KR940011699B1 - Method of extracting the image contour from binary image signal - Google Patents

Abstract

To obtain the image of characters more clearly and to sample it faster, the method that samples the outline of binary image effectively is suggested. The sampling process suggested in this invent is as follows. First, an expansion operator window is installed to four directions around the focal pixel after scanning the whole area of the binary image once, and then the outline points are defined in four directions. Next, the final outline points are defined again with consideration of the interrelation among the neighbor pixels. As the result, the outline can be sampled fast because the sampling is completed with one-time scanning, and the distinct outline can be obtained because the outline is sampled to the size of +1 larger than the pixel thickness of the character.

Description

Contour Extraction Method of Binary Image

1 is a 3x3 window defined in conventional gray image and contour extraction.

2 is a weighted 3 × 3 window mask.

3 is a 3 × 3 window weighted in the X direction.

4 is a 3 × 3 window weighted in the Y direction.

5 is an example of a general image to extract the contour.

6 is an X-direction contour extraction result of the image of FIG.

7 is a result of contour extraction in the Y direction in the image of FIG.

8 is a result of extracting the final contour from the image of FIG.

9 is a flowchart of a conventional horizontal contour extraction method.

10 is a flowchart of a conventional vertical contour extraction method.

11 is an example of a conventional binary image contour tracking method.

12 is an example of an eight-way code in a conventional binary image contour tracking method.

13 (a) and 13 (b) are examples of character outlines extracted by a conventional binary image contour extraction method.

14 (a) and 14 (b) are examples of character outlines extracted by the contour extraction method of the present invention.

Four neighboring operator window defined in the contour extraction method of the present invention of FIG.

16 is an extended operator window defined in the contour extraction method of the present invention.

17 is an example of contour extraction of the present invention.

18 is a block diagram of a document recognition apparatus in which the contour extraction method of the present invention is executed.

* Explanation of symbols for main parts of the drawings

p: reference point p0, p2, p4, p6: 1st outer point

r0, r2, r4, r6: 2nd outer survey point 2: hoist processor

3: video display unit 4: host interface

5: auxiliary memory device 6: scanner interface

7: scanner 8: buffer

9: address decoder 10: data memory

11: digital signal processor

The present invention relates to a method for extracting a contour of a binary image. In particular, an extension operator window is defined for four directions around a center pixel, and the contour points are first defined in four directions. By defining the final contour point from the correlation with the adjacent pixels, it is possible to extract the binary image of the characters clearly and at high speed.

Conventional image contour extraction methods are roughly divided into a contour extraction method of a gray image (an image having a monochrome grayscale of 16 to 256 gradations) and a binary image extraction method.

Contour extraction of the gray image defines 3 × 3 windows as shown in FIG. 1, and defines 3 × 3 window masks by assigning weights (X1-X9) to each part (9 rooms) as shown in FIG. The contour is extracted through the calculation process of FIGS. 9 and 10 with the weights defined in each direction as shown in FIG. 3 in the X direction (horizontal) and in FIG. 4 in the Y direction (vertical).

The window of FIG. 3 is for extracting the contour in the X direction, and the window of FIG. 4 is for extracting the contour in the Y direction, calculated from the weight of FIG. 3 for the X direction, GX = (X7 + 2X8 + X9)- Perform (X1 + X2 + X3) and perform calculation from the weight of FIG. 4 for the Y direction, and perform GY = (X3 + 2X6 + X9)-(X1 + 2X4 + X7).

First, contour extraction in the X-direction divides the contour of an image having a gray image level (e.g., 256 gray levels = 256 gray levels) into two stages, black (1) and white (0), as shown in FIG. The threshold value for the image is determined, and the pixel of the coordinate (i, j) to extract the contour in the horizontal / vertical (i, j) pixel area of the image corresponds to the position of the window (X5), and GX operation is performed to calculate the threshold value. If it is above, the pixel is defined as black (X5 = 1), and if the pixel is defined as white (X5 = 0), the contour of the X direction is extracted.

As an example of the result of contour extraction of the gray image in the X direction, if the image of FIG. 5 is contour-extracted in the X direction, the contour as shown in FIG. 6 is extracted.

On the other hand, in the contour extraction in the Y-direction, as shown in FIG. 10, the pixel having the coordinates (i, j) to extract the contour in the horizontal / vertical (i, j) pixel area of the image following the determination of the threshold is positioned at the window X5 position. After the correspondence, GY operation is performed, if the result is greater than or equal to the threshold, the pixel is defined as black (X5 = 1), and if the pixel is defined as white (X5 = 0), the contour of the Y direction is extracted.

As an example of the result of contour extraction in the Y direction as a result of the contour image extraction in the Y direction as shown in FIG.

Thereafter, when ORing the contour in the X direction and the contour in the Y direction, the contour of the gray image is finally extracted as shown in FIG. 8.

On the other hand, the contour extraction method of the binary image uses the contour tracking method as shown in FIG.

That is, each droplet of the 3 × 3 window defined in FIGS. 1 through 4 is divided into eight directions from 0 to 7 and each component is traced as represented by the arrow of FIG. 11 around the reference point p. To extract the contour.

13 (a) and 13 (b) are examples of outlines (characters) extracted by such a Y-direction tracking method. In the figure, the black hatched pixels represent the extracted contours, and the white pixels therein are original binary images.

As described above, the existing contour extraction methods mostly provide a method for extracting a contour from a gray image, and the method of extracting a contour from a binary image mostly depends on the contour tracking method.

Therefore, if the contour extraction method of gray image is applied to the contour extraction method of binary image, unnecessary computation routines are repeatedly performed, which takes a lot of time to extract the contour. In the contour tracking method, the contour is moved by moving 3 × 3 windows one by one. You have to take the inconvenience of detecting it.

In addition, since the contour tracking method does not consider the thickness of the character, when the thickness of the character is 2 pixels or less, if the contour is extracted, it is difficult to find an accurate feature in a later recognition step.

Fig. 13 (a) shows an example of this, and since the horizontal stroke of the letter A is shown to be 2 pixels thick ("F"), it is difficult to find an accurate tracking path (arrow path in Fig. 11) at the next contour tracking.

This is because accurate contour tracking is possible only when the opposite contours are separated from each other by at least one pixel.

In addition, since the conventional contour tracking method does not consider correlation with adjacent pixels (pixels), in the image having depressions F as in the case of the English letter W shown in FIG. Filling with contour pixels (black) prevented accurate contour extraction.

In addition, in the conventional contour tracking method, when an image is composed of several separated images, for example, in the case of an English letter as shown in FIG. This is possible.

That is, since the process of re-searching the image must be repeated from the initial stage in order to search the separated image, there is a problem in that the contour extraction time becomes longer.

The present invention extracts the contour through one search for the entire area of the binary image, defines the extension operator window in four directions around the center pixel, and defines the contour point in four directions, and then the adjacent pixels. We determine whether the adjacent pixels are already defined as contours, and define the final contour point based on the result of this determination, shortening the contour extraction time and extracting the correct contours even when the image thickness is 2 pixels or less. It is an object of the present invention to provide a method for extracting a contour of a binary image that enables the extraction of a stable feature. Hereinafter, the contour extraction process according to the present invention will be described with reference to FIGS. 14 to 18.

First, in the present invention, a 4-neighbor window consisting of 3x3 windows as shown in FIG. 15 is defined as a basic operator for defining an outline.

The four-neighbor window defines the survey points (p0, p2, p4, p6) in four directions (right, top, left, and bottom) around the current survey point (p), and each outer survey point (po, The extended operator window is defined by defining extended search points r0, r2, r4 and r6 as shown in FIG. 16 outside p2, p4 and p6).

In this way, the extended operator window is defined, and the first step is to search the entire area of the character image and search for the pixel whose pixel value is "1".

In this step, the pixel value of "0" is excluded from the investigation.

In the second step, the current survey point is defined as the reference survey point (p), and the survey points (p0, r1, p2, r2, p4, r4 p6, r6) are defined in four directions around the reference survey point (p). do.

Subsequently, in a third step, it is examined whether at least one of the first outer irradiation points p0, p2, p4, and p6 is "0", and all of the irradiation points p0, p2, p4 and p6 are " If it is 1 ", it repeats from a 1st step, and if at least 1 irradiation point is" 0 "among irradiation points p0, p2, p4, and p6, the secondary secondary irradiation point r0, r2, r4 which is adjacent in each case , the final contour point is selected according to whether r6) has already been determined as the contour point.

This step consists of a pair of adjacent survey points (p0, r0), (p2, r2), (the first outer survey point (p0, p2, p4, p6) and the secondary outer survey points (r0, r2, r4, r6). The correlation points are determined by p4, r4) and (p6, r6), and the final contour points are determined after determining the contour points that are the primary contour candidates.

That is, in the third step, when the first outer irradiation point p0 is "0", the fourth operation is to detect the right contour, and the second outer irradiation point r0 adjacent to the first outer irradiation point p is If it is already determined as the contour point, and if the determination result is determined as the contour point, the reference survey point (p) is selected as the final contour point, and the secondary outer survey point (r0) is not determined as the contour point. In the following, the first outer survey point (p0) is selected as the final contour point.

On the other hand, when the first outer irradiation point (p2) is "0" in the third step is a fifth step of detecting the upper contour, the secondary outer irradiation point (r2) adjacent to the primary outer irradiation point (p2) is If it is already determined as the contour point, and if the determination result is determined as the contour point, the reference survey point (p) is selected as the final contour point, and the secondary outer survey point (r2) is not determined as the contour point. In the following, the first outer survey point (p2) is selected as the final contour point.

On the other hand, in the third step, when the first outer irradiation point p4 is "0", the sixth step of detecting the left contour is a sixth step, and the secondary outer irradiation point r4 adjacent to the primary outer irradiation point p4 is If it is already determined as the contour point, and if the determination result is determined as the contour point, the reference survey point (p) is selected as the final contour point, and the secondary outer survey point (r4) is not determined as the contour point. In the following, the first outer survey point (p4) is selected as the final contour point.

On the other hand, in the third step, if the primary outer irradiation point (p6) is "0" as the seventh step of detecting the lower contour, the secondary outer irradiation point (r6) adjacent to the primary outer irradiation point (p6). If it is determined that the contour point is already determined, and if the determination result is determined as the contour point, the reference survey point p is selected as the final contour point, and the secondary outer survey point r6 is not determined as the contour point. In this case, the first outer survey point (p6) is selected as the final contour point.

By selecting the final contour point through the process as described above, it is possible to ensure the sharpness of the letter recessed portion and to extract the correct contour regardless of the pixel thickness.

An example of the character image from which the outline is extracted is the outline of the English letters A and W shown in FIGS. 14 (a) and 14 (b).

That is, the present invention compares the outermost irradiation points r0, r2, r4, and r6 of the extension operator window to the conventional contour extraction method in which the unclear contour is extracted when the pixel thickness is 2 pixels or less as shown in FIG. 13 (a). By selecting the final contour point by comparing it once more, as shown in Article 14 (a), the contour (or binary image) having a thickness of 2 pixels or less is extracted to a size of +1 larger than the thickness of the character, so that accurate contour extraction is performed. It is possible.

In addition, according to the conventional contour extraction method, as shown in FIG. 13 (b), the depression F is filled with the contour point and becomes unclear, whereas the present invention provides the outermost irradiation points r0, r2, r4, r6 of the extended operator window. By comparing the one more time and selecting the final contour point, as shown in FIG. 14 (b), the depressions are not clearly filled with the contour points and are clearly distinguished.

17 shows an example of such contour extraction, in which 1 represents a black pixel of a character, 0 represents a white pixel, and 2 represents a pixel already extracted as an outline.

If the current h position is defined as the reference survey point p of the extended operator window and the above processes of the first to seventh steps are performed, since the first irradiation point p4 (g position) on the left is “0” The sixth step is applied.

In this case, the reference survey point p = 1 and the secondary outline search point r4 (f position) adjacent to the primary outline point p4 (g position) is already determined as the contour point as r4 = 2. The reference point p is selected as the contour point, and the h position selected as the contour point is spaced apart by one pixel from the f position, so that the depression is not filled and is clearly distinguished as shown in FIG. 14 (b).

18 is a circuit diagram of a general document recognizing apparatus for carrying out the present invention, and includes a keyboard unit for directly inputting a command for contour extraction to the host processor 2 having an execution operation key for contour extraction. 1), a hoist process 2 for performing control of the entire apparatus, and a video display unit 3 for displaying contours of characters extracted by the digital signal processor 11 under the control of the hoist processor 2; ), A host interface 4 connecting the host processor 2 and each of the systems 5, 6, 8, 9, and 11, and data associated with device control by connecting to the host interface 4; A secondary storage device 5 to be stored, a scanner interface 6 that is responsible for a data input interface between the system and the scanner 7, a scanner 7 that reads an input image as a binary image, and the interface 6 ) In order to store the image information read into the scanner 7 or the contour extraction information extracted by the digital signal processor 11 in the data memory 10, an address for designating a buffer 8 and a memory address in charge of data input / output A decoder 9, a data memory 10 in which scanned image information and contour extraction information are stored and output through the buffer 8, and an individual character image stored in the data memory 10 is buffered. The digital signal processor 11 includes a digital signal processor 11 that reads through the digital signal, extracts a feature including an outline, and stores the contour in the data memory 10 through the buffer 8.

The contour extraction operation performed by the apparatus configured as described above is performed by the first to seventh steps, and when the user inputs a key command for instructing contour extraction by manipulating the keyboard unit 1, the host processor 2 Recognizes this and outputs data, address, strobe signals (D, A, DS, IS, MS) through the host interface 4, and the scanner interface 6, buffer 8, address decoder 9, The digital signal processor 11 is controlled, and the data necessary for the control is stored in the auxiliary memory device 5.

First, enable and strobe signals ENB and ST are output to scan a document with a scanner 7 to read a binary image, and the scanned image information is passed through the data output terminal D of the scanner interface 6. The buffer 8 stores the data at the predetermined address A of the data memory 10 designated by the address decoder 9.

When the binary image data is stored in the data memory 10, the digital signal processor 11 reads the individual character image information stored in the data memory 10 through the buffer 8 again and the digital signal processor 11 reads the second image data. A binary image contour extraction process of steps 1 to 7 is executed, and the contour information extracted as a result of the execution is stored in the data memory 10 through the buffer 8 again.

In this way, the contour image information stored in the data memory 10 can be read by the host processor 2 through the interface 4, the buffer 8, and the address decoder 9 and displayed on the video display unit 3. FIG.

As described above, according to the present invention, since the scanning is completed in one time by the conventional contour tracking method, the contour extraction can be performed at high speed, and the thickness of the image is extracted by extracting the contour by the size of +1 than the pixel thickness of the character. Even if the size is less than 2 pixels, accurate and clear outline extraction is possible, and especially when applied to a character recognition system, it is effective to extract accurate and stable features.

Claims (8)

The current pixel from which the contour of the image is to be extracted is the reference irradiation point p, and the first outer irradiation point p0 corresponding to the pixels in four directions (right, upper, left, and bottom) adjacent to the reference irradiation point p. , p2, p4, p6) and the secondary outer survey points (r0, r2, r4, r6) corresponding to the adjacent pixels are defined to define the extended operator window, and the extended operator window corresponds to the current pixel A method for extracting the contour of a binary image, characterized in that the final contour point is selected from the survey points (p0, p2, p4, p6) by comparing and verifying the correlation. The method of claim 1, wherein the correlation with the adjacent irradiation point is whether the first outer irradiation point (p0, p2, p4, p6) and the adjacent secondary irradiation point (r0, r2, r4, r6) are already selected as contour points. Contour extraction method of a binary image, characterized in that determined by. The method of claim 1, wherein the comparative verification of adjacent irradiation points is defined in the same direction and pairs (p0, p2, p4, p6, r0, r2, r4, r6) adjacent to each other. r0), (p2, r2), (p4, r4), and (p6, r6), respectively, to determine the final contour. The method of claim 3, wherein the primary outline point (p0, p2) with respect to the pair of primary and secondary outline points (p0, r0), (p2, r2), (p4, r4), (p6, r6) , p4, p6) is the case of white pixel (pixel value = 0), and the secondary paired with the reference point (p) or the first outer point (p0, p2, p4, p6) according to the result A method for extracting the contour of a binary image, characterized by selecting a final contour point from the outer survey points r0, r2, r4, and r6. The method according to claim 4, wherein when the primary outer survey point p0 on the right side of the reference survey point p is a department, it is determined whether the secondary outer survey point r0 paired with this is already selected as the contour point. If the secondary outline point (r0) is a contour point, the reference point (p) is selected as the outline point. If the secondary outline point (r0) is not selected as the contour point, the primary outline point (p0) is selected. Contour extraction method of a binary image, characterized in that the selection as the contour point. The method according to claim 4, wherein when the primary outer survey point p2 above the reference survey point p is a department, it is determined whether the secondary outer survey point r2 paired with this is already selected as the contour point. If the secondary outline point (r2) is a contour point, the reference point (p) is selected as the outline point. If the secondary outline point (r2) is not selected as the outline point, the primary outline point (p2) is selected. Contour extraction method of a binary image, characterized in that the selection as the contour point. The method according to claim 4, wherein when the first outer survey point p4 on the left side of the reference survey point p is a white pixel, it is determined whether the secondary outer survey point r4 paired with this is already selected as an outline point. If the secondary outline point (r4) is a contour point, the reference point (p) is selected as the outline point, and if the secondary outline point (r4) is not selected as the outline point, the primary outline point (p4) is selected. Contour extraction method of a binary image, characterized in that the selection as the contour point. The method according to claim 4, wherein if the primary outer survey point (p6) on the left side of the reference survey point (p) is a departmental pixel, it is determined whether the secondary outer survey point (r6) paired with this is already selected as the contour point. If the secondary outline point (r6) is a contour point, the reference point (p) is selected as the outline point, and if the secondary outline point (r6) is not selected as the outline point, the primary outline point (p6) is selected. Contour extraction method of a binary image, characterized in that the selection as the contour point.