JPS63245777A - Processing method for image pattern - Google Patents

Abstract

PURPOSE:To omit a memory to be used for blur processing and to simplify its constitution and address control by adding a blur effect to a normalized image only by one arithmetic operation by means of a scale converting function. CONSTITUTION:An unnormalized original image pattern is regarded as one belonging to a coordinate system (i, j), and when a normalized image pattern is picked up, a reference image pattern belongs to a coordinate system (i', j'). When it is defined that an unnormalized original image is f(i, j), a scale converting function is phi(x, y), a normalized image to which blur effect is added is h(t, u), scale factors are r'=t/i, s'=u/j, and the area of the original image to which blur is added is iS<=i<=iE, jS<=j<=jE, an objective normalized image (t, u) is obtained. Since the normalized image including the blur effect which is ordinarily obtained by coordinate transformation arithmetic can be obtained only by one arithmetic operation, address control can be reduced and it is unnecessary to prepare a memory for storing the normalized image to use it for the succeeding blur processing.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method of processing an image pattern, which is performed as preprocessing during pattern recognition by pattern matching.

[Conventional technology]

For example, if the character image consists of 24x32
Assume that there is a digital image obtained by dividing the image into meshes and expressing the density values at these meshes (lattice points) as digital values. When this digital image is pattern-recognized by the pattern matching method, the digital image is compared with a standard pattern consisting of
Recognition is performed based on the degree of matching.

When comparing a digital image to be recognized with a standard pattern in this way, if the sizes of the two patterns are too different, an effective comparison cannot be made. Therefore, as preprocessing, the size of the digital image is made equal to the size of the standard pattern, and this operation is called normalization. When normalizing a digital image to match the size of a standard pattern, the grid point coordinates that were taken for the digital image before normalization will be taken for the standard pattern (normalized pattern) after normalization. The coordinates of the lattice points (normalized coordinates) rarely match exactly, and a deviation inevitably occurs between them, so by interpolating the deviation, a normalized digital image can be obtained.

This interpolation will be explained below with reference to FIG.

FIG. 2 is an explanatory diagram of this interpolation operation. In the same figure,
The digital image (original image) before normalization belongs to a coordinate system where the horizontal axis is i and the vertical axis is j, and f (i.

j) is expressed as a function. Further, the digital image after normalization belongs to a coordinate system (normalized coordinate system) in which the horizontal axis is k and the vertical axis is l, and g (k.

β) (However, in Figure 2, the coordinate axes themselves are not shown)
.

Now, the grid points (
im, jn) is located within the normalized coordinate system after normalization, four grid points belonging to the normalized coordinate system, namely (kp, βq), (k
p+1. j! q), (kp+1.βq+1), (kp
, jlq+1) exists.

Then, in the normalized image, the possible grid points are no longer grid points (in, jn) belonging to the coordinate system with i on the horizontal axis and j on the vertical axis, but grid points belonging to the normalized coordinate system. In this case, the grid point (im, j
The value f(im, jn) at the surrounding grid points (
kp, j! q), (kp+1. 7!q), (kp+1
．． 1 q +1) and (kp, j2q+1), and this operation is an interpolation operation. The value at each grid point after interpolation is f (kp, Aq)
, f (kp+1.42q), f (kp+1.Nq
+1), f (kp, 6q+1>), and when a linear interpolation method is adopted as the interpolation method, it is expressed by the following equation.

However, the original image f (i, the size of D is (iQxjO)
Assuming that the size of the image g (k, j) after normalization is (kO x β0), the size conversion ratio due to normalization is r = kQ/i in the horizontal axis
Q, 5=A0/jO in the vertical axis direction.

・ (1-β) ・・・・・・・・・(1
)(1-β) ・・・・・・・・・(
2) α)・β ・・・・・・・・・(3
), β ・・・・・・・・・(4
) However, k-(r-1) 11= (s-j) α=r-4-(r-i) β-5-j-(s-j) Here, [] is a Gauss symbol, that is, [ ] This symbol means that when the number in brackets consists of an integer and a decimal part, it is an integer.

Calculations using equations (1) to (4) above are performed as im=1 to io,
By performing this for each grid point defined by jn=1 to jO, a normalized image g(k, 1) is obtained.

When performing pattern recognition using the pattern matching method, it may be effective to perform blurring processing on either the normalized image or the standard pattern (standard image) to be compared.

If one of the two images is expansive and the other is sharp, if you apply blurring to the latter, the images will match well when compared and will not cause erroneous recognition.

Such blur processing involves converting the normalized image to g (k.

l), the image after performing the blurring process is h (t, u), and the size of the area to be subjected to the blurring process. ×b. , the blur coefficient is Cv, (1≦V≦b0, 1≦W≦bo), it is expressed by the following equation.

W) ・・・・・・・・・・・・
・(5) [Problems to be solved by the invention] As explained above, conventionally, as preprocessing for pattern recognition, there are two steps: coordinate transformation for normalizing the original image and calculation for blurring. This method requires multiple steps, and there is a problem in that a memory is required to temporarily store the normalized image in preparation for the next blurring process.

SUMMARY OF THE INVENTION An object of the present invention is to provide an image pattern processing method as pre-processing that solves these problems and allows normalization and blurring to be performed in one operation.

To explain specifically with reference to FIG. 3, it can be said that FIG. 3(a) shows a conventional processing method, and FIG. 3(b) shows a processing method aimed at by the present invention.

[Means for solving problems]

In order to solve the above problem, the present invention defines and uses a function called a scale conversion function. This scale conversion function is defined for each grid point, and if we assume orthogonal axes x and y centered on the grid point, the scale conversion function is the same for symmetrical coordinate positions centered on the It is a function that assigns weighting to a value and similarly assigns the same weighting to coordinate positions that are symmetrical about the y-axis, and is expressed as φ(x, y).

FIG. 4 is an explanatory diagram of such a scale conversion function.

In the figure, the scale conversion function φ(
i-kp, j nq) is shown as a three-dimensional curve. That is, the height of this curve is the coordinate value (i-kp, j-
The scaling function φ(i−kp,
j-j! q).

It can be seen that it takes the same value for symmetrical coordinate positions centered on the coordinate axis 1=kp, and also takes the same value for symmetrical coordinate positions centered on the coordinate axis j-βq. Probably.

[Effect]

Next, the operation will be explained with reference to FIG. In the same figure,
The unnormalized original image pattern is in the coordinate system (i, j)
and the normalized image (standard image)
Assume that when the pattern is mapped, it is determined that the standard image pattern belongs to the coordinate system (i'', j').

At this time, the unnormalized original image is f (t.

j), the scale conversion function is φ(x, y), the normalized image with the blurring effect is h(t, u), and the scale rate is r'=t/i, s' out u/j, If the area of the original image to which blur is added is i3≦i≦i+:+Js≦j≦jt, then
The desired h (t.

U) is obtained.

・f (i, j) ・・・・・・(6)
The areas 1s+tE and Js+JE to which blur is added and the scale conversion function φ(x, y) are optimally selected depending on the size of the target image, the type of image, etc.

The scale conversion function φ(x, y) is an 8th degree polynomial φ(x, y)= (x”−a2) 2・(
y2 1,2)! Hyperbolic type φ(x, y) = (1-a・1x1
)・(1-b・1 y 1) etc. can be used.

Figures 6(i) and (ii) are graphs of these functions, i.e., (i) represents the scale conversion function of the 8th degree polynomial, and (
ii) represents a hyperbolic type.

〔Example〕

Next, the present invention will be described in detail with reference to the drawings.

FIG. 1 is an explanatory diagram showing one embodiment of the present invention.

In the figure, (A) shows a normalized image (standard image), and its size is assumed to be composed of 20×20 meshes. On the other hand, (b) shows the original image, and its size is assumed to be composed of 123×147 meshes. A case will be described in which this original image (123×147) is converted into a normalized image (20×20) and a normalized image is obtained after adding a blurring effect.

As already mentioned, after mapping the normalized image shown in (a) to the original image shown in (b), scale conversion defined as the sampling value of the original image is performed over the area (hatched area) shown in (c). Find the sum of products with the value of a function.

A quantitative explanation is as follows.

Since the scale factor is r-20/123,5=20/147, 1/r=6.15, 1/s=7.
It is 35. Using a hyperbolic type as a scale conversion function, a=
1/12. If b=1/14, φ(x, y) −, (1−l x l/12)
・(1-lyl/14). In this case, the normalized image with the blur effect (
10,'10) pixel, h (10,10) is 1s-C
61,5-12) +1= (49,5) +1=50
it = (61,5+12) = (73,5)”
'73j s= (73,514) +l=
(59,5) +1=60L -(73,5+14)
−(87,5) =87, so j). Other pixels are found in the same way.

〔Effect of the invention〕

According to this invention, a normalized image with a blurring effect, which was conventionally obtained by two coordinate transformation operations, can be obtained by a single operation, so there is less need for address control, and once the normalization is performed, This has the advantage of eliminating the need for memory to store images in preparation for the next blurring process.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram showing an embodiment of the present invention, FIG. 2 is an explanatory diagram of an interpolation operation, FIG. 3 is an explanatory diagram of a comparison between the conventional technology and the processing method targeted by the present invention, and FIG. FIG. 5 is an explanatory diagram of the scale conversion function used in the present invention, FIG. 5 is an explanatory diagram of a coordinate system for explaining the present invention in detail, and FIG. 6 is an explanatory diagram showing an example of the scale conversion function used in the present invention. . Explanation of symbols i, j...coordinate axis

Claims (1)

[Claims] 1) A normalized image as a standard pattern to be compared is inversely mapped onto a non-normalized original image, the original image is sampled according to the normalized coordinates in the normalized image, and the sampling value is Multiply the value by a given scale conversion function defined for each grid point of the normalized coordinates (a weighting function that weights pixels at symmetrical positions by the same amount when viewed from the coordinate axis passing through the grid point). A method for processing an image pattern, characterized in that a normalized image with a blurring effect added to the original image is obtained by weighting the original image and calculating the sum thereof.