional integration area excluding the multiple candidate areas is identified as the evaluation area. Then, in the following S903, based on the evaluation area identified in S902 and the binary image read out in S701, the black pixel density in the image area in the binary image corresponding to the evaluation area is calculated.

Next, in S904, a process for determining whether or not each provisionally integrated region can be integrated is performed based on the black pixel density calculated in S903. Specifically, if the calculated black pixel density is equal to or greater than a certain level defined in advance, it is determined that integration is possible, and if it is less than the certain level, it is determined that integration is not possible. The threshold value corresponding to the certain level in this case may be a value lower than the threshold value in S505 described above. This is because if the presence of a certain amount of black pixels can be confirmed, it can be estimated that the evaluation region is originally an area of black pixels (the "ground" area in the inverted area). By using the black pixel density to determine whether or not integration is possible, it is possible to determine whether the part set as a new candidate region by integration is appropriate as a candidate for the inverted area in the binary image. For example, in the case of the provisionally integrated region 820, there are no black pixels in the part 821 corresponding to the evaluation region in the binary image shown in FIG. 8(a) (the black pixel density is extremely low). In contrast, in the case of the provisionally integrated region 819, the density of black pixels in the part corresponding to the evaluation region in the binary image shown in FIG. 8(a) is high. Therefore, the provisional merged area 819 is determined to be mergeable, and the provisional merged area 820 is determined to be unmerged.

The above is the content of the candidate area re-evaluation process according to this modified example. Figure 8(d) shows the result of applying the process of this modified example. It can be seen that the inverted areas 800c and 800d in the binary image of Figure 8(a) are each processed as separate inverted areas and are appropriately uninverted. In this way, by determining whether or not merging is possible before performing it in the re-evaluation of the candidate areas, unnecessary merging can be avoided, and as a result, pixel parts that are not inverted areas can be prevented from being uninverted.

As described above, according to this embodiment, by reevaluating the identified candidate areas, it is possible to identify inverted areas within a document image with greater accuracy.

[Embodiment 3]
In the first and second embodiments, the size of the enhanced image is the same as the size of the binary image of the input scan image. Next, an aspect in which an enhanced image is generated using a reduced binary image will be described as the third embodiment. Note that the following describes the differences from the first and second embodiments, i.e., the generation of the enhanced image and the subsequent identification of the candidate region.

First, a binary image is generated from the input scanned image (S301), and before moving on to generating an enhanced image, the binary image generated in S301 is reduced. Next, a reduction and expansion process is performed on the reduced binary image to generate an enhanced image (S302). Then, a process is performed to extract black pixel blocks from the enhanced image, and the extracted black pixel blocks are converted to their pre-reduction size based on the scaling factor used in the reduction process to identify candidate areas (S303). At this stage, the size of the candidate areas remains the same as if the reduction process had not been performed, so the processes from S304 onwards can be applied in the same way.

By using a binary image reduced in size as described above, the load of morphological processing and the process of extracting connected pixel blocks can be reduced, and the smaller image size also reduces the memory capacity required for temporary storage.

However, when using a reduced binary image, it is necessary to keep in mind that there are cases where inverted areas with a small gap between them are combined. FIG. 10(a) shows a partial binary image obtained by performing a binarization process on a table portion including an inverted area of a certain scanned image. The partial binary image in FIG. 10(a) contains four inverted areas 1001-1004. FIG. 10(b) is a schematic diagram of a reduced binary image obtained by performing a reduction process on the partial binary image in FIG. 10(a) at a predetermined magnification. The four inverted areas 1001-1004 in the partial binary image shown in FIG. 10(a) are combined by the reduction process to become one black pixel block 1005. In this way, the reduction process may cause the boundaries between multiple inverted areas to disappear. When the pixel inversion process (S305) is applied to the inverted areas that have been unintentionally combined in this way, even pixels that are not actually in the inverted areas will be inverted.

To avoid the above problem, in the reevaluation of the candidate regions (S702) described in the second embodiment, prior to the integration process, a region division process using the binary image before reduction is performed on each candidate region. Specifically, a partial image corresponding to the candidate region is cut out from the binary image before reduction, and the cut-out partial image is projected in the x-axis and y-axis directions. Next, the generated projection is analyzed to identify a group of coordinate positions below a predefined threshold. Then, the region consisting of the identified group of coordinate positions is determined as a region outside the candidates for the inverted region (hereinafter referred to as an "exclusion region"). The candidate region is divided so as not to include the determined exclusion region. Here, a specific example is used for explanation. FIG. 11 shows the result of projecting the partial binary image shown in FIG. 10(a) in the x-axis direction. In this case, three regions below the threshold th (between x1 and x2, between x3 and x4, and between x5 and x6) are determined as exclusion regions. Then, as shown in FIG. 10(c), the candidate area corresponding to the black pixel block 1005 in FIG. 10(b) is divided into four areas 1006-1009 with the above three exclusion areas as boundaries. Then, each of these four areas 1006-1009 is identified as a new candidate area. After that, as described in the modified example of the second embodiment, provisional integration information is generated for all candidate areas including each candidate area after division, and the possibility of integration is determined for each provisional integration information. At this time, the candidate areas obtained by division are not integrated. As described above, the possibility of integration is determined based on the black pixel density between candidate areas that can be integrated in a binary image, but each candidate area obtained by division by projection usually has a low black pixel density, so it is determined that integration is not possible.

As described above, by using a reduced binary image, the load of morphology processing and connected pixel block extraction processing can be reduced, and the capacity of temporary storage devices can also be reduced. In addition, by additionally performing region segmentation processing on the candidate regions, it is possible to deal with cases where narrowly spaced inverted regions are combined due to the reduction processing, making it possible to reduce the processing load while maintaining accuracy.

Other Embodiments
The present invention can also be realized by a process in which a program for implementing one or more of the functions of the above-described embodiments is supplied to a system or device via a network or a storage medium, and one or more processors in a computer of the system or device read and execute the program. The present invention can also be realized by a circuit (e.g., ASIC) that implements one or more of the functions.

Claims (13)

a processing means for performing morphological processing including at least a contraction process for removing normal characters from a document image including an inverted region including characters in which the background and foreground are inverted, and normal characters in a region other than the inverted region, to generate an enhanced image in which connected pixel blocks that may constitute the inverted region remain;
an analysis means for analyzing a partial image of the document image corresponding to a candidate region for the inverted region identified based on information of the connected pixel blocks remaining in the highlighted image;
a specifying means for specifying the inverted area in the document image, the inverted area including the character in a state where the background and the foreground are inverted , based on the analysis result by the analyzing means ;
Equipped with
the specifying means, when there are a plurality of candidate regions, integrates the candidate regions based on their positional relationships, and specifies the inverted region based on an analysis result of a partial image of the document image corresponding to the integrated candidate region;
13. An image processing device comprising: The image processing device according to claim 1, characterized in that the processing means performs an expansion process on the result of the contraction process as the morphological process to generate the enhanced image. The image processing device according to claim 1 or 2, characterized in that the document image is a binary image obtained by binarizing a scanned image of a document. the analysis result includes a black pixel density in the partial image and a number of contours of white pixel portions in the partial image;
4. The image processing device according to claim 1, wherein the first and second inputs are input to the image processing apparatus. The analysis result is a size of a pixel cluster formed by connecting white pixels in the partial image or a distance between the pixel clusters.
4. The image processing device according to claim 1, wherein the first and second inputs are input to the image processing apparatus. 6. The image processing device according to claim 1, wherein the integration is a process of combining candidate regions into one when the distance between each candidate region and an adjacent candidate region is within a certain distance. The image processing device according to claim 6, characterized in that the identification means does not perform the integration if the black pixel density in the image area between the candidate areas in the document image corresponding to the distance is less than a certain level, even if the distance is within a certain distance. The document image is a reduced binary image obtained by performing a reduction process at a predetermined magnification on a binary image obtained by binarizing a scanned image of a document,
The image processing device according to any one of claims 1 to 7, characterized in that the identification means performs area segmentation processing using the binary image before the reduction processing for each of the candidate areas, and performs the integration on the candidate areas obtained by performing the area segmentation processing. an inversion means for inverting each pixel constituting the inversion region identified in the document image;
a character recognition means for performing character recognition processing on an image obtained by inverting each pixel constituting the inverted region;
The image processing device according to claim 1 , further comprising: a processing means for performing morphology processing, including at least a contraction process for removing normal characters, on a document image including an inverted region including characters in which the background and foreground are inverted, and normal characters in an area other than the inverted region, to generate an enhanced image in which connected pixel blocks that may constitute the inverted region remain;
an analysis means for analyzing a partial image of the document image corresponding to a candidate region for the inverted region identified based on information of the connected pixel blocks remaining in the highlighted image;
a specifying means for specifying the inverted area in the document image, the inverted area including the character in a state where the background and the foreground are inverted, based on the analysis result by the analyzing means;
Equipped with
The analysis result includes at least the number of contours of white pixel portions in the partial image.
13. An image processing device comprising: a generating step of performing morphology processing including at least a contraction process for removing normal characters from a document image including an inverted region including characters in which the background and foreground are inverted, and an ordinary character in an area other than the inverted region, to generate an enhanced image in which connected pixel blocks that may constitute the inverted region remain;
an analysis step of analyzing a partial image of the document image corresponding to a candidate region for the inverted region identified based on information of the connected pixel blocks remaining in the highlighted image;
a specifying step of specifying the inverted area in the document image, the inverted area including the character in a state where the background and the foreground are inverted , based on the analysis result of the analyzing step;
Including ,
In the identifying step, when there are a plurality of candidate regions, the candidate regions are integrated based on their positional relationships, and the inverted region is identified based on an analysis result of a partial image of the document image corresponding to the integrated candidate region.
23. An image processing method comprising: a generating step of performing morphology processing including at least a contraction process for removing normal characters from a document image including an inverted region including characters in which the background and foreground are inverted, and an ordinary character in an area other than the inverted region, to generate an enhanced image in which connected pixel blocks that may constitute the inverted region remain;
an analysis step of analyzing a partial image of the document image corresponding to a candidate region for the inverted region identified based on information of the connected pixel blocks remaining in the highlighted image;
a specifying step of specifying the inverted area in the document image, the inverted area including the character in a state where the background and the foreground are inverted, based on the analysis result of the analyzing step;
Including,
The analysis result includes at least the number of contours of white pixel portions in the partial image.
13. An image processing method comprising: A program for causing a computer to function as the image processing device according to any one of claims 1 to 10 .

Images