JPS619775A - Picture image processor - Google Patents

Abstract

PURPOSE:To increase the picture image processing speed by advancing the end point of an edge toward the maximum evaluation function value obtained about each peripheral point of the end point of a noticed edge. CONSTITUTION:A source selector 6 counts the number of end points and stores the count value to an internal register of an ALU5. Then an access is given to a frame memory 18 by address counters 16 and 17. Thus the first end point ¦E¦, acute angleE and the FLG data can be fetched by the ALU5. Each peripheral point of the end point has at least one edge point (flag=1). Thus this edge point is detected inside the ALU5 and deleted out of the extension candidate points. Then the evaluation function H is calculated between the remaining extension candidate point and a noticed end point. In order to shorten this calculation time, an operation acute angleEo-acute angleEi is performed by the 1st ROM table 11a and the address value of an angle calculation table is written to the 2nd ROM table 11b.

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field The present invention relates to an image processing device that processes digital image information captured by a chimp camera and converted into a multivalued image.

Background technology Conventionally, pattern recognition of grayscale images has been performed.

First, extract the edges in the original image as shown in Figure 1 and create a line drawing (an image consisting of a logical ``0'' and a logical ``1'', in which a line is drawn by connecting the logical ``1''s).

After that, the percentage characteristics of the line drawing are extracted and the pattern is recognized. In order to extract edges from a grayscale image, a differential method is generally used because it is sufficient to extract changes in density within one image.

FIG. 2 is a diagram showing an example of a 3×3-dimensional differential operator. @2 Diagram +I+ changes in the lateral direction (determined as ΔX)
is determined, and the change in the vertical direction (denoted as Δy) is determined using FIG. 2 (2). The magnitude of the two-dimensional differential value (IEI) is determined by the first equation of order H.

IE1=φ7]7F...+11 The direction of the change can be found using the following equation 21.

1E=tan-"(Δy/ΔX)-(21After this, we perform line thinning processing to thin the 1 wide edge line to a line with a width of 1 pixel, and perform a binarization processing session to extract points with even stronger differential values. , '@ Original drawing of figure 1.

As shown in FIG. 3, it roughly becomes an edge line drawing. Here, what I mean by "roughly" is that the original picture is quite accurate! This means that in images with a lot of noise, edge lines tend to become discontinuous.

Conventionally, to convert this discontinuous edge line drawing into a complete line drawing,
There were approximation methods and edge extension methods. The approximation method samples several points distant from the edge line and approximates a straight line or curve, but it has the drawback that if the accuracy of the approximation cannot be ensured, an incorrect edge line will be detected. On the other hand, the edge extension method.

Starting from the end point of a discontinuous edge, calculate a certain evaluation function between the first pixel of interest and each of its surrounding points, and extend the edge to the surrounding point with the largest value until it collides with another edge point. This is an extension. Here, the evaluation function
IEI) - - (31i = 1.2..., 8. However, /IEil is the magnitude of the differential value of one point among the surrounding points, lE, ij is the direction of the differential value. This turtle edge The extension method is an excellent method for extracting edge lines that were erased during thinning and binarization.

Since this method is generally realized by software using a digital computer, it has the disadvantage that it takes a long processing time and is difficult to put online.

OBJECTIVES An object of the present invention is to provide an image processing apparatus that solves the technical problems mentioned above and can perform edge extension processing at high speed as preprocessing for pattern recognition of grayscale images.

Embodiment FIG. 4 is a block diagram of an embodiment of the present invention. The processing device 11ζ has 1 condition code selector 2. Sequencer 3.
Program ROM (read only memory) 4. and A
An LU (arithmetic unit) 5 is provided. Two condition code selectors are connected to the sequencer 3 via line 14. The program ROM 4 is connected to the sequencer 3 via a pass line 15 . ALU 5, line l! 1
1'a[' and is connected to each output of the three-stage buffer circuits 7-9. Output of ALU5, connect to line 115□
The latch circuits 10a and 10b are connected to the latch circuits 10a and 10b through a line 1161ft: connected to the first ROM table lla, which is a calculation memory,
The first ROM table lla and the launch circuit Inc are connected through 116 lines to the second ROM table llb+ζ, which is a calculation memory. The second ROM table llb is connected to the three-state buffer 7Iζ via line 114. Frame memo via 3 step buffer circuit 8 line 112'e! 718, and a 3-step buffer circuit with 9 lines I! 13' via the buffer address counter 12. x address buffer 14. and Y address buffer 15. Buffer address counter 12 indicates line I! A clear signal is provided via line 20vil- and OR gate 13 via line /19'.
An output signal of is given. Is the output of the X address buffer 14 line/24? Intermediate L7 is connected to the X address counter 16, and the output of the Y address buffer 15 is line /2.
3 to the Y address counter 17. X address buffer 14 and Y address buffer 151
OX address counter 16 and Y address counter 17 are provided with an edge write signal to write the output signal of buffer address counter 12 via line 122.
are provided with a load signal on line 130 which causes all of these counters to be in full operation. Further, the X address counter 16 is given an X up signal via the line 131 to move the X coordinate of the end point of the edge in the positive direction.7-1. Further, an X down signal is applied to line 7732 to move the X coordinate of the end point of the edge in the negative direction. Y
Address counter 171rld, line I! 33?
Y coordinate of edge end point via r? A Y-up signal is given to move the line in the positive direction, and the line l! A Y down signal is applied 712 via 34'Th to move the Y coordinate of the end point of the edge in the negative direction. X address counter 16 via frame memory 18 and life 126.

to the Y address counter 17 via line Z25.

It is connected to the direction decoder ROM (read only memory) 19 through the line 127'l, to the 3-state buffer 8 through the line l1-2Th, and to the condition code selector 2 through the line 12, respectively. In addition, a flag write signal for writing an edge flag, which will be described later, is applied to the frame memory 18 via line 128.

The direction decoder ROM 19 outputs line 1 to condition code selector 2! Connected via 3'r. A start signal is applied to the condition code selector 2 via line /1. 6 source selectors.

Line I! by signals applied to lines 16 and 17, respectively. 8~l! ] Select the output signal of 0.

The operation of the block diagram shown in Figure 41 will be explained below. The sequencer 3I-i controls the flow t1 of the program, and normally increments the address in the program ROM 4, but if the instruction is a conditional jump or loop, the data in the condition code selector 2? Appropriate control is performed while observing. Address signal from sequencer 3? line! ! 5) and sends the program ROM l- to the microprogram that directly controls the 7) ware.

This 4 ROM bipolar fuse ROM that can be read at high speed is used. Does ALUS have multiple internal registers? The digital input data from line/11 and internal register data are operated on.

What is the result of performing an operation between some internal registers? Send to line 11.5.

The data that should be prepared before starting the edge extension of the image is the differential absolute value IE1.1 stored in advance in the frame memory 18. Differential direction code ZE.

These are an edge flag FLG and end point address data representing the end point of the edge. This is a coded version of the differential direction code ZE.

Is Figure 5 an example of hooding in 16 directions with 4 bits? FIG. This number of focuses is arbitrary, and may be further increased if accuracy is to be compromised. Edgeflag FLGH@6
As shown in the figure, the result of thinning and binarization is
1”, it is an edge point.

If it is a logical rOJ, it is 1-bit data that is a non-edge point. This frame memory 18H, X, Y address counter 1.6. +E1. of the address indicated by j7. lE,
It is output in parallel to the FLG data. These data calculations can be written using software, or real-time processing can be performed using hardware. FIG. 6 shows an example when the direction codes are 1 to 8.

Next, the end point address data will be explained.

Examples of end points are shown in FIGS. 7(a) to 7(d). Figure 7 (a).

Figure 7(b) has endpoints, while (c) and (d) do not have endpoints.

Here, it is clear that an end point can only be used in a case where there is one pixel of logic "1" or two consecutive pixels of logic "1" in one place at the surrounding points of the pixel of interest.

It would be difficult to determine these conditions using software.

fJI as a pattern care dress for the flags of 8 surrounding points
ROM table lla and second ROM table 11b
? If used, it can be discriminated quickly.

Externally write the X and Y addresses of the pixels that meet the end point conditions for all grooves. In other words, Line I! The buffer address counter 12 is incremented by the edge amplifier signal applied to the X and Y address buffers 14 and 18, and then the edge write signal applied to the X and Y address buffer lines 14 and 15 is used to perform a pre-write operation. this? If this is done over one entire screen, the total number of endpoints will remain in the buffer address counter 121, and the XIY address of the number of endpoints will remain in the address buffer 14゜1.
5 in order from address O. The above is the preparation operation.

Hereinafter, the entire operation of edge extension will be explained with reference to FIG.

First, the source selector is set as a 6-digit expansion address counter, and its count value (the number of end points) is stored in the internal registers of all ALUs 5. Next, move the buffer 77 and dress counter 12 to line l! 201 Doyo Eladreste (Sofa 14.15
The contents of address θ of line 124. l! Sent on 23rd. Whisper this, X.

The Y address counters 16 and 17 are filled. (Address of end point O in FIG. 8) Next, this address counter 16
．． By accessing the frame memory 18 using 17, the first end point +E+, /E, and FLG data can be taken into kALU5. Next, data i of 8 surrounding points of this end point
, x, y address counters 16 and 17 separately and then given to the ALU 5.

As mentioned above, since there is at least one edge point, that is, a point with flag = logic rlJ, in the surrounding points of the end point, this point is searched for within the ALU 5 and deleted from the extension candidate points. Next, the above-mentioned evaluation function Hk is calculated between the remaining extension candidate points and the end point of interest, and the third equation is multiplied by the cosine (
cos) Since it takes a long time to perform the calculation, the first ROM table lla and the second ROM table l shown in @4 are
Use lb. Line I from ALU 5! 15, the calculation data 1Eil, ZEo, and ZEi of the evaluation function H are sent to the latch circuits 10a to 10c. Calculated data/
EO is launched into the latch circuit 10a, and the calculation data lE
I have a 10bl lunch circuit.

It is sent to the first ROM table 11a via line /16a as the address of the angle calculation table. Eo
-ZEi is calculated using the angle calculation table. Calculated data 1 EilFi Launched by the launch circuit 10C, calculated data lE o sent from the 1st ROM table
2nd R via line /16b with -l E i
It is sent to OM table llb as the address of the evaluation function table.

Since the angle calculation table is provided in the first ROM table lla, the capacity of the second ROM table can be reduced, reducing the volume of the nodeware. Can be made smaller. what if,
Calculated data l launched into launch circuits 10a to 10c
E il I Z E o + /E i is sent directly to the second ROM table llb as an address. Assuming that 1Eil is 6 bits, lEo is 4 bits, and Ei is 4 bits, the total data length is 14 bits. This 14 pits? If the second ROM table llb is configured as an address, a ROM with a capacity of 16 words is required. As in the present invention, 笥lRO
When the M table 11a is provided, the first ROM table 1
Since the calculation of ZEO-.DELTA.Ei is performed at '1a and the output is 4 bits, the length of the data sent as an address to the second ROM table llb is 10 bits. Therefore, the capacity of Ir2ROM table llb is I
It becomes the K word. Content tfl of the first ROM table 11a
The accounting ROM capacity is 10 with the H'lK word.
88 words, which can significantly reduce the ROM capacity. For example, a fuse ROMk with an access time of about 40 n5ec is used for the first ROM table lla, and an access tie A of 20 n5ec is used for the second ROM table 11b.
It can be used for a programmable ROM' of 0 to 300 n5ec.

The table value of the evaluation function H is read into the ALU 51 from the second ROM table l'lb. This operation is performed on the extension candidate pit, and the edge is extended to the pixel in the direction indicating the maximum value. The flag of the pixel designated as the frame memory edge point is transmitted through the 128ft line and written as an X in the flag write signal.

For edge extension processing, repeat all of the above operations.

When extending further from a pixel that has newly become an edge point, it is known that the direction in which the pixel has progressed is always an edge point, so it is necessary to import data in that direction and extend the 0 pixel. If you find a flagged point in a direction other than the direction you were traveling, it means that you have touched another edge line, so stop the extension at that point. ((7) in Figure 8)
At this point, the extension from end point 0 is completed, and in order to extend from end point 1, the expansion address counter 12 is incremented, and the x, y address counters 16, 171 are set to the logical "1" address indicated by the end point. Load the file and perform the same process as described above.

The above processing is repeated, but since the initial number of endpoints is within ALUS, all extension processing ends when the processing is repeated that number of times.

When performing extension, only three pixels in the direction of the differential value of the current pixel can be used as full extension candidate points. This situation is shown in FIG. 9. If the differential value direction of the current point is direction code 8 in FIG. 6, the extension candidate points are A and A.

3 pixels. In order to perform this method 4, it is necessary to determine the direction ZE of the current point using the ALU 5.

For this purpose, a complicated process is required in the ALU S to perform an exclusive OR of the numbers 1 to 8 and lE to see if it becomes 0. Therefore, as shown in Figure 4.

Give the differential direction code LE to the direction decoder ROM 19,
Speeding up can be achieved by supplying data indicating which direction the code is in from 1 to 8 to the condition code selector 2 and checking which of the eight code inputs is in focus.

In the configuration of FIG. 4, lEt in the frame memory 18
Direction decoder ROM19 and direction decoder ROM1
No. 9 selects one of the eight directional focusing directions. Flowchart of this direction determination) k$10 Figure number is shown. As is clear from this flow chart, the direction determination # operation requires a maximum of 8 steps and an average of 4 steps. In order to eliminate this loss, the circuit shown in Figure 11 is added. That is, direction decoder R
The output of the OM 19 is given to the address of the manipulating ROM 20 connected to the sequencer 3. Therefore @4
The direction decoder ROM 19 shown in the figure is deleted. Jump destination addresses of microprograms corresponding to each direction code are written in the mapping ROM 20. For example, when the differential direction code lE is 4 bits, there are 16 codes, but since they are divided into 8 directions, the code 2"
The jump destination address for determining and processing the direction "2" is entered in the place where "3" is the address. With the addition of this mapping ROM 20, microprogramming and JUM
One step of the PMAP instruction makes it possible to read and determine one pixel.

Effects As described above, according to the present invention, edge extension processing as pre-processing for pattern recognition of grayscale images can be executed at high speed.

[Brief explanation of drawings]

Figure 1 C original drawing? Figure 2 is a diagram showing an example of a one-dimensional differential operator, Figure 3 is a differential, thinning, and 211 formula t
Also, FIG. 1 is a diagram showing a corresponding image, FIG. 4 is a block diagram of an embodiment of the present invention, and FIG. 5 is an example of a direction code. Figure 6 is an example of when the direction code is set to 1 to 8, and Figure 7 is an example of the end point of an edge. 8 and 9 are diagrams for explaining the entire process of extension, FIG. 10 is a flowchart for determining direction, and FIG. I: This is an added block diagram. DESCRIPTION OF SYMBOLS 1... Processing device, 6... Source selector, 7.8°9... 3-state bumper, 10a-1oc... Latch circuit* 11a... First read-only memory table, 11b... -Second read-only memory table. 1゛2...Banfer address counter, 13...O
R gate, 14...X address buffer, 15...
Y address buffer, 16...X address counter,
17...Y address counter, 18...frame memory.

Claims (1)

[Claims] An address buffer that stores coordinate data of pixels that are end points of discontinuous edges representing the periphery of an image in grayscale image processing, and an address data output from the address buffer is loaded, and this address data is An address counter for calculating the data of pixels adjacent to the pixel of interest based on the image, and an image such as a code that indicates the differential direction with respect to the direction extending from the end point of the edge based on the output of the address counter. A frame memory that stores data; and a frame memory that uses the data to indicate the relationship between a pixel of interest and its surrounding pixels, and that is used to identify mutually different points of the calculation steps in an evaluation function that includes a plurality of calculation steps. a processing device that controls the address buffer, the address counter, and the plurality of calculation memories and calculates the image data; The feature is that the end point of the discontinuous edge that is being looked at begins to be extended, the evaluation function value is obtained for each point around the end point of the edge of interest, and the end point of the edge is advanced in the direction of the maximum function value. Image processing device.