JPH03269782A - Character area extracting method for character reader - Google Patents

Abstract

PURPOSE:To accurately read characters by using the two-dimensional fractal matrix of a multilevel picture to classify a texture pattern and extracting a character area. CONSTITUTION:A multilevel picture 5 consisting of a background area 3 and a character area 4 having characteristic texture patterns is inputted from an image pickup device 1 to a character reader 2. A rectangular coordinate system of (x) and (y) is considered and the density value at each point of the picture is read on the (z) axis to obtain the multilevel picture which has a picture surface 6 of a curved surface corresponding to the density value. The character reader 2 operates the trace value of the two-dimensional fractal matrix and the value of a determinant based on inputted picture data and uses the difference in texture pattern between the background area 3 and the character area 4 based on both operated values to extract the character area 4. A character is recognized in the extracted character area 4. Thus, the character is accurately read.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a character area extraction method in a character reading device that reads characters from a multivalued image containing characters captured using an imaging device such as a CCD camera. .

[Prior Art] As a method for extracting a character area in a conventional character reading device, first, filter processing such as smoothing is performed on the image shown in FIG. 4(a), and the image shown in FIG. 4(b) is Then, as shown in FIG. 4(C), the local noise 20 is removed, the histogram distribution of image density values is obtained, and the image is binarized with appropriate density values based on this. A common method is to separate the background area and character area and extract the character area, as shown in FIG. 4(d).

[Problem to be solved by the invention]

However, in the conventional character area extraction method described above, even if the background image has a characteristic texture pattern, as shown in FIG. Since the background region and the character region are separated using only the histogram distribution of the density values shown in Fig. 5(b) without using the There was an unresolved problem that characters were cut off and it was not possible to extract an appropriate character area.

Therefore, the present invention has been made by focusing on the unresolved problems of the above-mentioned conventional example.
It is an object of the present invention to provide a character area extraction method for a character reading device that can accurately extract character areas.

[Means to solve the problem]

In order to achieve the above object, a method for extracting a character region in a character reading device according to the present invention is provided. For each point in the multivalued image,
The trace value and determinant value of a two-dimensional fractal matrix obtained from the number of curves covered when the length of orthogonal cut curves on the image plane formed by the image density values are measured in unit length, and the values of the trace and determinant of the two-dimensional fractal matrix are calculated. The present invention is characterized in that the background area and the character area are separated from each other based on a calculated value, and the character area is extracted.

[Effect]

In this invention, when extracting a character area, instead of simply extracting the character area based on only the density value of the image, the value of the trace of the two-dimensional fractal matrix and the value of the determinant are calculated, and both calculations are performed. To accurately extract a character area from a value by using the difference in texture pattern between a background area and a character area.

〔Example〕

Embodiments of the present invention will be described below based on the drawings.

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

In the figure, 1 is a CCD that scans documents and outputs image data.
It is an imaging device such as a camera, and image data output from this imaging device 1 is input to a character reading device 2.

This character reading device 2 calculates a trace value and a determinant value of a two-dimensional fractal matrix based on input image data, and based on both calculated values, the difference in texture pattern between a background area and a character area. Extract the character area using .

That is, a multivalued image 5 consisting of a background area 3 and a character area 4 having a characteristic texture pattern as shown in FIG. 2(a) is now input from the imaging device 1 to the character reading device 2. shall be. In this FIG. 2(a), x
, y, and taking the density value at each point of the image as the Z axis, we obtain a multivalued image having a curved image plane 6 according to the density value, as shown in Figure 2). be able to.

In this FIG. 2 (b), any pixel P on the multivalued image
Cut curve 7 passing through (x, y) and orthogonal on the image plane 6
Considering x and 7y, this cut curve is 7x.

When the length of 7y is measured with unit length U, the value in the X-axis direction of the number of curves covered is X(u), and the value in the y-axis direction is Y
(u), the vector-valued function V (
u) can be obtained.

V (u) = tCX (u), Y (u))
--(1) At this time, (V(u)) = u-' (V(1):l...
If there exists a (2,2) matrix H that satisfies -(2), it can be defined that ■ is a two-dimensional fractal quantity, and matrix H is a fractal matrix associated with ■.

Therefore, in order to estimate the matrix H, a two-dimensional vector W(s) expressed by the above equation (2) is obtained.

Since the right side of this equation (3) is the solution of the differential equation W' = -HW, each component of the matrix H (h++, h+z.

h2°hgz) can be obtained by calculation.

Each of the calculated components hll+ hl□, hz+
, calculate the trace value T of the matrix H and the determinant (iD) from the hz□O value.

Then, the above calculation is performed for all pixels on the image, and the distribution of the value T of the trace of the matrix H and the value of the determinant, which is the calculation result, is as shown in FIG. Samples of the same category corresponding to character region 4 form a cluster with each other, and are included in region 8 included in the stable helical focal region and in the stable nodal focal region separated by the curve l represented by D=T”/4. The area 9 is classified into two texture patterns.

Therefore, by finding an appropriate boundary 10 for separating the background area and character area from FIG. 3, pixels corresponding to the character area can be accurately extracted using this boundary 10.

Then, by performing character recognition on the extracted character area, it becomes possible to perform accurate character reading.

In addition, in the above embodiment, the case was explained in which the fractal matrix is traced and the determinant value is calculated for each pixel on the image, but this is not limited to this, and depending on the resolution of the image and the size of the characters, For example, it is possible to first extract a global character area by sampling every few pixels, and then perform more detailed calculations within that area, which has the advantage of speeding up the character reading process. There is.

〔Effect of the invention〕

As explained above, according to the present invention, two
A dimensional fractal matrix is used to classify the texture pattern, separate the background area from the text area, and extract the text area. Even if there is a problem, the character area can be extracted more accurately and the characters can be read more accurately than in the case of extracting the character area only based on the density value of the image as in the conventional example.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIGS. 2(a) and (c) are explanatory diagrams for explaining the character extraction method of the present invention, and FIG. 3 is a diagram showing trace values and determinants. Graphs showing the relationship with values, FIGS. 4(a) to 4(d) and FIGS. 5(a) and 5(b) are explanatory diagrams for explaining the conventional character extraction method, respectively. In the figure, 1 is an imaging device, 2 is a character reading device, 3 is a background area, 4 is a character area, 5 is a multivalued image, 6 is an image plane, 7x, 7
y is the cut curve and 10 is the boundary. i Carbon concentration value

Claims (1)

[Claims] In a character reading device that extracts a character area from a multi-valued image consisting of a background area and a character area and reads the characters, an orthogonal cut on the image surface made by the density value of the image is made for each point of the multi-valued image. The trace value and the determinant value of the two-dimensional fractal matrix obtained from the curve coverage number when the length of the curve is measured in unit length are calculated, and the background area and the character area are separated using both calculated values. 1. A method for extracting a character area in a character reading device, characterized in that a character area is extracted using a character reading device.