JPS63216185A - Composite pattern recognition system - Google Patents

Abstract

PURPOSE:To attain high-speed recognition by obtaining correlation with an input image while using recognition dictionary data on its bases for each cell and allowing a host computer to select with the highest correlation thereby recognizing a composite pattern. CONSTITUTION:The host computer 3 is provided with a dictionary designation means 31 for designating the pattern in the recognition dictionary of plural cells 20, 21..., an input image broadcast means 32 for broadcasting the input image data to the cells 20, 21... and a decision means 33 for deciding an element pattern of an input image from the correlation data obtained by each cell. Since the cells 20, 21... take the correlation with the image data different in the recognition dictionary and the input image data, many candidates of patterns constituting the input image serving as the composite pattern exist, but it is possible to retrieve the constitution graph by the processing of the host computer. Since the parallel processing is applied by lots of cells, the recognition speed is quickened.

Description

[Detailed Description of the Invention] [Summary] A method for recognizing elemental figures constituting a composite figure from the results of parallel processing of a plurality of cells to find a correlation between an input image and a standard image of a specified figure in a recognition dictionary. It is composed of
Complex figures, which have been difficult to recognize in the past, can be easily and quickly recognized.

[Industrial Application Field] The present invention relates to pattern recognition technology, and particularly to a system for recognizing complex figures.

BACKGROUND ART In recent years, the method of inputting information into electronic computers by directly inputting characters and figures has been increasing, instead of using a keyboard.

Accordingly, powerful and high-speed recognition technology that can handle various patterns is required. Among these, there is a need for a method for easily and quickly recognizing the constituent elements of a complex figure as shown in FIG.

[Conventional technology] A conventional pattern recognition method is shown in FIG.

In the figure, 71 is an image input device such as a scanner,
74 is image data of the input image 70 inputted by the image input device 71 and stored in the main memory of the calculation m73.

76 is a recognition dictionary stored in a magnetic disk device;
75 is image data extracted from the recognition dictionary 76.

The input image data 74 and the image data in the recognition dictionary 76 have generally been subjected to image processing such as normalization and feature extraction.

In conventional pattern recognition methods, features are first extracted from the input image, and then the features obtained are used to extract data from a large amount of image data (generally several thousand to ten thousand times image type value) in the recognition dictionary. I was searching for something that would fit better.

[Problems to be solved by the invention] In the conventional pattern recognition method, (1) When feature extraction is performed on a human image, in the case of a complex figure, the features of the component figures are mixed, and it is difficult to extract them. It becomes difficult to separate them, and therefore it is difficult to match them with the image data in the recognition dictionary, making it difficult to recognize complex figures.

(2) As mentioned above, a large amount of image data is stored in the recognition dictionary, and it takes time to match it.

Such problems arose.

The present invention aims to provide a composite pattern recognition method that solves these conventional problems.

[Means for Solving the Problems] FIG. 1 shows a block diagram of the principle of the complex pattern recognition method of the present invention.

In the figure, reference numeral 1 denotes a recognition dictionary, which stores standard images of elemental figures constituting a complex figure.

2 o, 21. - is a cell, which is equipped with a correlation calculation means 21 that reads out a standard image of a designated figure in the recognition dictionary 1 and calculates the correlation with the input image.

3 is a host computer, which includes a plurality of cells 20 and 21;

Dictionary specifying means 31 for specifying the figure to be used in the recognition dictionary l for each cell 2 o, 2 +, - input image broadcasting means 32 for broadcasting input image data for each cell 2 o, 2 +, -; It is provided with a determining means 33 that determines the elemental figure of the input image from the correlation data.

[Operation] In the above configuration, each cell has input image data and recognition word S.
In order to correlate with different image data in the input image, a large number of candidate figures are generated that make up the input image, which is a complex figure, but it is possible to narrow down the constituent figures through processing by the host computer.

Furthermore, since parallel processing is performed using a large number of cells, recognition speed can be increased.

[Example] The present invention will be explained in more detail below with reference to Examples shown in FIGS. 2 to 5.

FIG. 2 is a block diagram showing the configuration of an embodiment of the present invention.

In the figure, 5 is an input image, and 4 is an image input device.

20.21, 22. - is a plurality of cells, each of which is a unit that includes a processor and memory and constitutes a parallel computer.

3 is a host computer, and each cell is 20.21, 22°.
- Pass the input image to each cell 20.21, 22.・−
- Take in the output from and determine the one that has the highest correlation with the input image.

(2) is a recognition dictionary stored in the magnetic disk device, which consists of image data of a large number of standard patterns.

Each cell 20.21, 22. -- and -- have in common the input image data 22 passed from the host computer 3, and each takes in different image data 23 specified by the host computer 3 from the recognition dictionary 1, and combines the data in the recognition dictionary and the input image. Correlate with the data and send the results to the host computer 3
give it to

A terminal 6 is connected to the host computer 3 and provides a human-machine interface with the operator.

FIG. 3 is a diagram showing an example of recognition of a complex figure according to an embodiment of the present invention.

In the figure, 5 is an input image, 3 is a host computer, 20, 21, 22, and 23 are cells, and 1 is a recognition dictionary.

It is now assumed that the recognition dictionary l stores four types of dictionary images: ○1, △, and ×.

Both input image data and dictionary image data are binary image data.

The host computer 3 has each cell 20, 2 +, 22.2
After assigning the dictionary image to cell 3, the input image data is broadcast to all cells 20, 21, and 22.2t.

Each cell 20.21, 22.23 takes out the assigned dictionary image data from the recognition dictionary 1 and stores it within the cell.

Here, it is assumed that each cell 20, 21, 22.23 is assigned one dictionary image.

FIG. 1 is a flowchart showing the flow of processing in each cell.

In the figure, (a) shows the main flow (MAIN), and (
bl is subflow 1 (SUBI (d, P)) indicating the content of one step in the main flow, and (C)
is a diagram showing the output of subflow 2 (SUB2 (IMAG, d, v)) showing the contents of two steps in subflow 1.

The example shown in the figure is a case where the number of divisions d=4, and the image is divided into 4×
It is divided into 4 areas. The flow of processing of each cell will be explained below according to the flowchart of FIG.

This is a process in which the number of divisions d of the 0 image is determined as n.

■This is the process of subflow 1 (SUBI (d, P)), which is the process of calculating the proportion P in which dictionary images are included in the input image.
)).

■Let the human image be IIMAG, and let the dictionary image be DIMAG.

■Input image density value data vl@-1 for division number d
The human image IIMAG is divided into two 2'' parts. As shown in FIG. 4, if d=4, the area is divided into 4×4−16 areas.

Let l be the area number divided by o-2, and let V be the density value of that area. First 1=0. Let v=none.

■ Increase -31 by 1 (l=f+1).

Determine whether @-41 has reached 2''2. If Yes, return to step @ of subflow 1, N.

If so, proceed to steps 0-5.

@-5 Find the density value V of the small region l. If the number of black pixels in the small area l is greater than or equal to a certain value, it is designated as B (black), if the number of black pixels is F, then it is designated as W (white), and if the number of black pixels is within a certain range in between, it is designated as B (black). Set it to G (gray) and return to step 0-3.

By repeating this process, v=BW as shown in FIG.
Get data like WGGWG) 38BBGGWWW.

The density value data vD of the 0 dictionary image DMAG is determined.

Details are below @-1 as in ■.

(2) Find the part that contains the most sequences in v[l in the vl node, and find the proportion P that it occupies in vDO. For example, if vI=BWWG; vD=BWWW, then Pm3/4=0.75.

When this process is finished, the process returns to the main flow.

■Broadcast the value Pk of P obtained in step ■ to all other cells.

■Receive the value Pi of P from another cell.

(2) Find the maximum value Pm among the PiOs from other cells.

(2) Compare Pk and Pm obtained by the own cell, and determine whether Pk<Pm. If Yes, stop the process (STOP) L,
If No, proceed to step ■.

(2) Determine whether there is a cell whose processing has been canceled. If No, proceed to step (2); if Yes, proceed to step (2).

■Increase the number of divisions d by one (d=d+1) and return to step ■.

■When there are no more cells whose processing has been canceled, send a message to the host computer to the effect that there are no more cells whose processing has been canceled.

According to the above processing flow, for example, the first time, the number of divisions d=
4, calculate P for each cell, and when there are no more canceled cells, increase the number of divisions d by 1 and obtain more detailed P for the remaining cells as d-5. When there are no more canceled cells, the host The computer recognizes that the input image is composed of dictionary images assigned to cells.

In the example of FIG. 2, in the first time, the P of ○2 mouths, △, and × are respectively O', 5.1.0.0.7. If it is set to 1.0, cells with ○ and △ marks will be discontinued.

For the second time, cell 20 and cell 2I, which have an opening and a cross, continue processing, and P = 1.0 for cell 20 and P = 1.0 for cell 2I.
When this happens, the cells 20 and 2I send a message to the host computer 3 that it is a mouth and an x, so the host computer 3 can recognize that the input image consists of a mouth and an x.

Although it is desirable that the number of cells be equal to the number of images in the recognition dictionary, a configuration may be adopted in which one cell can be assigned a plurality of images.

Even in this case, the speed can be significantly improved compared to the conventional example.

[Effects of the Invention] As explained above, according to the present invention, each cell calculates the correlation with the input image based on the recognition dictionary data, and the host computer selects the one with the highest correlation among them.
It makes it possible to recognize complex figures, and since each cell is processed in parallel, high-speed recognition is possible, which has great practical effects.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of the principle of the present invention, Fig. 2 is a block diagram showing the configuration of an embodiment of the present invention, Fig. 3 is a diagram showing an example of processing according to an embodiment of the present invention, and Fig. 4 is a block diagram of the present invention. Flowchart showing the flow of processing in an embodiment of the invention. FIG. 5 is a diagram explaining divided area density according to an embodiment of the invention. FIG. 6 is a diagram explaining a composite figure. FIG. 7 is a diagram explaining conventional recognition. FIG. 2 is a block diagram showing the method. In the drawing, 1.76 is a recognition dictionary, 20, 21, 22.2t, -.
- is a cell, 3 is a host computer, 4, 71 is an image input device, 5.70 is an input image, 6 is a terminal, 21 is a correlation calculation means, 22.74 is input image data,
23 and 75 are recognition dictionary data; 31 is a dictionary specifying means;
32 is an input image broadcasting means, 33 is a determining means, and 73 is a computer. Input image Fig. 3 (a) Flowchart showing the flow of processing in an embodiment of the present invention Fig. 4 (Part 1) (c) Flowchart showing the flow of processing in an embodiment of the present invention Fig. 4 (Part 2 ) Divide into 4 x 4 = 16 areas B, W,
Density value data expressed by any of G = BWWGGWGB
BBBGGWWW Figure 5 illustrates the density of divided regions using JelJ, which is a part of the present invention. Figure 6 is a composite figure consisting of two parts. Figure 7 is a block diagram showing the conventional recognition method.

Claims (1)

[Claims] A recognition dictionary (1) that stores standard images of elemental figures constituting a composite figure; and a standard image of a specified figure in the recognition dictionary (1) is read out and a correlation with the input image is determined. A plurality of cells (2_0, 2_1, ...) equipped with means and a plurality of cells (2_
0, 2_1, ...), specify the figure to be used in the recognition dictionary (1), and after broadcasting the input image, calculate the element figure of the input image from the correlation data obtained for each cell. A composite pattern recognition method characterized by comprising a host computer (3) for determination and configured to recognize elemental figures constituting an input image.