JPS59208667A - Labelling device - Google Patents

Abstract

PURPOSE:To attain high speed labelling by outputting a labelling condition on the basis of data on a processing point and pattern and label data on the preceding points from the processing point by 1H and one piece and relabelling the same pattern area. CONSTITUTION:It is defined that the processing point is A, the preceding point by one piece is B and the preceding point by 1H is C. A pattern signal is inputted to a terminal 15, and if a sweeping point is A, both ends of a 1H data line 1-1 are B and C data. Data labelled by the point A are inputted to a terminal 16 and data labelled by B and C are inputted to both ends of a 1H data line 1-2. When the A is ''0'', the program shifts to the succeeding operation without any processing. If the B and C are ''0''s when the A is ''0'', the output of the gate 7 adds ''1'' to the contents of a counter 11, the added value is outputted from a terminal 17 and a new number is assigned to the A. When the B is ''1'' and the C is ''0'', the same number as the B is assigned to the A, and when the B is ''0'' and the C is ''1'', the same number as the C is assigned to the A. Subsequently, the contents of a memory 14 are arranged and label numbers are assigned again in accordance with the contents of the memory 14.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to an image processing apparatus capable of executing multiple identical processes such as image transmission processing at high speed, and particularly to a high-speed labeling apparatus.

The configuration of the conventional example and its problems
Compact, high-speed processors and memories are used for various signal processing using rated circuit technology. Especially for advanced processing, DSP (Dig
A processor called itatSignaA Processor) is used. This (is ALU (
In addition to the Arithmetic Logic 'Unit), it has a dedicated multiplier, etc., and can perform high-speed data processing.Using this DSP, voice recognition and synthesis, various band compressions for digital transmission, etc. can be processed in real time. can.

On the other hand, considering the combination of image processing and image processing such as medical and pattern recognition, a processing speed of two orders of magnitude or more is required compared to audio signal processing 31H.

One possible way to achieve this is to improve the performance of devices. Since current DSPs are composed of MO8 type LSIs, the speed can be increased by using bipolar type LSIs for this DSP. However, with current technology, it is only possible to speed up the speed by about - orders of magnitude.

By the way, the amount of data of a video signal is enormous, and it is said that a data memory for one field and a capacity of about 4 Mbit are required. MOS type is suitable as a large-capacity memory, but it takes time to load and read data, and real-time data input/output is impossible.

Simple image processing includes detecting edges of patterns, assigning numbers to regions of multiple patterns and labeling to distinguish between regions, extracting pattern features/calculating the area of the pattern, calculating the perimeter of the pattern, calculating the double center of gravity. and so on.

In these processes, in order to process a certain image, it is necessary to also process image data around that point. For this reason, it takes a long time to extract one pattern from the pattern memory.

The conventional two-hering method will be explained based on the binarization pattern shown in FIG. The square part is 1°“
The surrounding area is assumed to be 0''. Sweep in the X direction, P
Add a number of 1 A to the first pattern shown in . If the shapes are the same, assign the same number. Labeling is easy if it is a simple pattern, but
In the case of the pattern shown by Q, the processing is 4 m long. Yi
When sweeping the line shown in , it is determined that there are two patterns, and they are labeled 2゛, ゛'3''.
A problem occurs when the Xi point of the j line is neglected.

In other words, the areas labeled 2"° and 3" are turning areas.Therefore, return to the beginning and perform the sweep again.Then, the area labeled 3°" is the turning area. It is necessary to change all of them to ``P2''.

Re-labeling is performed in this manner, and the subsequent labeling "2" is performed again from the Xi point of the Yj line.

In this example, it may be sufficient to renumber once. Since the pattern is generally more complex, the sweep time for renumbering is several times to several tens of times, making it difficult to process at high speed.

OBJECTS OF THE INVENTION In view of the drawbacks of the prior art, the present invention provides a high-speed labeling device for binarized patterns.

The structure of the invention, that is, the present invention takes a point to be processed in a binarized pattern, the pattern data and 0 before this point, the label data of that point, the pattern data 1H before the point to be processed, and means for outputting a signal for determining five conditions based on the label data at that point; means for generating a label number based on this signal; A labeling device is provided with a memory for storing these label numbers when the labeling device is labeled, and after performing labeling once, stores a plurality of label numbers and performs labeling according to the contents of the memory. It is a sacrifice.

Description of Example The label N of a certain 1ilii element is the pixel before that point and I H1iiJ (1) ll! Let lj elementary data be data. Consider the labeling of pixels in FIG. B for the previous pixel.

Let C be the pixel 1H ago. There are five states of these three 1fii elements shown in the following table.

If the pixel to is ○"', no processing is performed and the next pixel is processed. If the pixel of A is 1", the following four processes are performed.If both E and C are O", then the next pixel is processed. For example, give a new number to the pixel of A. If B is '°1°', add the same lift to the ifj element as B (= J. If C is '1', give the pixel of A Give the same number to C. If both B and C are 1'', give the same number to the pixel to B, and set the number of C to p
It is necessary to show that the exaggerated area is the same as B.

FIG. 3 shows an embodiment of the present invention. 1 represents the 1H digital delay line, 1-1 represents the 1H delay line of the input pattern, and 1-2 represents the 1H delay line of the pattern labeled with the input pattern. 2 to 6 are inverters, 7 to 1Q are ant gates, 11.12 are counters,
13 is a multiplexer, and 14 is a memory. 16 is 2
16 is an input terminal for a labeled pattern, 17 is an output terminal for a label number, and 18 is a flag output terminal.

The input pattern signal is continuously input to the terminal 15 according to the sweep in the X direction. Suppose that the sweep point has just entered the point. In this case, the contents of the 1H delay line have B and C pixel data at both ends. At the same time, terminal 16
The labeled data of point A (still O゛ at this point) is input to B/ and C/, and the labeled data of pixels B and C are input to B/ and C/. When the value of is 0"', a flag signal A=O is output to the terminal 18 via the inverter 6, and the process moves to the next pixel without performing any processing on the pixel. When the value of is 1, If B and C are ○', the output of the AND gate at 7 is 1, the counter at 11 is thickened by 1, and this r direction is The signal is output from terminal 17 via multiplexer 13 controlled by the gate output of gate 7. The data is then written to the memory at the coordinates corresponding to point A connected to terminal 1. This value becomes the label number of point A. In the case where B is "1" and C is °O", the output of the AND gate of 8 becomes 1, and the label number of point B is outputted via the multiplexer 13 controlled by this signal. That is, The label number for point A is the same as the label number for point B.C is 1'' and B
When is '0°'', the output of the AND gate in case 9 becomes 1, and the multiplexer 13 controlled by this signal
The label number C of the 0 point 1H before is also outputted via. In this case, the same label number as point 0 is assigned as the label number of point A.

Finally, if both B and C are '1''(#:j: j
The output of the AND gate of O becomes 1, and the label number B' of point B is first outputted via the multiplexer 13 controlled by this signal. Zunawochi B is "'1°",
It is the same as when C is ○.Furthermore, anj gate 1
When the output is 0, the counter 12 is advanced by 1. This output becomes the address of memory 14. That is, the straddle address is advanced, and the label number of point 0 and the label number of point B are stored at this address. This means that the label number of point 0 is the same as the label number of point B. This example shows that the area labeled 3 is the same as the area labeled 2. In the same way, sweep at the last inch with j~.

However, in this 1-stop, there are patterns in which a plurality of label numbers are "f" even though they are in the same area.

Therefore, it is necessary to allocate label numbers again according to the contents of the memory 14. First, the contents of the memory 14 are cleaned up.

As shown in the example of Figure 3, 6.6. Since the regions labeled y and s are the same, they are unified to the smallest number as shown in 14-2. That is, the label number areas 6, 7, and 8 are all rewritten to the area 5, such as 1 to 5. After that, the contents of the memory containing the label number are transferred to this memory contents 1.
According to 4-2, the sweep label numbers are reassigned once in sequence.0 At this time, the label numbers are skipped (・), so you may reassign them so that they are in the order again.

Effects of the Invention The pattern labeling device of the present invention can perform labeling by performing two sweeps even if the pattern becomes complex, without repeating the pattern sweep many times as in the conventional method. The part that is the bottleneck in speeding up can be greatly speeded up. Not only can input image data be labeled from the memory using a protrusion l~, but also an incoming one-sided image can be labeled and stored in memory in real time. 1st method without using the play line (1) By dividing the point to be processed from memory and the data of the pixel 1H before and reading it sequentially into 2#s, the processing speed will be 2 A, but it will be faster than the conventional method. I can do something like Jy.

[Brief explanation of the drawing]

The first section is a diagram explaining the image labeling device,
FIG. 2 is a circuit diagram of an embodiment of the labeling apparatus of the present invention. 1...1H digital delay line, 2~ra...
Inverter, 11.12... Counter, 13 Multiplexer, 14 ・Memory, 15.16-Input terminal, 1
γ, 18 ・Output l#1i child.

Claims (1)

[Claims] Regarding the binary pattern labeling device, the pattern processing P
means for outputting a labeling condition based on the data of the point to be processed, the pattern data and label data of the point 1H before, and the pattern data and label data of the point immediately before the point to be processed; A means for generating a label number based on the above conditions, and storing the plurality of label numbers in a mixture where a plurality of labels are assigned to the same turn area under the above conditions. A labeling device comprising a memory, and re-labeling the same pattern area into one label number according to the contents of the memory.