JPS63251866A - Picture normalizing method - Google Patents

Abstract

PURPOSE:To extract the stable feature value by obtaining the density value for each of vertical and horizontal components based on a size ratio between a subject pattern and a standard pattern. CONSTITUTION:When the vertical and horizontal component values of a pattern are normalized, the density of each component is corrected in response to the vertical and horizontal reduction rates. IN other words, the vertical component is multiplied by a (y) directional reduction rate y/Y, e.g., 8/24=1/3 is satisfied. In the same way, the vertical component is multiplied by an (x) directional reduction rate x/X. Thus two normalized pictures having the maximum density value '6' are obtained. In such a way, the stable feature value is obtained with no difference of density caused between the vertical and horizontal components even with deformed characters.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for normalizing the size required in pattern recognition and the like.

[Conventional technology]

Normal pattern recognition processing is performed after converting binarized or digitalized image data for each pixel into vectorized data with size and direction, as shown in Figure 4.
, the size is normalized (■ span), and recognition is performed according to a predetermined algorithm (see ■).

Conventionally, the following methods are known as normalization methods in such processing steps. This means that input image 1 with a size of AxxAy as shown in FIG. 5(a),
A certain size BxXBy as shown in the same figure (b))
When performing K normalization, AX * Ay t” of the input image
Divide BX e By into equal parts, (Ax / BX)
This is a method of mapping feature amounts included in a rectangular area with a size of (Ay/By) to one of the output images.

An example of normalization using the Kaheru normalization method is shown in FIG. This is "+" character pattern 1A, IB, IC
In the nine examples, character 1B has the same vertical and component density as in (e), while characters IA and iC have (
As shown in (d) and (f), a difference in concentration occurs between the vertical component and the constituent components.

[Problem that the invention seeks to solve]

However, with the above-described method, there is a problem that a character or mark that has been deformed due to the writing person's habits or optical factors has a difference in density between the vertical component and the component, making it impossible to obtain stable feature quantities.

Therefore, it is an object of the present invention to provide a normalization method that does not cause density differences between vertical components and constituent components even in the case of deformed characters, and can obtain stable feature amounts.

[Means for solving problems]

Using the size ratio between the target pattern and the reference pattern, density values are determined for each vertical component.

[Effect]

When normalizing the height of the pattern and the size of its components,
By correcting the density of each component according to the vertical and horizontal scale factors, stable feature quantities can be obtained.

〔Example〕

FIG. 1 is an explanatory diagram for explaining in detail the present invention in the case of normalizing the vertical component, FIG. 2 is an explanatory diagram for explaining in detail the present invention in the case of normalizing the horizontal component, FIG. 6 is an explanatory diagram for explaining an example of an input image.

Now, when there is an input image "+" as shown by reference numeral 1 in FIG. 5, first, the vertical component and constituent parts are extracted. The extracted vertical components are shown in Figure 1 (A), and the extracted components are shown in Figure 2 (A).
Shown below. When these vertical components and components are normalized using the conventional normalization method, the maximum density value is "18" for the vertical component and "18" for the component, as shown in Figure 1 (b) and Figure 2 (b), respectively. In this case, the value becomes "12", and a density difference occurs as in the case of FIG. Therefore, in this invention, the following steps are taken. For the vertical component, multiply by the scale factor y/Y in the X direction. In this example, 8/24=1/3. Similarly, the horizontal component is multiplied by the scale factor x/X in the X direction.

In this example, 8/16=1/2. As a result, the maximum density value is 16# as shown in Figure 1 (C) and Figure 2 (C).
A normalized image can be obtained, and the density difference as described above can be eliminated.

Here, x / X m y / Y is multiplied by each direction component, but if the constant used to multiply the vertical component is A, and the constant used to multiply the horizontal component is B, then A: B = y/Y: x/X
= (1) In the case of constants A and B, especially X m y,
It can be seen that it is sufficient as long as the constants A and B satisfy.

By the way, the normalized correlation value S(A, B) of the two patterns A and H is defined as follows (3). Here, ABke BJjk represents one component in the feature plane. 1. j is the characteristic surface X.

It corresponds to y, and k corresponds to the number of feature planes. Here, the input cover turn t-P (size: IXJ), resignation? Q (
Size: 1Xj), the input cover pattern P is normalized to the size of ixj using this normalization method, and the correlation value is calculated as shown in the following equation (4) (α: a constant according to the feature surface )00111. (4) α is a constant for each feature surface,
k=2 (vertical component.

I is any positive number). Therefore, when normalization is performed as in the present invention, the correlation value can be obtained using equation (4).

〔Effect of the invention〕

According to the present invention, it is possible to correct the difference in concentration between the vertical component and the horizontal component, which occurs with the conventional normalization method, and therefore it is possible to extract a stable I¥f Sosha.

[Brief explanation of the drawing]

FIG. 1 is an explanatory diagram for explaining in detail the present invention in the case of normalizing the vertical component, FIG. 2 is an explanatory diagram for explaining in detail the present invention in the case of normalizing the horizontal component, FIG. 3 is an explanatory diagram for explaining an example of an input image, FIG. 4 is a flowchart showing an overview of general pattern processing,
FIG. 5 is a conceptual diagram showing the concept of a conventional normalization method, and FIG. 6 is an explanatory diagram for explaining a specific example of the conventional normalization method. Code explanation 1, IA, IB, IC...Input image, 2,
2'. 2A, 2B, 2C... Normalized images. Agent Patent Attorney Akio Namiki Agent Patent Attorney Kiyoshi Matsuzaki Figure 1 (A) 112 Figure (A) Figure 3 Figure 1fA Figure 5 @ Figure 6

Claims (1)

[Claims] A normalization method for normalizing a target pattern having a certain size and having concentration information for each component divided vertically and horizontally to a reference size, the target pattern and An image normalization method characterized by determining density values for each vertical and horizontal component using a size ratio with a reference pattern.