JP2008225739A - Feature amount measuring method and feature amount measuring apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem that a feature amount of obliquely connected pixel components may not be measured conventionally by one image scanning. <P>SOLUTION: A feature amount measuring method uses a feature amount table storing a frequency of merging and a feature amount, and a feature amount pointer table storing an index of the feature amount table, refers to pixels within a window including a target pixels to be measured, and measures while matching the feature amounts of referred pixels and the target pixels in accordance with the presence of pixels to which pointer indexes indicating data locations in the feature amount pointer table within the window are given and the presence of discrepancies of the pointer indexes given to the pixels within the window, other than the target pixels. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a feature amount measuring method and a feature amount measuring apparatus for measuring a feature amount measurement process of a binary image used in an image processing apparatus by one scan.

  Usually, the feature value measurement processing of a binary image is performed by, for example, labeling a natural number for the same connected component in a connected component in which pixels indicating 1 are connected vertically, horizontally, and diagonally, and the area of each connected component It is divided into feature amount measurement for measuring feature amounts such as. When the connected component has a complicated shape, a plurality of labels are given to the same connected component in a single raster scan, and the image is scanned a plurality of times until the labels are matched.

  In order to solve such a problem, an algorithm for measuring a feature amount by one scan by managing pointers in a memory has been proposed (for example, see Patent Document 1). In this method, 2 × 2 pixels are measured and a pattern is applied, and the label of each raster is pointer-managed using three memories based on the pattern.

Japanese Patent Laid-Open No. 5-12444 (pages 3-5, FIGS. 1-13)

  However, the above conventional technique has a limitation that pixel components connected obliquely are not measured. Further, in the above-described prior art, 2 × 2 pixels must be measured even when the target pixel for performing feature amount measurement is 0 in order to apply the pattern. In general, since the processing speed of software is slower than that of hardware, if the number of pixels to be referenced is large, the number of accesses to the memory increases, and the software processing performance deteriorates.

  The present invention has been made to solve such a problem. In the image processing, the feature amount is measured by one scanning, the feature amount of the pixel component obliquely connected can be measured, and the number of pixels to be referred to is determined. The purpose is not to decrease the processing performance of the software by reducing it.

The feature amount measuring method according to the present invention includes:
In the feature quantity measurement method for measuring the feature quantity of an image,
A feature quantity table for storing the feature quantity and the number of merges of the feature quantities;
A feature amount pointer table for storing a feature amount index indicating a storage location of the feature amount and the number of merges stored in the feature amount table;
Use
Refer to the pixel in the window including the target pixel for which the feature amount measurement is performed, and if there is no pixel assigned a pointer index indicating the data storage location of the feature amount pointer table in the window other than the target pixel, new A first step of creating a feature quantity and storing it in an unused area of the feature quantity table, and storing a feature quantity index indicating the new feature quantity in the feature quantity pointer table;
When there is a pixel assigned a pointer index in the window other than the pixel of interest with reference to a pixel in the window including the pixel of interest, and the pixel assigned the pointer index is consistent, a feature amount pointer A second step of incrementing the feature value indicated by the feature value index stored in the pointer index of the table;
If there is a pixel assigned a pointer index in the window other than the pixel of interest with reference to a pixel in the window including the pixel of interest, and there is a contradiction in the pixel assigned the pointer index, the feature amount Stores the feature quantity that is the result of merging the corresponding feature quantity in the table and the corrected number of merges, and merges the merged feature quantity index among the feature quantity indexes stored in the pointer index of the feature quantity pointer table A third step of rewriting to the feature amount index and indicating the feature amount after merging;
It is characterized by having.

  As in the present invention, by performing the above processing using a window including the left, upper left, upper, and upper right pixels with respect to the target pixel, it is possible to match the feature amount with the pixel diagonally above the target pixel. Therefore, it is possible to measure the feature amounts of pixel components that are obliquely connected by one scan.

Embodiment 1 FIG.
FIG. 1 is a diagram showing a data structure and an image scanning direction of a feature amount measuring method according to Embodiment 1 of the present invention. FIG. 1A is a diagram illustrating a data structure used for feature quantity measurement of the input image 11 in the first embodiment. The input image 11 is a binary image of 0/1. Among them, a square frame with hatching is a pixel (hereinafter referred to as 1 pixel) 12 indicating a value of 1, and a square frame surrounded by other dotted lines is 0. A pixel (hereinafter referred to as 0 pixel) 13 indicating the value of.

  The input image 11 is captured as image data 14 when image processing is performed. Accordingly, in the initial state of image processing, the image data 14 and the input image 11 are the same. A square frame with dots in the image data 14 is a target pixel 15 for measuring a feature quantity. In the feature quantity measurement, a pixel to which a pointer index of a feature quantity pointer table 17 described later is added is shown. A thick frame including the target pixel 15 is a window 16 indicating a region of interest when the feature amount measurement of the target pixel 15 is performed.

  The feature value pointer table 17 is a table that stores a feature value index of a feature value table 18 described later. Therefore, when a plurality of the same feature amount indexes are stored in the feature amount pointer table 17, it means that they point to the same feature amount index. The pointer index indicates a storage location of data (feature amount index) in the feature amount pointer table 17.

  The feature quantity table 18 is a table that includes a feature quantity storage area 19 that stores measured feature quantities and a merge count storage area 20 that stores the number of times feature quantities have been merged in the past. Each element of the feature amount table 18 has a valid / invalid state including an initial state. The initial state of both the feature amount storage area 19 and the merge count storage area 20 is “0”. The feature amount index indicates a storage location of data (feature amount and merge count) in the feature amount table 18.

  FIG. 1B shows the scanning direction of the image data 14 to be subjected to feature quantity measurement with arrows.

  FIG. 2 shows a flowchart of the feature amount measurement process. In the flowchart shown in FIG. 2, the process of step S16 described later is shown in FIG. 3, the process of step S18 described later is shown in FIG. 4, the process of step S19 described later is shown in FIG. 5, and the process of step S19 described later is shown. It is shown in FIG.

  The specific processing content of the feature amount measurement will be described with reference to the flowcharts of FIGS. Specific processing contents will be described with reference to FIGS. 7 to 10 as appropriate together with the flowcharts of FIGS. Note that the feature value taken up in the embodiment of the present invention is an area, that is, the number of one pixel 12, and the feature value merge process indicates a process of adding two areas. Further, when invalidating the storage area of the feature quantity table 18 in the merge processing, the area value as the feature quantity is invalidated by setting it to 0.

  In step S11 of FIG. 2, the input image 11 which is a binary image is input as the image data 14 for measuring the feature amount. Therefore, the initial state of the image data 14 is the same as that of the input image 11, and the initial data of the feature amount pointer table 17 and the feature amount table 18 is set to “0”.

  In step S12, a scanning loop process of the input image 11 is started, and feature quantity measurement is performed according to the flowchart of FIG. In this loop processing, the scanning of the input image 11 is repeated for the pixels, and the loop processing is terminated when the scanning is completed.

  In step S <b> 13, it is determined whether the target pixel 15 is the 1 pixel 12 or the 0 pixel 13. FIG. 7 is a diagram showing a series of processes of step S13, step S15, and step S16. FIG. 7A is a diagram illustrating a scanning direction in which scanning has been performed up to the target pixel 15. In FIG. 7A, the scanning direction is indicated by an arrow. In the image data 14 shown in FIG. 7A, since the pixel 13 of 0 continues to the target pixel 15, NO is selected in step S13, and the process jumps to step S14 to repeat the input image scanning loop.

  As shown in FIG. 7A, the order of referring to the pixels other than the target pixel 15 with respect to the target pixel 15 is, for example, the order of A, B, C, and D in the first embodiment. . However, since all the pixels other than the target pixel 15 are referred to, the reference order is not limited to this.

  FIG. 7B is a diagram showing one pixel 12 detected first in the scanning process. In the feature amount measurement, measurement is first performed with the one pixel 12 as the target pixel 15. Since the target pixel 15 is one pixel 12, step S13 is YES.

  In step S15, it is determined whether there is a value greater than 0 in the window 16 other than the target pixel 15. In FIG. 7B, since all the pixels in the window 16 excluding the target pixel 15 are 0 pixels 13, NO is determined in step S15, and the process proceeds to step S16.

  In step S16, a new feature amount is created and the target pixel is updated. FIG. 3 shows details of the processing in step S16. The process of step S16 is comprised from step S31 and step S32 as shown in FIG.

  In step S31, unused areas of the feature quantity table 18 and the feature quantity pointer table 17 are secured, and the feature quantity index of the secured feature quantity table 18 is stored in the secured feature quantity pointer table 17. The newly secured feature value pointer table 17 is set to indicate the new feature value created in step S32. FIG. 7B shows the processing so far.

  In step S32, a new feature quantity is created and stored in the area of the secured feature quantity table 18. Specifically, “1” is stored as the area in the feature amount storage area 19 of the feature amount table 18. The merge count storage area 20 corresponding to the stored feature quantity storage area 19 is set to “0” (since the area storing the new feature quantity has never been merged). The pointer index assigned to the target pixel 15 of the image data 14 is the pointer index of the secured feature amount pointer table 17 (m in FIG. 7). Then, the process returns to step S16. The result of the above processing is shown in FIG. When the process returns to step S16 in FIG. 2, the process proceeds to step S14, and the loop process for feature quantity measurement is repeated.

  FIG. 8 is a diagram in which the loop processing for feature quantity measurement is continued after FIG. 7C, and is a diagram showing a series of processing of step S13, step S15, step S17, and step S18 of FIG.

  In FIG. 8A, the scanning direction in which scanning has been performed up to the target pixel 15 is indicated by an arrow. In the scanning direction shown in FIG. 8A, a loop process with the pixel 13 of 0 is performed after FIG. 7C. Thereafter, since there is one pixel 12, the same processing as that shown in FIG. 7 is performed, and in order to make this one pixel 12 a new feature amount, a feature index is added to the pointer index (m + 1) of the feature amount pointer table 17. (N + 1) indicating the unused area of the quantity index is stored.

  In the feature quantity table 18, the area “1” as the feature quantity is stored in the feature quantity storage area 19 of the feature quantity index (n + 1). Since the number of merges is 0, “0” is stored in the merge number storage area 20 of the feature index (n + 1). The corresponding pixel is assigned (m + 1) which is the pointer index of the secured feature value pointer table 17. Thereafter, the scanning is repeated according to FIG. 8A, and the state of processing the next target pixel is FIG. 8B.

  Consider the pixel of interest 15 in FIG. In step S13, since the target pixel 15 is the one pixel 12, it determines with YES and progresses to step S15. In step S15, since there is one pixel 12 whose pointer index is m above the target pixel 15, it is determined YES, and the process proceeds to step S17.

  In step S17, it is determined whether or not there is a contradiction in the pointer index in the window 16. For example, in the case of FIG. 8, in the window 16 excluding the target pixel 15, all the pixels 13 other than the pixel of the pointer index m are 0 pixels 13, so it is determined that there is no contradiction (NO) and the process proceeds to step S 18.

  Here, since one pixel 12 in the window 16 must be the same connected component, the feature index stored in the pointer index assigned to the one pixel 12 in the window 16 is the same. Must point to the feature quantity.

  Therefore, “the pointer index is inconsistent” means that the feature amount index stored in the assigned pointer index is different in the pixels in the window 16 excluding the target pixel 15. On the other hand, “the pointer index is consistent” means that the feature amount index stored in the assigned pointer index is the same in the pixels in the window 16 excluding the target pixel 15. Also, in this specification, “matching feature amounts” means making the state in which the contradiction occurs without any contradiction.

  In step S18, the feature amount and the target pixel are updated. FIG. 4 shows details of the processing in step S18. The process in step S18 includes step S41 as shown in FIG.

  In step S41, the feature amount is first updated. In the case of FIG. 8B, since the pointer index of the upper pixel in the reference order B is m, the feature quantity pointed to by the feature quantity index n stored in the pointer index m of the feature quantity pointer table 17, that is, The feature amount (area) of the target pixel 15 is incremented with respect to the feature amount index n of the feature amount table 18. Therefore, the feature quantity storage area 19 in the feature quantity index n of the feature quantity table 18 is “1” + “1” = “2”.

  In step S41, the pointer index of the target pixel 15 is the pointer index of the associated pixel. That is, the pointer index m of the upper pixel is set as the pointer index of the target pixel 15. Then, the process returns to step S18. The result of the above processing is shown in FIG. When the process returns to step S18 in FIG. 2, the process proceeds to step S14, and the loop process for feature quantity measurement is repeated.

  FIG. 9 is a diagram in which the loop process for feature quantity measurement is continued after FIG. 8C, and is a diagram showing a series of processes of step S13, step S15, step S17, and step S19 in FIG.

  In FIG. 9A, the scanning direction in which scanning has been performed up to the target pixel 15 is indicated by an arrow. Consider the pixel of interest 15 in FIG. In step S13, since the target pixel 15 is the one pixel 12, it determines with YES and progresses to step S15.

  In step S15, since one pixel 12 having the pointer index m is located on the upper left and left of the target pixel 15 and one pixel 12 having the pointer index (m + 1) is present on the upper right, there is a value greater than 0 ( YES), the process proceeds to step S17.

  In step S17, it is determined whether or not there is a contradiction in the pointer index in the window 16. For example, in the window 16 excluding the target pixel 15, the pointer index of the upper left and left pixels of the target pixel 15 is m, and the pointer index of the upper right pixel of the target pixel 15 is (m + 1), so there is a contradiction ( YES), the process proceeds to step S19.

  In step S19, feature amount merge processing and feature amount pointer table 17 matching processing are performed. FIG. 5 and FIG. 6 detail the processing contents of step S19. Among these, FIG. 5 shows processing related to the feature amount merge processing, and FIG. 6 shows processing related to the matching processing of the feature amount pointer table 17.

  The feature amount merging process includes step S51 and step S52 as shown in FIG. The matching process of the feature amount pointer table 17 includes steps S61 to S67 as shown in FIG. It should be noted that either the feature amount merging process or the matching process of the feature amount pointer table 17 may be performed first, but the present embodiment shows an example in which the feature amount merging process is performed first.

  In FIG. 5, in step S51, the feature amount of the target pixel 15 is added and the feature amount is merged. In step S51, the target pixel 15 is compared with the feature amount pointed to by the feature amount index n stored in the pointer index m of the feature amount pointer table 17, that is, the feature amount index n of the feature amount table 18. The feature amount (area) of 15 is incremented. Accordingly, the feature quantity storage area 19 in the feature quantity index n of the feature quantity table 18 is “2” + “1” = “3”.

  Note that there are two types of pointer indexes m and (m + 1) for pixels other than the target pixel 15 included in the window 16. Among these, referring to the pixels other than the target pixel 15 in the order of A to D shown in FIG. 7A, in FIG. 9B, the reference to the upper left pixel (pointer index m) of the target pixel 15 is first performed. (Reference order A). Accordingly, the pointer index assigned to the target pixel 15 is m in the reference order, and the feature amount is incremented as described above.

  In step S51, the pointer index of the target pixel 15 is the pointer index of the associated pixel. That is, the pointer index of the target pixel 15 is m as described above. FIG. 9B shows a series of processes so far.

  In step S51, feature quantities having inconsistent pointer indexes are merged. Specifically, the feature quantity “3” stored in the feature quantity index n of the feature quantity table 18 and the feature quantity “1” stored in the feature quantity index (n + 1) of the feature quantity table 18 are merged. Then, “3” + “1” = “4” is stored in the feature amount index n of the feature amount table 18 as a merge result.

  In the case of FIG. 9, the feature amount is merged toward the feature amount index n when merging the feature amounts, but may be merged toward the feature amount index (n + 1).

  In step S52, the emptied feature amount area is invalidated. In the example of FIG. 9, the feature quantity storage area 19 becomes empty in the feature quantity index (n + 1) of the feature quantity table 18 as the feature quantities are merged, so the feature quantity index (n + 1) of the feature quantity table 18 is invalidated. To do. When the process of step S52 is completed, the process returns to step S19.

  Next, matching processing of the feature amount pointer table 17 is performed. In the matching processing of the feature value pointer table 17, as described in FIG. 6, first, in step S61, the values in the feature value pointer table 17 indicating the feature values to be invalidated are merged for the pixels in the window 16. Rewrite to point to the feature value later.

  For example, in the example of FIG. 9, the feature quantity index (n + 1) is invalidated. The pointer index of the feature value pointer table 17 storing this is (m + 1), and the pixel corresponding to this pointer index (m + 1) is the upper right pixel of the target pixel 15. Since the feature quantity index (n + 1) in the feature quantity table 18 becomes invalid, the value (n + 1) stored in the pointer index (m + 1) in the feature quantity pointer table 17 is rewritten to n, and the feature quantity index in the feature quantity table 18 Point to n.

  Since the feature quantity index (n + 1) of the invalid feature quantity table 18 has never merged the feature quantities in the past, the loop relating to the matching of the feature quantity pointer table outside the window 16, that is, steps S62 to S65 are executed. Otherwise, the process proceeds to step S66 (the processing related to step S65 from step S62 will be described later).

  In step S66, the number of merges in the feature value table 18 is added, and the merge number is stored in the merge number storage area 20 of the feature value table 18. In the example of FIG. 9, “1” corresponding to the current merge process is added to the initial value “0” in the merge count storage area 20 of the feature value index n in the feature value table 18, and “0” + “1” = “ 1 ”is entered.

  In step S67, the number of merges of the invalid feature quantity table 18 is set to zero. In the example of FIG. 9, since the feature value index (n + 1) in the feature value table 18 becomes invalid, “0” is stored in the merge count storage area 20 of the feature value index (n + 1). The result of the above processing is shown in FIG. When the process returns to step S19 in FIG. 2, the process proceeds to step S14, and the loop process for feature quantity measurement is repeated.

  FIG. 10 is a diagram in which the loop process for feature quantity measurement is continued after FIG. 9C, and is a diagram showing a series of processes of step S13, step S15, step S17, and step S19 in FIG.

  In FIG. 10A, the scanning direction in which scanning has been performed up to the pixel of interest 15 is indicated by an arrow, and a diagram in which the feature amount measurement loop is continued after FIG. 9C is shown. In FIG. 9C to FIG. 10A, the one pixel 12 indicating the pointer index (m + 2) is not included in the window 16, and therefore the index (m + 2) is a new feature amount. Is given.

  Further, an index m is assigned to the subsequent one pixel 12. This is because, according to the reference order shown in FIG. 7A, there are pixels in which the pointer index m is assigned to the upper right, upper, upper left, and upper left pixels for the corresponding pixels.

  In FIG. 10B, since the pointer index m of the feature amount pointer table 17 indicates the feature amount index n, n is stored. The pointer index (m + 1) of the feature value pointer table 17 stores n in order to indicate the feature value index n after merging. The pointer index (m + 2) of the feature quantity pointer table 17 stores (n + 2) to indicate the feature quantity index (n + 2) in which the new feature quantity is stored.

  Further, in the feature quantity storage area 19 of the feature quantity table 18, the feature quantity index n is merged with the feature quantity index n (area “7”) and the feature quantity (n + 1) feature quantity (area is “1”). “8” is stored. In the feature quantity storage area 19 of the feature quantity table 18, the feature quantity index (n + 1) is invalid, and therefore “0” is stored. In the feature quantity storage area 19 of the feature quantity table 18, “1” which is a new feature quantity is stored in the feature quantity index (n + 2).

  In the merge count storage area 20 of the feature quantity table 18, since the merge quantity “1” is invalidated for the feature quantity index n and the feature quantity index (n + 1) is invalid, “0” is merged for the feature quantity index (n + 2). The merge count “0” is stored.

  Consider the pixel of interest 15 in FIG. In step S13, since the target pixel 15 is the one pixel 12, it determines with YES and progresses to step S15.

  In step S15, since one pixel 12 with the pointer index m is on the left of the target pixel 15 and one pixel 12 with the pointer index (m + 2) is on the upper right, a value greater than 0 exists (YES) ) And the process proceeds to step S17.

  In step S17, it is determined whether or not there is a contradiction in the feature amount in the window 16. For example, in FIG. 10, in the window 16 excluding the target pixel 15, the left pointer index of the target pixel 15 is m, and the upper right pointer index of the target pixel 15 is (m + 2), and different feature amount indexes are stored. Therefore, it is determined that there is a contradiction (YES), and the process proceeds to step S19.

  In step S19, the feature amount merge processing and the feature amount pointer table 17 matching processing are performed as described above.

  In the feature amount merging process shown in FIG. 5, in step S51, the feature amount is merged after adding the feature amount of the target pixel 15. In step S51, the target pixel 15 has the feature point indicated by the feature amount index (n + 2) stored in the pointer index (m + 2) of the feature amount pointer table 17, that is, the feature amount index (n + 2) of the feature amount table 18. ), The feature amount (area) of the target pixel 15 is incremented. Therefore, the feature quantity storage area 19 in the feature quantity index (n + 2) of the feature quantity table 18 is “1” + “1” = “2”.

  In addition, there are two types of pointer indexes m and (m + 2) for pixels other than the target pixel 15 included in the window 16. Among them, referring to pixels other than the target pixel 15 in the order of A to D shown in FIG. 7A, in FIG. 10B, pixels to which a pointer index is assigned at the upper left and above the target pixel 15 There is a pixel with a pointer index (m + 2) on the upper right. This pixel is referenced before the left pixel to which the pointer index m is assigned. Therefore, the pointer index assigned to the target pixel 15 is (m + 2).

  In step S51, the pointer index of the target pixel 15 is the pointer index of the associated pixel. That is, the pointer index of the target pixel 15 is (m + 2) as described above. FIG. 10B shows a series of processes so far.

  In step S51, feature quantities having inconsistent pointer indexes are merged. Specifically, the feature quantity “8” stored in the feature quantity index n of the feature quantity table 18 and the feature quantity “2” stored in the feature quantity index (n + 2) of the feature quantity table 18 are merged. Then, “8” + “2” = “10” is stored in the feature amount index (n + 2) of the feature amount table 18 as a merge result.

  In the case of the above example, when the feature amounts are merged, the feature amount index (n + 2) is merged. However, the feature amount index n may be merged.

  In step S52, the emptied feature amount area is invalidated. In the example of FIG. 10, the feature quantity storage area 19 becomes empty in the feature quantity index n of the feature quantity table 18 as the feature quantities are merged, so the feature quantity index n of the feature quantity table 18 is invalidated. When the process of step S52 is completed, the process returns to step S19.

  Next, matching processing of the feature amount pointer table 17 is performed. In the matching processing of the feature value pointer table 17, as described in FIG. 6, first, in step S61, the values in the feature value pointer table 17 indicating the feature values to be invalidated are merged for the pixels in the window 16. Rewrite to point to the feature value later.

  For example, in FIG. 10, the feature quantity index n is invalidated. In order to invalidate the feature quantity index n of the feature quantity table 18, the value n stored in the pointer index m of the feature quantity pointer table 17 is rewritten to (n + 2), and the feature quantity index (n + 2) of the feature quantity table 18 is changed. Try to point.

  Since the feature quantity index n of the invalid feature quantity table 18 has merged the feature quantities once in the past, a loop related to matching of the feature quantity pointer table outside the window 16, that is, steps S62 to S65 are executed. .

  The above contents mean the following. The pixels in the window 16 include pixels to which two types of pointer indexes of m and (m + 2) are assigned. Since these pixels indicate different feature amount indexes, they are in an “inconsistent state”. Therefore, in order to resolve the contradiction and to match the feature amounts, the feature amounts indicated by the pointer index m and the pointer index (m + 2) are made the same. This is equivalent to indicating the feature amount index (n + 2) in the above example.

  Here, the feature quantity index originally pointed to by the pointer index m is n, but “1” is stored when the merge count storage area 20 is referenced. This indicates that as a result of merging once in the past, there is another pointer index other than m indicating the feature amount index n. The other pointer index is assigned to a pixel outside the window 16. Then, the step S62 to the step S65 are performed in order to match the feature amounts of the other pointer index indicating the feature amount index n.

  In step S62, the feature quantity index n of the invalid feature quantity table 18 has merged the feature quantity once in the past as in the merge count storage area 20 of FIG. 10B. Therefore, the feature amount pointer table 17 is searched in step S62.

  In step S63, it is determined whether or not the feature quantity that has been emptied (or emptied) is indicated. For example, when the pointer index (m + 1) of the feature value pointer table 17 is referred to, the pointer index (m + 1) points to the feature value index n of the feature value table 18 that is emptied. Therefore, step S63 becomes YES and proceeds to step S64. If no empty (or empty) feature value is indicated, step S64 is skipped and the process proceeds to step S65.

  In step S64, the pointer is replaced with a pointer indicating the merged feature amount. For example, in the case of FIG. 10B, the pointer index (m + 1) in the feature value pointer table 17 is rewritten from n which has become invalid to (n + 2) in which the feature values are merged.

  Since the pointer has been corrected once, the other pointer index is the pointer index (m + 1), and since the rewriting of the feature value pointer table has been completed, the feature values are consistent and the contradiction is resolved. Exit the loop. If there are two or more pointer indexes outside the window 16 that must be matched with the feature amount, the loop from step S62 to step S65 is repeated.

  In step S66, the number of merges in the feature value table 18 is added, and the merge number is stored in the merge number storage area 20 of the feature value table 18. In the example of FIG. 10, the merge number storage area 20 of the feature value index (n + 2) in the feature value table 18 is set to the initial value “0”, and the number of merges so far “1” stored in the feature value index n. “1” for the current merge process is added, and “0” + “1” + “1” = “2” is entered.

  In step S67, the number of merges of the invalid feature quantity table 18 is set to “0”. In the example of FIG. 10, since the feature amount index n of the feature amount table 18 is invalid, “0” is stored in the merge count storage area 20 of the feature amount index n. The result of the above processing is shown in FIG. When the process returns to step S19 in FIG. 2, the process proceeds to step S14, and the loop process for feature quantity measurement is repeated.

  When all the scanning processes are completed and a series of loop processes from step S12 to step S14 in FIG. 2 are completed, the process proceeds to step S20. In step S20, the sum of the areas of connected components, which are feature amounts obtained by merging by the loop processing, is output.

  As described above, by performing the above processing using the window including the left, upper left, upper, and upper right pixels with respect to the target pixel 15, it is possible to match the feature amount with the diagonally upper pixel of the target pixel 15. Therefore, the feature amounts of pixel components connected obliquely as shown in FIG. 1 can also be measured by one scan.

  In merging the feature amounts, as a criterion for merging to which feature amount index, merging to the one with the larger number of merges, merging to the smaller feature amount index of the feature amount table 18, etc. can be considered, The criteria for merging to which direction may be determined as appropriate.

  For example, when merging feature amounts in FIG. 11, the merged feature amounts are stored in the feature amount index n or the feature amount index (n + 1) of the feature amount table 18. Here, when the merged feature value is stored in the feature value index n of the feature value table 18, the values stored in the pointer indexes (m + 1) to (m + 3) of the feature value pointer table 17 are changed from (n + 1) to n. Must be rewritten to

  On the other hand, when the merged feature value is stored in the feature value index (n + 1) of the feature value table 18, the value stored in the pointer index m of the feature value pointer table 17 need only be rewritten from n to (n + 1). .

  Therefore, when merging feature amounts, the feature amount pointer table 17 is updated by storing the merged feature amount in the one where the past number of merges stored in the merge number storage area 20 of the feature amount table 18 is larger. The load to be performed can be further reduced.

Embodiment 2. FIG.
FIG. 12 shows the image data 14 to be processed and the order of reference pixels in the window 16 with respect to the target pixel 15 from A to D. Note that the order of reference pixels other than the target pixel 15 in the first and second embodiments is different.

  FIG. 13 shows a flowchart of feature quantity measurement in the second embodiment. FIG. 13 is a flowchart in which a procedure for referring to a pixel is added to the first embodiment.

  About step S11, step S12, and step S13, when the attention pixel 15 is not 1 (NO), since it is the same as that of Embodiment 1, description is abbreviate | omitted.

  If the target pixel 15 is one pixel 12 in step S13, YES is selected and the process proceeds to step S111.

  When performing feature quantity measurement, in Embodiment 1, in determining whether or not there is a value larger than 0 in the window 16 in addition to the target pixel 15 in step S15, the target pixel 15 is compared with the target pixel 15. In other words, the determination was made with reference to all pixels on the left, upper left, upper and upper right. However, in the second embodiment, in the processing after step S13, the priority of the pixel to be referred to is determined and determined.

  For example, if the pointer index of the upper pixel of the pixel of interest 15 is greater than 0, it is not necessary to refer to another pixel because YES is determined in step S111. The feature values of the upper left pixel and the upper pixel are matched when the upper pixel is the target pixel, and the feature values of the upper pixel and the upper right pixel are matched when the upper right pixel is the target pixel. This is because the feature values of the left pixel and the upper pixel are matched when the left pixel is the target pixel. When it determines with YES by step S111, it progresses to step S18.

  Since the process of step S18 is the same as that of FIG. 4 and FIG. 8 of Embodiment 1, description is abbreviate | omitted. After the process of step S18 is complete | finished, it progresses to step S14 and repeats the loop process of a feature-value measurement.

  If the upper pixel of the target pixel 15 is the zero pixel 13, the process proceeds to step S 112 to determine NO in step S 111.

  In step S112, it is determined whether the left or upper left pixel is greater than zero. If YES, the left pixel or the upper left pixel and the upper right pixel have never been in the same window in the past. Accordingly, there is a possibility that the feature quantities are not matched and may indicate different feature quantities, so it is necessary to check in step S113 and step S115. If the left and upper left are 0 (NO) in step S112, the process proceeds to step S114.

  In step S113, it is determined whether the upper right pixel is greater than zero. If the upper right pixel is 0 (NO), the process proceeds to step S18, and the feature amount / target pixel update is performed so that the left or upper left pixel is matched. Since the process of step S18 is the same as that of FIG. 4 and FIG. 8 of Embodiment 1, description is abbreviate | omitted. After the process of step S18 is complete | finished, it progresses to step S14 and repeats the loop process of a feature-value measurement. If it is determined in step S113 that the upper right pixel is greater than 0 (YES), the process proceeds to step S115.

  In step S115, it is determined whether the feature quantity of the left or upper left pixel is the same as the feature quantity of the upper right pixel. If YES in step S115, the process proceeds to step S18, and the feature amount / target pixel update is performed on the left or upper left pixel. Since the process of step S18 is the same as that of FIG. 4 and FIG. 8 of Embodiment 1, description is abbreviate | omitted. After the process of step S18 is complete | finished, it progresses to step S14 and repeats the loop process of a feature-value measurement. If the feature amount of the left or upper left pixel and the feature amount of the upper right pixel are not the same (NO) in step S115, the process proceeds to step S19. This corresponds to the case where there is a contradiction (YES) in step S17 of the first embodiment.

  In step S19, the feature amount merging process and the feature amount pointer table 17 matching process are performed so that the left or upper left pixel can be matched. Since the process of step S19 is the same as that of FIG. 5, FIG. 6, FIG. 9, FIG. 10 of Embodiment 1, description is abbreviate | omitted. After the process of step S19 is complete | finished, it progresses to step S14 and repeats the loop process of a feature-value measurement.

  In step S112, when the left and upper left are 0 (NO), the process proceeds to step S114. In step S114, it is determined whether or not the upper right pixel is greater than zero. If the upper right pixel is larger than 0 (YES), the process proceeds to step S18, and the feature amount / target pixel update is performed so that the upper right pixel is matched. Since the process of step S18 is the same as that of FIG. 4 and FIG. 8 of Embodiment 1, description is abbreviate | omitted. After the process of step S18 is complete | finished, it progresses to step S14 and repeats the loop process of a feature-value measurement. If the upper right pixel is 0 (NO) in step S114, the process proceeds to step S16.

  In step S16, new feature value creation / target pixel update is performed. Since the process of step S16 is the same as that of FIG. 3 and FIG. 7 in Embodiment 1, description is abbreviate | omitted. Moreover, since the process of step S14 and step S20 is the same as that of Embodiment 1, description is abbreviate | omitted.

  Thus, when creating a new feature amount, it is necessary to confirm that all the pixels other than the target pixel 15 are zero. Accordingly, the target pixel 15 is 1 in step S13, the upper pixel of the target pixel 15 is 0 in step S111, the left and upper left pixels are 0 in step S112, and the upper right in step S114. All decisions must be made that the pixel is zero.

  FIG. 14 shows an example of processing using the flowchart. First, consider Example 1 shown in FIG. Since the upper pixel A is pear with respect to the target pixel 15, NO is selected, and the process proceeds from step S111 to step S112. Next, since the left pixel B and the upper left pixel C are pears, NO is selected, and the process proceeds from step S112 to step S114. Next, since the upper right pixel D is pear, NO is selected, and the process proceeds from step S114 to step S16. In step S16, new feature amount creation / attention pixel update is performed. Therefore, the pixel is referred to four times (A, B, C, D).

  Next, consider Example 2 shown in FIG. Since the upper pixel A is an ant with respect to the target pixel 15, YES is selected, and the process proceeds from step S111 to step S18. In step S18, the feature amount / target pixel update (upper pixel) is performed. Therefore, the pixel is referred to once (A).

  Next, consider Example 3 shown in FIG. Since the upper pixel A is pear with respect to the target pixel 15, NO is selected, and the process proceeds from step S111 to step S112. Next, since the left pixel B is an ant, YES is selected, and the process proceeds from step S112 to step S113. Next, since the upper right pixel D is an ant, YES is selected, and the process proceeds from step S113 to step S115. Next, since the feature amounts of the left pixel and the upper right pixel are not the same, the process proceeds from step S115 to step S19. In step S19, feature amount merging / pointer table matching is performed. Therefore, the pixel is referred to three times (A, B, D).

  Finally, consider Example 4 shown in FIG. Since the upper pixel A is pear with respect to the target pixel 15, NO is selected, and the process proceeds from step S111 to step S112. Next, since the left pixel B is an ant, YES is selected, and the process proceeds from step S112 to step S113. Next, since the upper right pixel D is pear, NO is selected, and the process proceeds from step S113 to step S18. In step S18, the feature amount / target pixel update (left or upper left pixel) is performed. Therefore, the pixel is referred to three times (A, B, D).

  FIG. 15 shows the number of reference pixels excluding the target pixel when the feature amount is measured for the input image 11 of FIG. 1 according to the flowchart of FIG. Note that the numerical values in the square frame shown in FIG. 15 do not represent the assigned pointer indexes shown in FIGS. 7 to 10 and indicate the number of reference pixels as described above.

  From the reference order in FIG. 12 and the flowchart in FIG. 13, it can be seen that there are places where only three reference pixels are required instead of four, and where only one reference pixel is required, as shown in FIG. 15.

  As in the second embodiment, the process of determining the priority of the reference pixel with respect to the target pixel 15 is performed, so that it is not necessary to refer to a pixel that does not need to be referred to. Will improve.

  In order to describe the embodiment of the present invention, the area, that is, the number of pixels of the connected component, was used as the feature quantity to be measured. However, the feature quantity to be measured need not be limited to the area. For example, it is good also as a feature-value which measures the gravity center of a connection component.

  In that case, in addition to adding 1 to the area in step S32 shown in FIG. 3 in the process of step S16 for creating a new feature amount, the process of adding the X coordinate and the Y coordinate of the target pixel 15 is performed. Do.

  In addition, in step S18 shown in FIG. 4 in the process of step S18 for performing the feature amount / target pixel update process, the process of adding the X coordinate and the Y coordinate of the target pixel 15 is performed in addition to the process of incrementing the area.

  By processing in this way and taking the average of the number of connected component pixels obtained by adding the X and Y coordinates, the center of gravity of the connected component pixels can be measured as a feature amount.

  Note that the method described in the first and second embodiments is performed by, for example, a CPU (Central Processing Unit). Data can be stored using a storage element such as a RAM (Random Access Memory).

It is a figure which shows the data structure and image scanning direction of the feature-value measuring method in Embodiment 1 of this invention. It is a figure which shows the flowchart of the feature-value measurement in Embodiment 1 of this invention. It is a figure which shows the flowchart of step S16 (new feature-value production and attention pixel update) in Embodiment 1 of this invention. It is a figure which shows the flowchart of step S18 (feature amount and attention pixel update) in Embodiment 1 of this invention. It is a figure which shows the flowchart of step S19 (feature amount merge process) in Embodiment 1 of this invention. It is a figure which shows the flowchart of step S19 (matching process of a feature-value pointer table) in Embodiment 1 of this invention. It is the figure which showed a series of processes of step S13, step S15, and step S16 among the feature-value measurement in Embodiment 1 of this invention. It is the figure which showed a series of processes of step S13, step S15, step S17, and step S18 among the feature-value measurement in Embodiment 1 of this invention. It is the figure which showed step S13, step S15, step S17, and step S19 among the feature-value measurement in Embodiment 1 of this invention. It is the figure which showed step S13, step S15, step S17, and step S19 among the feature-value measurement in Embodiment 1 of this invention. It is the figure which showed the example of the reference | standard which performs a merge process among the feature-value measurement in Embodiment 1 of this invention. It is a figure which shows the order of the pixel referred in a window with respect to the image data and the attention pixel which perform the process in Embodiment 2 of this invention. It is a figure which shows the flowchart of the feature-value measurement in Embodiment 2 of this invention. It is the figure which showed the example of a process using the flowchart in Embodiment 2 of this invention. It is the figure which showed the reference pixel number except the attention pixel at the time of measuring the feature-value in Embodiment 2 of this invention.

Explanation of symbols

11. 13. pixel 13.0 of input image 12.1; Image data 15. Pixel of interest 16. Window 17. Feature amount pointer table 18. Feature amount table 19. Feature amount storage area 20. Merge count storage area

Claims (7)

In the feature quantity measurement method for measuring the feature quantity of an image,
A feature quantity table for storing the feature quantity and the number of merges of the feature quantities;
A feature amount pointer table for storing a feature amount index indicating a storage location of the feature amount and the number of merges stored in the feature amount table;
Use
Refer to the pixel in the window including the target pixel for which the feature amount measurement is performed, and if there is no pixel assigned a pointer index indicating the data storage location of the feature amount pointer table in the window other than the target pixel, new A first step of creating a feature quantity and storing it in an unused area of the feature quantity table, and storing a feature quantity index indicating the new feature quantity in the feature quantity pointer table;
When there is a pixel assigned a pointer index in the window other than the pixel of interest with reference to a pixel in the window including the pixel of interest, and the pixel assigned the pointer index is consistent, a feature amount pointer A second step of incrementing the feature value indicated by the feature value index stored in the pointer index of the table;
If there is a pixel assigned a pointer index in the window other than the pixel of interest with reference to a pixel in the window including the pixel of interest, and there is a contradiction in the pixel assigned the pointer index, the feature amount Stores the feature quantity that is the result of merging the corresponding feature quantity in the table and the corrected number of merges, and merges the merged feature quantity index among the feature quantity indexes stored in the pointer index of the feature quantity pointer table A third step of rewriting to the feature amount index and indicating the feature amount after merging;
A feature quantity measuring method characterized by comprising:   The window includes the target pixel and left, upper left, upper, and upper right pixels with respect to the target pixel, and measures a feature amount by referring to at least one of the left, upper left, upper, and upper right pixels. The method of measuring a feature quantity according to claim 1. In the feature quantity measurement method for measuring the feature quantity of an image,
A feature quantity table for storing the feature quantity and the number of merges of the feature quantities;
A feature amount pointer table for storing a feature amount index indicating a storage location of the feature amount and the number of merges stored in the feature amount table;
Use
In the first to third steps, if there is no contradiction in the feature quantity of the pixel in the window with respect to the target pixel in the feature quantity measurement up to one pixel before the target pixel, one of the consistent pixels 3. A feature quantity is measured with reference to one, and a new feature quantity is created when the point index is not assigned to all pixels other than the target pixel in the window. The characteristic amount measuring method described.   4. The feature amount merging includes storing a merge result in a feature amount having a larger number of merges out of the number of merges stored in the feature amount table. The characteristic amount measuring method according to claim 1.   5. The feature amount measuring method according to claim 1, wherein the feature amount is an area of a connected component of a pixel indicating 1 in a binary image indicating 0/1. .   5. The feature amount measuring method according to claim 1, wherein the feature amount is a centroid of a connected component of a pixel indicating 1 in a binary image indicating 0/1. . A storage element for storing data to be stored in the feature quantity table and the feature quantity pointer table;
A central processing unit for performing the third process from the first process;
The feature amount measuring apparatus according to claim 1, further comprising:

Images