JPS62114077A - Picture element feature extracting device - Google Patents

Abstract

PURPOSE:To extract a feature without being influenced by various deformations of a graphic and a noise, by deciding a notice area or plural sensitive areas to be white or black, and outputting a result of decision, as a picture element feature code to a notice picture element. CONSTITUTION:An image which has been converted to a binary signal by an image input device 1 is stored in a binary image buffer memory 3. A picture element feature extractor 4 reads out a picture element in a notice area 8 containing a notice picture element 7 from the binary image buffer memory 3, and a picture element in plural sensitive areas 10a-10h provided on the periphery of a dead zone area 9 which is in the periphery of the notice area 8 and also not related to a picture element feature code, compares the number of black picture elements contained in each area, and a prescribed threshold value which has been set to a threshold value setting device 5, at every area of the notice area 8 and the sensitive areas 10a-10h, decides '1' or '0' at every area, and its result is arranged in order of the areas 10a, 10b, 10c, 10d, 8, 10e, 10f, 10g, and 10h, and outputted as a picture element feature code.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Application in Industry) The present invention relates to a pixel feature extraction device for a device that recognizes characters and figures.

(Prior art) Conventionally, this type of device thins a figure, and then divides the thinned figure into eight pixels around the pixel of interest and A as shown in FIGS. 2(a) to (d). Features were extracted using a mask defined by . Furthermore, when line thinning is not performed, features are extracted using a mask with a wide field of view as shown in FIG. 2(e).

(Problem to be solved by the invention) However, when line thinning is performed, it is not always possible to extract features stably due to whiskers, etc. that occur along with this, and there are cases where line thinning is not performed. However, it has not always been possible to perform sufficient feature extraction due to various deformations of the figure, including broken line segments, protrusion, unevenness, etc., and noise. Furthermore, when a drawing or the like in which figures of different sizes coexist is to be recognized, there is a problem in that it is difficult to extract and cover features using the same method.

The present invention provides a pixel feature extraction device that is capable of extracting basic features that are not affected by noise or various deformations of figures including line segment breaks, unevenness, etc., and is not limited to recognition targets. With the goal.

(Means for Solving the Problems) In order to solve the above problems, the present invention analyzes the states of each pixel (target pixel) in a binarized image and the surrounding pixels. In a pixel feature extraction device that extracts the feature of each pixel by detecting it, two image buffer memories for temporarily storing the binarized image and the two image buffer 7 memories set in advance are used. A sensing region including at least a pixel in a region of interest including the pixel of interest before -b and at least one pixel set in advance with a dead zone region not involved in feature extraction surrounding the region of interest. means for selectively reading out pixels, and determining that the region of interest or the plurality of sensing regions is white or black depending on the black and white state of the pixel read from the region of interest or the plurality of sensing regions; and a pixel feature extractor serving as a means for outputting determination results for the region of interest and the entire sensing region as a pixel feature code for the pixel of interest.

(Function) According to the present invention, the pixels in the width direction of a line segment with a large line width are canceled by the dead zone area, the same features as those of a line segment with a small line width are extracted, and the lines with breaks are When the break belongs to the region of interest or the sensing area, the break is detected as being interrupted, and when the break belongs to the dead zone area, it is detected as being continuous, but in the case of the end of the line segment, in either case, A break is detected, and it is determined whether the line segment is broken or not.

(Embodiment) FIG. 1 is a functional block diagram showing an embodiment of the present invention. In the figure, 1 is an image input device, 2 is a pixel feature extraction device, and the pixel feature extraction device 2 has two image buffer memories 3.
, a pixel feature extractor 4, and a threshold value setter 5.

The image input device 1 is a well-known optical reading device equipped with, for example, a light source, a photoelectric conversion element, etc., and converts images to be recognized, such as characters and figures, into binary symbols. The 2-image buffer memory 3 stores the image converted into the 2-digit code (hereinafter referred to as 2).
It is called an individual image. ) has the necessary capacity to store
2IIT] image output from the image input device 1 is temporarily stored. The pixel feature extractor 4 extracts from the two image buffer memories 3 pixels in a region of interest including at least a preset pixel of interest, and special features around the region of interest.
4 - means for selectively reading out pixels in a plurality of sensing regions including at least one pixel set in advance across a dead zone region not involved in feature extraction; Determining the region of interest or the plurality of sensing regions as white or black according to the black and white state of the pixels read out,
This constitutes a means for outputting the determination results for the region of interest and all 1 sensing regions as a pixel characteristic for the pixel of interest. Specifically, the pixel is read out according to the mask of the present invention, which will be described later, and a preset threshold value is output. The setting device 50 determines whether the pixel is white or black according to the threshold value, and outputs a pixel characteristic code for each pixel.

FIG. 3(a) shows a mask used in this embodiment.
a region of interest 8 that includes n pixel features around the region of interest 8; a dead zone region 9 that does not involve the
A plurality of sensing areas 1iJ110a, 10 provided around the
b, 10c, 10d.

It consists of 10e', 10f, 100, and 10h.

Here, the region of interest 8 and the sensing regions 10a to 10h each consist of 9 pixels (3 pixels x 3 pixels), and the dead zone area 181i9 has a width of 3 pixels, and the entire mask 6 has 15 pixels x 15 pixels. It is made up of 225 pixels.

The pixel feature extractor 4 extracts the number of black pixels included in each region and a predetermined value set in the threshold value setter 5 for each of the target region 8 of the mask 6 and the sensing regions 10a to 10h. Compare with the threshold value and set "1" or "0-1" for each area.
As shown in FIG. 3(b), the region t
dlOa, 10b.

10c, 10d, 8.10e, 10f, 100°10h
The pixel features are arranged in the order of pixel features]-11.

FIG. 3(C) shows each output of the pixel feature extractor 4 described above based on the positional relationship shown in FIG. 3(a), and below, this is also shown in FIG. 3(b). Similar to the signal string shown in , this is called a pixel feature code 12.

FIG. 4 shows how the mask 6 absorbs variations in line width. Here, the image has a resolution of 8 lines/mm, that is, the area per pixel is 0125 mm x
If it is 0.125m and the threshold value is "3", then the line width is 0.
．． The line segment 13 of 125# is shown in Figure 4(a).
I area 8. When intersecting the sensing region 10d and the sensing region 10e, three pixels in each of the regions become black pixels, and a pixel characteristic code 14 of "rooolloooo" without a horizontal line segment can be obtained. Also, as shown in FIG. 4 (1), a line segment 15 with a line width of 1.125 m is located in the area of interest 8. When intersecting the sensing region 10d and the sensing region 10e, all pixels, that is, 9 pixels, in each of the regions are black pixels, but the pixel feature representing a horizontal line segment can be set to 1 in the same manner as described above.

FIG. 5 shows an example of image feature extraction by the pixel feature extractor 4, in which the mask 6 scans an angled binary image -L having a line width of two pixels in the left direction. The following shows the pixel characteristics obtained at this time. FIG. 5(a) is a binary image @attention area 8. Sensing area 10d.

10e and 10f, and in this case, the pixel feature code is converted to 17 of roool 11100J. In addition, Fig. 5(b) shows the scanning position as shown in Fig. 5(a).
This shows the case of moving one pixel to , in this case I'0O
Pixel characteristics of O011100J]-18 are obtained, and FIG. 5(C) shows the case where the scanning position is further moved by 4 pixels with respect to FIG. 5(b), and in this case roo('
)011010'l pixel feature code 19 is obtained.

FIG. 6 shows a pattern 20 based on the pixel characteristic code obtained when the entire top of the 2-candy image having a line width of 2 pixels is scanned, and the shaded area 21 indicates that the area of interest is a black pixel. Indicates the judged part. According to the pattern 20, it can be seen that a pixel feature representing an angle shape is extracted approximately at the center.

FIG. 7 shows another example of image feature extraction by the pixel feature extractor 4. FIG. 7(a) shows the input figure, and does not show the figure 22 in which a line segment of about 2rrvn is connected to one end of a circle with a diameter of about 2m. FIG. 7(b) shows a pattern 23 based on a pixel characteristic code in which the region of interest 8 is only black pixels, and a pixel characteristic code 271 indicating a horizontal Y-shaped three-branch is present at the connecting part of a circle and a line segment. It can be seen that C is extracted. FIG. 7(C) is a pattern 25 based on pixel characteristic codes of all pixels.
It can be seen that on the binary image, a pixel characteristic code 26 indicating that a white pixel is surrounded by black pixels is extracted. In addition, Fig. 7 (b
) (c), the pattern 23. is based on the pixel feature code.
The reason why 25 is flat is due to the nature of the printing press.
It is.

FIG. 8 shows still another example of image feature extraction by the pixel feature extractor 4.
7 is shown when scanning.

FIG. 8(a) The tj line segment 27 is the sensing area 10d and 1
Indicates the case where 0e intersects, in this case rool
A pixel feature code 28 of 01000J is obtained. Further, FIG. 8(11) shows a case where the pixel is moved by one pixel with respect to FIG. 8(a), and in this case, a pixel characteristic code 29 of [0O0111000J is obtained. Here, at the end of a line segment or between different line segments, the previous pixel feature ]-do 28 and the pixel time Wi]-do 29 cannot be 1q, so when the pixel feature code 28 is obtained, By referring to the pixel feature codes before and after the pixel feature code, it is possible to determine whether or not the line segment is broken.

FIG. 9 shows a specific embodiment of the pixel feature extraction device of the present invention (region of interest 13×3 pixels, dead zone width 3 pixels, sensing region 3×3 pixels).

The image input device 1 includes a one-dimensional sensor 110 and a binarizer 12
0, and a scanning circuit (not shown) that scans the one-dimensional sensor 110 in the main scanning direction X and the sub-scanning direction Y with respect to the input drawing 130. 2 image buffer memory 3
is an input buffer 310 in which shift registers having the same number of bits as the number of bits in the main scanning direction A shift register having the same number of bits as the number of bits of x is made up of a code buffer 320 connected to the target area, the dead zone area, and the same number of columns as the number of pips 1 to 1 in the sub-scanning direction Y of the sensing area. The pixel feature extractor 4 includes a black and white judgment register 410 having a shift 1 register configuration having the sizes of the region of interest and the sensing region, a black pixel number counter 420, a comparator 430, and the size of the region of interest, the dead zone region, and the sensing region. It consists of three shift registers having the same number of bits as the number of bits in the main scanning direction X.

In the above configuration, the input drawing 130 is scanned by the one-dimensional sensor 110 in the main scanning direction is input. The contents of the manual buffer 310 are sent to a black-and-white determination register 410 that is connected independently for each column Y01 in the sub-scanning direction, and has the same effect as when the black-and-white determination register 410 scans the input drawing 130. The individual bits of the black/white determination register old 0, ie, the data representing whether the pixel is black or white, are sent to the black pixel number counter 420, and the number of black pixels is counted. The output of the black pixel number counter 420 is compared with the value of the threshold value setter 5 having a predetermined value set in advance by a comparator 430, and a signal of “1” or “0” is outputted to the code buffer 320. Therefore, the output of the comparator 430 corresponds to each bit of the pixel feature code being output in sequence. Also, from the code buffer 320, pixel features are extracted from the shift register corresponding to the center position in the sub-scanning direction Y of the region of interest and each sensing region]
- outputs a signal for code conversion to the -code conversion register 440. ]- code conversion register 440 sets the bit corresponding to the center point of the region of interest and each sensing area as a pixel feature ]- code τ
Output.

FIG. 10 illustrates pixel feature codes representing only the features of the partial figures of the input drawing and the partial figures extracted by the apparatus of the present invention. FIG. 10(a) shows a figure 30 representing an end point and the corresponding rooool 10
A pixel characteristic code 31 of 00J is shown. Figure 10 (
b) shows the figure 32 representing the refraction point and the corresponding “0”
00011010” pixel feature code 33 and Mll
The pixel feature code 34 of "LOOLOO" is shown.

FIG. 10(C) shows a figure 35 representing a branch point and the corresponding pixel feature code 36 of rol 0011010J.
, pixel feature code 37 of rl 00100111J and pixel feature code 38 of rl 11100100J. FIG. 10(d) shows a figure 39 representing another branch point and the corresponding pixel characteristic code 40 of [100001100J and pixel characteristic code 41 of rl 00011100J. In addition, FIG. 10(e) shows a figure 42 representing an intersection and the corresponding ro101.
11010] pixel feature code 43, rool 001
111” pixel characteristics] -do 44, rl 0010011
1'' pixel feature code 45, rllloolool'' pixel feature code 46, and rl 11100100''
The pixel feature code 47 is shown.

(Effects of the Invention) As explained above, according to the present invention, a dead zone area is set around a region of interest, a sensing area is set around the dead zone area, and features of a partial figure are inputted using a small number of bits. Since it is possible to extract corresponding to all pixels of the image, it is stable even against broken lines, unevenness, etc., and it is possible to extract features that are very effective for practical implementation of figure recognition devices, etc. be able to. Further, by using only the features of the partial figure corresponding to the figure to be recognized, it is also possible to detect a desired feature point such as a refraction point or one branch point. Furthermore, a region of interest and a dead zone region according to the nature of the input image.

Image identification by setting the sensing area, threshold, etc.
It has the advantage that it can also be applied to analysis equipment.

[Brief explanation of drawings]

FIG. 1 is a functional block diagram showing an embodiment of the pixel feature extraction device of the present invention, FIGS. 2(a) to (e) are explanatory diagrams showing conventional feature extraction masks, and FIG. 3(a) is Explanatory diagrams showing an example of the mask of the present invention, FIGS. 3(b) and (C); Figures 4(a) and 4(b) showing the contents of the pixel characteristic code of the present invention;
) is a diagram showing that line segments with different line widths are expressed by the same pixel feature code, Figures 5 (a), (b), and (c) are diagrams showing an example of dimensions for feature extraction, and Figure 6 is a diagram showing that line segments with different line widths are expressed by the same pixel feature code. A diagram showing an example of a pattern based on a pixel feature code, FIG. 7(a) (bl
(cl is a diagram showing another example of a pattern based on a pixel feature code, Figure 8 is a diagram showing another example of feature extraction, Figure 9 is a diagram showing another example of a pattern using pixel feature codes,
The figure is a block diagram showing an example of a specific configuration of the pixel feature extraction device, and FIGS. 10(a) to 10(0) are J diagrams showing still other examples of feature extraction. 2... Pixel feature extractor, 3... Binary pixel buffer memory, 4... Pixel feature extractor, 5... Threshold value setter, 6... Mask. Patent Applicant: Oki Electric Industry Co., Ltd. Patent Attorney Furu 1) Sei Takashi Is this invention actually sold?
d) (e) Figure 2 (a) (b) 25 (C) Diagram showing the conventional characteristic cti-derived mask (a) (b) Figure 1 of the balun's Ise (a) (b) Ike 11 shows l Figure 8

Claims (2)

[Claims] (1) In a pixel feature extraction device that extracts features for each pixel by detecting the states of the pixel of interest and its surrounding pixels for each pixel (pixel of interest) in a binarized image, a binary image buffer memory that temporarily stores a binarized image; means for selectively reading out pixels in a sensing area including at least one pixel set in advance across a dead zone area that is not involved in feature extraction, and pixels read out from the area of interest or the sensing area; Pixel feature extraction having means for determining the region of interest or a plurality of sensing regions as white or black according to the black and white state of the pixel of interest, and outputting the determination results for the region of interest and all sensing regions as a pixel feature code for the pixel of interest. A pixel feature extraction device comprising: (2) If the region of interest or multiple sensing regions consists of two or more pixels, compare the number of black pixels (or number of white pixels) in the region with a predetermined threshold, and based on the result, select white or black pixels. The pixel feature extraction device according to claim 1, characterized in that the pixel feature extraction device determines that.