JPH0122954B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an improvement in an image position measuring device for positioning an object in a pattern recognition device.

In the case of an automatic white blood cell image classification device where the target object is blood cells, due to the limitations of the microscope, as shown in Figure 1,
The imaging area 1 of the TV camera is generally a wide area containing blood cells other than the target white blood cells.
Therefore, when extracting features, a necessary small part (sample area 2) containing white blood cells is placed in imaging area 1.
It is necessary to cut it out from The position of this sample area 2 is set so that the white blood cells are centered, but in order to do so, it is necessary to find the center position (on the screen) of the imaged white blood cells.
Conventionally, a method using a projection histogram as shown in Figure 2 has been common for this purpose, but in this case, when there are multiple objects as shown in Figure 2, the positions of each cannot be determined. There is a drawback. That is, from the projection histograms 3 and 4 of the real particle images A and B, virtual images C and D can also be considered, so the positions of the particles cannot be determined. Another problem is that the apparatus becomes complicated in order to obtain a projection histogram at high speed. Not only in automatic white blood cell image classification devices but also in other fields, there is a strong demand for high-speed determination of the positions of individual imaged particles, but until now there has been no suitable device that satisfies this requirement.

The present invention was made to eliminate the above-mentioned drawbacks of image position measuring devices, and realizes a high-speed image position measuring device that can accurately measure the center position on an image even when there are multiple objects. It is intended to.

According to the present invention, when a mask pattern obtained by scanning a binary image with unique regions detects a specific shape, the area number of the particle to which the mask pattern belongs and this specific mask are detected.・
Data as a pair with the x or y coordinate of the pattern.
The above object can be achieved by storing data in a memory and determining the center position of each object on the image from the contents of this data memory.

The present invention will be explained below based on the drawings.

FIG. 3 is a block diagram illustrating an embodiment of the image position measuring device according to the present invention. 5 is an input terminal to which the scanned binary image signal is input; 8 is the input terminal to which the binary image signal is applied; This is a unique region generation circuit that generates a region (region where holes and right and lower depressions in a particle image are filled), as shown in one embodiment in FIG.
Pixel information g and f are applied to an AND circuit 82, and the output thereof and a binary image signal from the input terminal 5 are applied to an OR circuit 81, and a target output Sp is obtained from its output terminal. 9 is the output 2 from the unique area generation circuit 8.
It is a one-line memory that stores one line of image information of the value signal Sp, and is composed of a shift register and the like.
10 is the output g from the 1-line memory and the output binary signal from the unique area generation circuit 8.
A shift register 12 holds pixel information of 2 vertical pixels x 3 horizontal pixels by inputting Sp, and 12 inputs the output from this shift register 10, and each control signal Sn, Sd controls the numbering circuit 15 and the data memory circuit 16. control that generates
A logic circuit consisting of, for example, ROM.
13 is one of the control logic circuit outputs.
The flip-flop is set or reset by the input signal and outputs to the control logic circuit.

In the numbering circuit 15, 31 is a particle number counter that generates a number for labeling a particle area, and a bus select buffer 32
Output to. Numerals 34 and 35 are number memories whose addresses are designated by area counters 33 and 36, respectively, for temporarily storing numbers for labeling, and which are read and written via the bus select buffer 32. .

Reference numeral 18 denotes a numbered output terminal for sending the numbered output from the number memory 34 or 35 to the pattern bus via the bus select buffer 32, and is used when calculating the area of a particle region externally.

In the data memory circuit 16, 41 is a U-shape counter that generates a set address of two area numbers of a concave pattern (hereinafter referred to as U-shape), and 42 is an internal numbering bus 1.
The address bus select buffer to which the output from the bus select buffer 32 is added via the bus select buffer 7 specifies the address of the data memory 43. 7 is a microcomputer, etc.
This is an address bus input terminal for specifying the address of the data memory 43 via the address bus select buffer 42 when reading data to the outside using a signal from the CPU. 45 and 46 are x counters and y counters which are countered in synchronization with the scanning of the mask pattern and output the x and y coordinates, respectively. 44 is this x
a data select buffer to which the output from the counter, y counter or the output from the bus select buffer 32 via the internal numbering bus 17 is added;
3. Select and output any of the above data. 4
A data output buffer 7 sends the contents of the data memory 43 to an external bus such as a microcomputer via the data bus terminal 19 in response to an address signal applied to the address bus input terminal.

Reference numeral 6 represents a clock input terminal to which a clock is applied from the outside, and reference numeral 14 represents a timing generation circuit that inputs the clock and supplies a clock in accordance with the signal delay to each circuit within the image position measuring device.

Reference numeral 11 denotes a feature point detection circuit for detecting characteristic points (hereinafter referred to as feature points) for detecting a U-shape in the x direction in the binary image signal input from the input terminal 5, the first of which is shown in FIG. An example is shown. Fifth
In the figure, an OR circuit 112 receives the binary image signal Si input from the input terminal 5, the pixel signal f, and the pixel signal g to its inverting input terminal.
The output of is added to the AND circuit 111, and this AND
The output from the circuit 111 is sent to the control logic circuit 1 via a 2-bit shift register 114.
2 as feature point output Sc. said AND
The output of the circuit 111 is also fed back via the 1-line memory 113 to the positive input terminal of the OR circuit.

The operation of the above block diagram will be explained below. When a binary image signal Si (1 for the particle region and 0 for the background region) is applied to the input terminal 5, the unique region generation circuit 8
becomes the input. In the unique area generation circuit 8 of FIG. 4, the binary image signal Si or the AND circuit 8
OR circuit 81 when either one of the outputs of 2 is 1
The output of is 1. The output of the AND circuit 82 becomes 1 only when the fed back pixel signals g and f are both 1. The pixel signals held in the shift register 10 are those that were added first in the order of a, b, c and d, e, f. Here, the pixel signal f from the shift register 10 is one point before the pixel signal Sp from the unique area generation circuit 8 (on the left in the image).
, and the pixel signal g is the pixel signal g.
This is the pixel signal one line before Sp (upper on the image). Figure 6 1 and 2 are the output Sp of the eigenregion generation circuit.
A 2×2 pattern of the input binary image when is 1 is shown. That is, the input pixel signal Si at a certain point is the sixth
In FIGS. 1 and 2, pixels 60 and 63 are displayed, respectively. In FIG. 6, the pixel 61 adjacent to the top of the input pixel 60 and the pixel 62 to the left of the input pixel 60 are 1 (shaded area in the diagram), so this corresponds to the case where the feedback pixel inputs f and g in FIG. 4 are both 1. The output corresponding to pixel 60 of the resultant unique area generation circuit becomes 1. In Fig. 6 2, the input pixel 63 is 1, so the binary signal input Si in Fig. 4 is 1.
This corresponds to the case of , and the output of the eigenregion generation circuit becomes 1 with the input signal Si as it is. The pixel labeled A in FIG. 6 may have a value of either 1 or 0. By sequentially performing the above operations for each pixel during raster scanning, a binary image with unique regions (hatched areas) filled with particle holes and dents on the right and bottom sides as shown in Figures 10 and 11 is created. Output Sp is obtained.
From now on, this unique region will be treated as a particle region.

The binary signal input Si is simultaneously connected to the feature point detection circuit 1.
It can also be added to 1. Feature point detection circuit 11 in FIG.
In this case, the output of the AND circuit 111 becomes 1 when the binary signal input Si is 1 and the pixel signal f is 0.
And only when the output from the OR circuit 112 is 1. The output of the OR circuit 112 becomes 1 when the pixel signal g or the output of the AND circuit 111 of the previous line outputted from the 1-line memory 113 becomes 1. The output of the AND circuit 111 is sent to a 2-bit shift register 114 in order to match the timing with the mask pattern of the shift register 10.
After a delay of two pixels is caused by
Sent to 2. As a result, 2 of shift register 10
The parts of ×2 mask patterns a, b, d, and e are the sixth
Feature point detection circuit 1 for the pattern shown in Figure 3 or 4
The output Sc of 1 becomes 1, and pixels 64 and 6 respectively
7, the feature point is detected. The circle mark of pixel 68 in FIG. 6 indicates a pixel in which a feature point has already been detected. The feature points are stored in the one-line memory 113 corresponding to each pixel in the one-line memory 9.

Unique area generation circuit 8 and 1 line memory 9
When the output from
b, e, d).

FIGS. 1 to 16 show all possible patterns of a mask pattern of two vertical pixels. Two sets of area counters and number memories (33, 3
4, 35, and 36) are alternately switched for each line of image scanning, and one operates for the upper line and the other for the lower line. Control signals Sn and Sd are output from the control logic circuit 12 to the numbering circuit 15 and the data memory circuit 16 in accordance with the shape of the mask pattern generated in the mask generation shift register 10 for each clock. As a result, the numbering information is the particle number counter 3.
1, and upper line number memory (number memory 3
4, 35) via the bus select buffer 32 to the lower line number memory (the other number memory 34, 35) and the data.
It is output to the memory 16. Bus select buffer 32 receives control signals from control logic circuit 12 and switches signal paths. The contents stored in the lower line number memory are used for the vertical propagation of the next line as the contents of the upper line number memory when the next line is scanned, and are also applied to the bus select buffer 32 as the determined numbering force. from the numbered output terminal 18 to the pattern bus.

Next, the operating principle of labeling will be explained using a specific example using the pattern to be tested shown in FIG. In FIG. 8, reference numeral 80 indicates a background region, 81 and 82 indicate particle regions, and M1 to M5, M2' and M4' indicate mask patterns at respective positions of the binary image. 83 is a line pattern showing the n-th line of the binary image, a ₁ to
a ₃ indicates the area on the line, a ₁ and a ₃ are the background areas,
_a2 is the particle region. 84 is a mask pattern, 85 is an upper line pixel, and 86 is a lower line pixel.

Numbering is done for each area on the line. That is,
When the mask scans line n, the lower line area counter increments by +1 at the first pixel of each area a ₁ to _{a 3} in the line pattern 83 (the leftmost pixel is the part where the area changes), and the lower line number is increased. Numbering information is entered into memory. When scanning the next line (line n+1), the number memory and area counter for the lower line work as for the upper line (the area counter is reset each time the line changes), and the area change point is placed on the upper line of the mask pattern. When appears, the upper line area counter is counted up by +1 and the contents of the upper line number memory corresponding to the next area are output. As a result, numbered outputs corresponding one-to-one to the pixels of the particle region of the binary image are obtained from the upper line number memory in synchronization with the clock.

Labeling is performed using a mask pattern while examining area propagation from the upper line to the lower line. Below area counter 33 and number memory 34
is for the upper line, area counter 36 and number memory 3
The explanation will be made assuming that 5 is operating for the lower line. For example, in the case of pattern M1, there is area propagation from the upper line, so the upper line number memory 3
4 is immediately transferred to the lower line number memory 35 via the bus select buffer 32. In the case of patterns M2, M2' without vertical propagation, the new number from the particle number counter 31 is sent to the lower line number memory 35 via the bus select buffer 32. Since this new number has not yet been determined at this stage, the temporary number flip-flop 13 is simultaneously set as a temporary number flag. In the case of pattern M2, pattern M3 appears after that, and since the temporary number flip-flop has already been set at that point, this particle area is determined to be a new area, and the new number is determined. As a result, the temporary number has been used, so the temporary number flip-flop 13 is reset, the particle number counter 31 is counted up by +1, and the next particle area is prepared. In the case of pattern M2', a pattern M4' with vertical propagation appears after that, and at that point it is determined that this particle area is not a new area, and the contents of the upper line number memory 34 are sent via the bus select buffer 32. Lower line number memory 3
5 and rewrites the contents of the lower line number memory 35. At the same time, the temporary number flip-flop 13 is reset.

Pattern M4 has the same shape as pattern M4', except that temporary number flip-flop 13 has been reset. In this case, it is determined that there is a concave pattern upward on the screen, and the numbered output from the upper line number memory 34 (particle area 82
Particle number corresponding to ) is sent to the data memory 43 via the bus select buffer 32, internal numbering bus 17 and data select buffer 44, and stored as a U-shape number. In the case of U-shape, pattern M5 always appears before pattern M4. This pattern M5
In the pattern M1 preceding , the contents of the upper line number memory 34 (particle number corresponding to the particle area 81) are transferred to the lower line number memory 35 via the bus select buffer 32 and simultaneously transferred to the internal numbering bus. 17 to the data select buffer 44, and has already been stored in a latch circuit therein. At the time of pattern M5, the contents of this latch circuit are written into the data memory 43 as the lower number for U-shape. The U-shape memory area in the data memory 43 is divided into an upper number block and a lower number block, and the above-mentioned upper number and lower number are written at the same location addressed by the U-shape counter 41 in each block. . The above pair of two numbers (upper number and lower number) related to the U shape indicates that the area with these two numbers is a single area, and when combining these two areas later with software. It can be used effectively.

The 2×2 mask pattern shown in FIG. 9 is a mask pattern that shows a specific shape that provides the maximum and minimum values of the X, y coordinates of the particle area. That is, when a particle region is scanned by mask patterns indicated by a, b, d, and e of the shift register 10 in FIG. 3, the mask pattern Xmin shown in FIG. e: The x-coordinate of the particle area (hereinafter the same) indicates the minimum value of the x-coordinate of the particle area, and the y-coordinate of the first item appearing in the area in the mask pattern Ymin of FIG. 92 indicates the minimum value of the y-coordinate of the particle area. 9, and the x coordinate of the particle appearing at the end of the region in the mask pattern Xmax of FIG. 3 is the x of that particle region.
The y-coordinate of the mask pattern Ymax in FIG. 9, which appears at the end of the region, indicates the maximum value of the y-coordinate of that particle region. The above mask pattern from shift register 10
When one of Xmin, Ymin, Xmax, and Ymax is output, control logic circuit 12 outputs to data memory 43 a block address signal for selecting a block address corresponding to this mask pattern. Next, the area number is designated as an address within the block of the data memory 43. That is, in the case of mask patterns Xmin and Xmax, the area number corresponding to the upper right pixel b is read from the upper number memory (either number memory 34 or 35) and sent to the internal numbering bus 17.
to the data memory 43 via. mask·
In the case of patterns Ymin and Ymax, the area number corresponding to the left pixel d is set to the address bus select
It is output from the latch circuit in the buffer 44 to the data memory 43. The contents written to the lower line number memory 34 or 35 via the bus select buffer 32 as described above are simultaneously written to the latch circuit in the address bus select buffer 44. The contents of the x counter 45 or the y counter 46 are written into the data area in the data memory 43 addressed as described above by the memory write signal from the control role logic circuit 12. That is, when the mask patterns are Xmin and Xmax, the contents of the x counter 45 are the mask patterns Ymin and Xmax.
When Ymax, the contents of the y counter 46 are written.

In the case of a simple figure as shown in Fig. 10 (100 is a particle region with a hole, 101 and 102 are parts where unique regions are generated), data and
In the memory 43, the maximum and minimum values of the x and y coordinates of each particle region can be stored at addresses corresponding to the region number of each particle. In FIG. 10, mask patterns M6, M7, M8, and M9 are the 9th mask pattern.
Mask pattern Ymin, Xmin, Xmax,
Each corresponds to Ymax. Therefore, by specifying the address of the data memory 43 from the external processor via the address bus input terminal 7, reading the maximum and minimum values of the x and y coordinates from the data bus terminal 19, and performing calculations in the processor, the particle area can be determined. The center position of can be found.

As shown in FIG. 11, in the case of a particle region with a U shape (in both x and y directions), adjacent region numbers for each U shape are stored in memory, and
The minimum value of the x or y coordinate for each region is stored in a memory, and the minimum x or y coordinate between adjacent regions is taken as the minimum value for the entire particle region. The following description will be made based on the specific example shown in FIG. In FIG. 11, pixels marked with a circle indicate pixels in which feature points have been detected, as described above. As mentioned above, the U shape in the upper direction (y-axis direction) of the screen is defined by the lower number (area 11) in the mask pattern M12.
4 (number corresponding to unique area 111)
However, in the mask pattern M13, the upper number (area 11
3) is stored in the data memory 43. Mask patterns M10 and M11 are mask patterns Ymin of the area 114 and area 113, respectively, and Ymin of the data memory 43.
The y coordinates of the blocks are stored in portions corresponding to areas 114 and 113, respectively. The same applies to the U-shape in the left direction (x-axis direction) of the screen, but in this case, feature points are used for detection. That is, by detecting the x-axis U shape with the M16 mask pattern, two number areas 114, 1
15 (including the unique area 112). When a U-shape in the x-axis direction is detected in the mask pattern M16, the area number of the upper right pixel b is sent from the upper line number memory 34 or 35 to the data memory 43 as the upper number. Next, the mask
In pattern M17, the area number of pixel d on the left side is written to the data memory 43 from the latte circuit in the data select buffer 44 as the lower number. As a result, it can be seen by reading the contents of data memory 43 that area 114 and area 115 are adjacent to each other and form a U-shape. Mask pattern M1
4, M15 are area 114 and area 11, respectively.
As the mask pattern Xmin of No. 5, its x coordinate is stored in the data memory 43 in the same manner as in the case of the y-axis U shape. x for each area written in the data memory 43 as described above,
The minimum value of the y coordinate is read out to the processor along with the combination information of the regions that make up the U shape, the minimum x, y coordinate within the same particle region is determined within the processor, and the maximum value of the x, y coordinate (mask/ By performing calculations together with patterns M18 and M19 (detected using patterns M18 and M19), the center position of the particle region when it has a U shape can be determined.

By using the above-mentioned apparatus, unlike the conventional method using a projection histogram, the center position is determined for each region, so even when a plurality of particle images exist, the position of each particle image can be accurately measured.

Note that although the above description has focused on particle images, the present invention is similarly applicable not only to particles but also to any particle-like target image.

As described above, according to the present invention, it is possible to realize a high-speed image position measuring device that can accurately measure the center position on an image of a plurality of objects even when there are a plurality of objects.

[Brief explanation of drawings]

Fig. 1 is a diagram illustrating the imaging area and sample area of a TV camera, Fig. 2 is a diagram illustrating the principle of a conventional image position measuring device, and Fig. 3 is a block diagram illustrating an embodiment of the present invention. , FIG. 4 is a circuit configuration diagram showing a portion of the unique area generation circuit in FIG. 3, FIG. 5 is a circuit configuration diagram showing a portion of the feature point detection circuit in FIG. 3, and FIG. A diagram showing mask patterns for explaining the operation of the feature point detection circuit, FIG. 7 is a diagram showing all mask patterns indicated by two vertical and horizontal pixels, and FIG. 8 is a diagram showing the pattern to be tested for explaining numbering. Figures 9 and 9 are diagrams showing mask patterns for detecting the maximum and minimum values of x and y coordinates of particle images, and Figures 10 and 11 are test samples for explaining how to determine the center position of particle images. It is a figure showing a pattern. 8... Unique region generation circuit, 11... Feature point detection circuit, 12... Control logic circuit, 1
5...Numbering circuit, 43...Data memory, 45...x counter, 46...y counter,
81, 82, 100, 113, 114, 115,
a ₂ ... particle area, 101, 102, 111, 11
2... Unique area, M1 to M19, M2', M4'...
...mask pattern, Sp...binary image output (from the unique region generation circuit), Sn, Sd...control signal.

Claims (1)

[Scope of Claims] 1. An image position measuring device that measures the center position of each particle region of a binary image consisting of a particle region and a background region, comprising at least means for generating a unique region in the binary image; When the mask pattern obtained from the means for generating the unique region shows a specific shape, the region number of the pixel of the particle region belonging to the mask pattern and the region number obtained from the means for generating the unique region of the mask pattern are determined. 2
An image position measuring device characterized by comprising a data memory that stores x or y coordinates on a value image as a set. 2. In an image position measuring device that measures the center position of each particle region of a binary image consisting of a particle region and a background region, a unique region generation circuit that generates a unique region of the binary image; a feature point detection circuit that detects feature points; and a control circuit that generates a control signal according to the shape of a mask pattern obtained from the binary image output from the unique area generation circuit and the output from the feature point detection circuit. a logic circuit; a numbering circuit that is controlled by a control signal from the control logic circuit and numbers the particle regions of the binary image output from the unique region generating circuit for each region; An x, y counter that gives the x, y coordinates on the binary image obtained from the area generation circuit, and data that is addressed by the area number output from the numbering circuit and in which the output from the x or y counter is written. a memory, when the mask pattern exhibits a particular shape, the x or y coordinates of the mask pattern are data-addressed by the region number of the pixel of the particle region in the mask pattern; An image position measuring device characterized by being configured to store information in a memory. 3. Image position according to claim 2, in which a combination of two area numbers respectively corresponding to two arbitrary adjacent protruding parts of the upward and leftward concave patterns in the binary image is stored in the data memory. Measuring device.