JPS62160592A - Picture normalizing device - Google Patents

Abstract

PURPOSE:To obtain a stable normalization picture in which a line width after a normalization is not fluctuated by a change, etc., due to dislocation or a scale ratio by corresponding one picture element at a normalization picture to a partial area at an original picture, and preventing each partial area at the original picture corresponded to the picture element adjacent to the normalization picture from overlapping with each other. CONSTITUTION:When a correspondence between the picture element of a normalization picture KP and that of an original picture NP to allocate the picture element value of the NP to the KP is performed, each partial area of the NP corresponding to the picture element neighboring on the KP is set so as to coverlap with each other with a constant ratio (alpha). The start point coordinate of an LP in the coordinate system of the NP can be found from the start point coordinate (XM, YM) of an MP at adders 31 and 32. A normalization computing element 5 has a memory on which the original picture NP is stored in its inside, and finds and outputs the sum of picture elements within the LP by being given XL+XM, YL+YM, and the lengths of the partial area LX and LY. An outputted value is stored on a normalization picture memory 6 with an address outputted from a counter 4. By advancing the counter 4 with the completion signal SF of the normalization computing element 5, all of the picture elements on the KP can be calculated.

Description

[Detailed description of the invention] [Industrial application field] The present invention is useful for character recognition, 5, drawing reading, etc.

The present invention relates to an image normalization device for preprocessing when recognizing line figures such as symbols.

[Conventional technology]

The correlation method is generally good (known) as a means of recognizing two characters and figures.
, j) and the image Qk(i, j) of the Dictionary 8 (standard) pattern
This is a method in which the correlation Rk between Rk and Rk is defined as the following equation, and the pattern belongs to k that maximizes this Rk.

However, since such correlation methods are sensitive to changes in pattern position and size, in order to compensate for this, it is customary to align the spread of the image of the unknown pattern and the position deviation prior to correlation calculation. . This work is called normalization. Furthermore, from the above equation (1), the amount of product-sum operations required to calculate the correlation is the square of the side length of the image size (N x N),
In other words, it increases in proportion to N, and when there are many types of symbols to be recognized (for example, characters exceed 2000 m)
In general, normalization is performed along with image reduction, since the amount of computation is an obstacle to realizing a practical device. In other words, the general expression for normalization for a recognition device based on correlation is MXM(MIXM2)O with the coordinates (XM, YM) on the NxN (fixed) image NP as the reference point, as shown in Figure 3.
This is to reduce the partial image MP to a normalized image KP of (KXK).

Therefore, M≧, and the scale factor /M is generally about 1 to 1/10.

By the way, as a conventional normalization means, a method using interpolation has been used exclusively. That is, the image of normalized image KP: AK
Coordinate values X1 and Yl on the original image NP corresponding to P (t l J ) are determined by a coordinate transformation equation such as the following equation (2).

Since the values of Xl and Yl are generally not integers, the surrounding four points NPO and NPl are expressed as follows.

From the pixel value of N P 21 N P 3, NPo-NP(
CX1:) , (Yl) )NP+=NP((
Xl) , (Yl:l+1)NP2=NP((Xl
)+1n [Yl) )NP3-NP((Xl)+
1. (Yl:l+1)K (i
, j). Note that [ ] indicates a Gauss symbol, and therefore (X) represents the largest integer that does not exceed X.

This method has the advantage that the amount of calculation is small and the circuit is simple because it is determined only by K without increasing or decreasing by M, but on the other hand, if the scale factor when normalizing the line figure is large, Because NPs are sampled only sparsely due to differences in sampling, there is a drawback that lines disappear.

Therefore, the literature [Improvement of correlation method for character recognition] (Yasuda, Fujisawa), Trans, IECE'79/3Vo
l.

62-dl13 was devised.

This paper is difficult to understand, but the essence is normalization by performing the opposite transformation to the above.

[Problem that the invention seeks to solve]

Although this method does not cause lines to disappear, it has the following problems.

b) There are a lot of product operations. In particular, 8M' calculations including (floating) decimal points are required to allocate the values of drawing lines (8 calculations are required for each point), which creates a heavy burden and makes it difficult to improve the recognition speed.

Generally, the value of one pixel on NP is distributed to four points on KP, so the minimum line width on KP is 2 pixels, but t
If x and ty are normalized to be 0 (if they overlap), the line width may be 1. That is, since the normalized result is unstable with respect to slight positional deviations and scale factors, the recognition rate decreases.

Therefore, the present invention eliminates the drawbacks of (a) 1) above,
The purpose of the present invention is to provide a normalization device that can be implemented at high speed with a small amount of calculations and has stable recognition performance for practical use in character gR devices and drawing recognition devices.

[Means for solving problems]

A region information generating means for generating partial region information obtained by enlarging a partial region of the original image that corresponds in position to the normalized pixel by a predetermined coefficient (α), a memory for storing the original image information, and a memory for storing the contents of the memory. A memory successive means for reading based on the area information and an integrating means for integrating the read pixel values are provided.

[Effect]

That is, in order to allocate the pixel values of the original image NP to the normalized image KP, when associating the pixels of KP with the pixels of NP, each part of NP corresponding to each adjacent pixel on KP is The regions have a certain ratio to each other (α
) so that they overlap. In this way, adjacent pixels on the normalized image have redundant information about the original image, so that the normalized result becomes stable regardless of positional deviation or scale. Note that the pixel value of KP is expressed as the sum of each pixel value in the partial area. This eliminates the need for product calculations, so the load on the device can be reduced.

〔Example〕

FIG. 1 is a block diagram showing an embodiment of the present invention.

In addition, in the figure, 11 to 14 are buffers, 21 to 2
4 is a memory table, 31.32 is an adder, 4 is a counter, 5 is a normalization arithmetic unit, and 6 is a normalization image memory.

Further, as shown in FIG. 1A, the normalization calculator 5 includes buffers such as a digit buffer BX1p, a digit buffer BX2p, a line buffer BY, a digit counter CXp, a digit counter CTx, and a line counter CY.

Counters such as line counter CTY, digit comparator CPx
, comparators such as row number comparator CPy, address table AT
, an original image memory M, and an integrator T.

Note that the digits correspond to the X direction, and the rows correspond to the Y direction.

Now, as shown in Figure 3, the original image NP (size.

NXN) partial image MP (size MXM, reference coordinates (X
M, YM)) to normalize the normalized image KP (size p
Consider the case where Kx x ) is output. At this time, N
, are constant for the device. Further, it is assumed that the parameters XM, YM, M (Ml, M2) representing MP have already been determined by the symbol extraction circuit before normalization. In other words, the starting point X coordinate of MP, the starting point Y
Coordinates YM are stored in digit buffer 111 and row buffer 12, respectively, and width M1 and height M2 of MP are stored in buffer 1.
3° buffer 14 respectively.

In this process, as shown in FIG. 2, a partial region LP of MP for an arbitrary pixel KP (i, j) on the normalized image KP is considered and determined as follows. First, the starting points XL and YL of the partial area LP are determined as follows. In addition, α is 1 × α × 2
This is a constant.

Here, the reason why the value in parentheses is compared with @0 and takes the maximum value is to prevent the partial area LP from protruding outside the MP. This means that XL and YL set the reference point of MP to (0,
This is because the coordinates are defined as 0).

Next, similarly, the end points XF and YF of the partial area LP are determined using the following equations. Note that () is the length of the partial area LX, L
Y, and the value should be rounded up.

Therefore, becomes.

Since K and α are constants for XL, YL, LX, LY, i,
The numbers are determined by j, Ml, and M2, and therefore, these are tabled as memory tables 21, 22, and 22 in FIG. 1, respectively.
Stored in advance on 23.24 (.

In this way, the output of the counter 4 giving the normalized pixel positions i and j, and the buffer 13 in which the values of Ml and M2 are stored.
．． Memory table 21.22.23.2 with output of 14
By accessing 4, XL+X at 31.32
By determining M and YLLYM, the coordinates of the starting point of LP in the coordinate system of NP can be determined. The normalization calculator 5 has a memory for storing the original image NP inside it, and
Given L+XM, YL-1-YM, LX, and LY, the total sum of pixels in LP is calculated and output. The output value is stored in the normalized image memory 6 according to the address of the output of the counter 4. By incrementing the counter 4 using the completion signal SF of the normalization calculator 5, all pixels of KP are calculated.

Next, the operation of the normalization calculator 5 will be explained with reference to FIG. 1A. Starting point coordinates of LP on NP XL+XM, YL+Y
M is each digit buffer BX1.

Each is stored in the row counter Cy as an initial value. X
L+XM is simultaneously stored as the initial value of the digit counter CX, and the digit counter CTxy line counter CTY is '0'.
The integrator 6 is also reset to 0. The number of digits buffer BX2y and the number of lines buffer BY include:
The width LX and height LY of each region LP are stored. The digit and row values are output from the digit counter CXs and the row counter Cy, and the original image NP is output from the address table AT.
is converted as the address of the image memory M where it is stored, and the value of the pixel of NP is read out.

At the same time as the digit counter Cx counts up, the digit number counter CTX also counts up, and the digit number buffer BX
When @LX of the area LP stored in 2 is exceeded, a line feed signal SL is output from the digit number comparator C'Px, the line counter CYp and the line number counter CTY are counted up, and the digit counter Cx is filled with digits. Area L from buffer BX1
The starting point X coordinate XL+XM of P is stored again, and the digit counter CTX is reset to ``O''. In this way, the values of pixels within the area LP are successively read out from the image memory M and integrated into the integrator T. Then, the area L where the value of the line number counter CTY was stored in the line number buffer BY
The operation ends when the line number comparator CPy detects that the height of P has exceeded LY.

〔Effect of the invention〕

According to the present invention, since the configuration includes a means for generating a partial area from the position and size of the partial image to be normalized according to the above equation (4) and the size of the normalized image, one part of the normalized image Pixels are associated with partial regions of the original image, and the respective partial regions of the original image, which are respectively associated with adjacent pixels of the normalized image, overlap with each other, resulting in misalignment and scaling. It is possible to obtain a stable normalized image in which the line width after normalization does not fluctuate due to changes, etc., resulting in an effect of improving the recognition rate.

In addition, by providing a means for scanning the associated partial area and a means for integrating the read pixel values, the product operation is no longer necessary, so the normalization operation can be performed at high speed, resulting in faster recognition speed. can be improved.

[Brief explanation of drawings]

FIG. 1 is a configuration diagram showing an embodiment of the present invention, and FIG. 1A is a configuration diagram showing an embodiment of the present invention.
FIG. 2 is a block diagram showing a specific example of the normalization calculator in the figure, FIG. 2 is an explanatory diagram for explaining the present invention in detail, and FIG. 3 is an explanatory diagram for explaining a general normalization method. Code explanation 11, 12, 13.14, Bxl, BX2
+ BY...Buffer, 21, 22, 2;, 2
4...Memory table, 31.32...
... Adder, 4. CX, CTX, CY. CTY・・・Counter Ra・kawa・Sho 1 Ihin space g-A・・・Normalized image memory, AT...
...address table, M...original image memory, T...integrator, NP...original image, M
P...partial image, LP...partial area,
KP... Normalized image. Agent Patent attorney Akio Namiki Agent Patent attorney Kiyoshi Matsuzaki Figure 1 Figure 1 To memory Figure 2 Figure 3

Claims (1)

[Scope of Claims] An image normalization device for converting an original image consisting of a plurality of pixels including gray scale information into a normalized image, comprising: a memory for storing the original image information; and each image on the normalized image. area information generation means for generating partial area information obtained by enlarging a partial area of the original image corresponding to the pixel by a predetermined coefficient; and accessing the memory based on the enlarged partial area information to An image normalization device comprising: memory reading means for reading contents; and integrating means for integrating pixel values read from the memory.