JPH05266251A - Character extractor - Google Patents

Abstract

(57) [Abstract] [Purpose] To provide a character extraction device that can accurately extract only characters from a read object such as a full-color document as well as a document represented by a black and white binary image. A color line sensor 1 collects color image information on an object P to be read, and a color image input circuit 2 is used to collect the color image information.
The color image information of R, G, and B is stored in the frame memories 3, 4, and 5, respectively. First, the CPU 9 uses the color image information in the frame memories 3, 4 and 5 to determine R,
The area is divided based on the distribution of the G and B spaces. Next, CP
U9 is, for each of these divided binary images,
Data in which character-likeness is extracted from the horizontal projection data is obtained. Next, the CPU 9 measures the height, width, and sharpness of the data extracted from the character-likeness to determine the character-likeness, and outputs only the binary image determined to be character-like as the final result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character extracting device for extracting only characters from a full color document such as a gravure magazine.

[0002]

2. Description of the Related Art For example, a binarization process is well known as a character separation method for separating only a character part from a multicolor image in which a character part and a picture part are mixed. This is to extract characters using only the brightness information of the image.
When the picture part is complicated, the character cannot be extracted by the simple binarization process. Therefore, for example, Japanese Patent Laid-Open No. Hei 2
As disclosed in Japanese Patent Laid-Open No. 253765, an improvement measure has been devised in which binarization processing is adaptively performed according to the characteristics of each image such as the maximum density difference between neighboring pixels.

[0003]

As described above, the character portion is separated by using the brightness information of the image and the local spatial information, but in such a method, for example, a white character or It was difficult to extract characters with high performance from a read object such as a full-color document like a gravure magazine.

Therefore, an object of the present invention is to provide a character extracting device capable of accurately extracting only characters not only from a document represented by a black-and-white binary image but also from a read object such as a full-color document. To do.

[0005]

A character extracting device of the present invention includes an information collecting means for collecting information on each color component from an object to be read in which information represented by a plurality of colors is mixed, and the information collecting means. And a character extracting unit for comprehensively determining the color information and the spatial information, and extracting a character region from the object to be read.

Further, the character extracting device of the present invention includes an information collecting means for collecting information about each color component from the object to be read in which the information expressed by a plurality of colors are mixed, and the color information collected by the information collecting means. Image division means for dividing the area based on the image division means, and character image determination means for determining the character likeness of the area divided by the image division means.

Further, the character extracting device of the present invention includes an information collecting means for collecting information on each color component from the object to be read in which the information expressed by a plurality of colors are mixed, and the color information collected by the information collecting means. Image dividing means for dividing the area on the basis of the area, accumulating means for accumulating the extracted pixels of the image information in the horizontal direction in the area divided by the image dividing means, and the characteristic of the accumulated information accumulated by the accumulating means. It is provided with a feature extracting means and a determining means for determining character likelihood from the features extracted by the feature extracting means.

[0008]

According to the present invention, it is possible to extract characters using not only the brightness information and local spatial information as in the past, but also color information and global spatial information. Only the characters can be extracted accurately from.

[0009]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows the configuration of a character extracting device according to this embodiment. In the figure, P is an object to be read, for example, a full-color document such as a gravure magazine having a multicolor image in which a character portion and a pattern portion are mixed,
It shall be conveyed in the direction of the arrow shown. The color line sensor 1 performs photoelectric conversion by optically scanning the conveyed object P to be read in a direction perpendicular to the conveying direction, and sends the photoelectric conversion signal to the color image input circuit 2. The color image input circuit 2 converts the input photoelectric conversion signal into red (R),
Converted to digital signals of green (G) and blue (B), R, G,
The digital image information of B is stored in the frame memories 3, 4, and 5, respectively.

The frame memories 3, 4 and 5 are controlled by the memory control circuit 6 and are connected via an address bus 7 and a data bus 8 to a CPU (Central Processing Unit) 9 which controls the overall operation. There is.
The CPU 9 has a program memory (ROM) 10 in which a control program of the CPU 9 and the like are stored, a work memory (RAM) 11 of the CPU 9, and a data output circuit for outputting data to the outside via the address bus 7 and the data bus 8. 12 are connected.

Next, the character extraction processing operation in such a configuration will be described. The object P to be read is conveyed in the direction of the arrow in the figure, and when passing through the line sensor 1, the color image information on the object P is photoelectrically converted by the color line sensor 1, and the photoelectric conversion signal is the color image input circuit 2. Sent to. The color image input circuit 2 converts the photoelectric conversion signals from the color line sensor 1 into red (R), green (G) and blue (B).
Convert to digital image information. This converted R,
The G and B color image information is stored in the frame memories 3, 4, and 5 under the control of the memory control circuit 6, respectively.

Then, when the storage of the image information in the frame memories 3, 4, 5 is completed, the frame memories 3, 4, 5
The control of 4, 5 is switched to the CPU 9 side. Then C
The PU 9 executes the processing described below according to the program of the processing procedure stored in the program memory 10.

That is, the program memory 10 stores, for example, a program for executing the process shown in the flowchart of FIG. 2, and the CPU 9 performs the process in accordance with this program. First, an area is divided according to each color image information in the frame memories 3, 4, and 5 (S1), and character-likeness is determined for each of the divided binary images (S2). Only the binarized image thus extracted is extracted and output as the final image (S3).

The color image division process of step S1 is specifically executed by the process procedure shown in the flowchart of FIG. 3, for example. That is, first, all the image areas for storing the division result are initialized to “0” (S1
1), the image resulting from the division is copied to the mask image area (S12). Then, in the unmasked image, that is, in the mask image, R of the pixel having a value of “0”,
Check the G and B values and create a histogram of them (S
13). This can be understood as a numerical value placed in a three-dimensional space as shown in FIG.

Next, the principal components of the three-dimensional distribution are analyzed to determine the principal axis direction (S14). In other words, this is a process for obtaining the direction in which the maximum dispersion is obtained. For example, when the main axis is obtained with respect to FIG. 5, it becomes the axis S represented by the direction vector s shown in FIG.

Next, the R, G, B histograms are projected onto the principal axes (S15). This becomes, for example, one-dimensional data as shown in FIG. In this projection one-dimensional data, if there is only one distribution peak, that is, one mountain (S16), all the images for that distribution are set to 1
One region is registered in the division result image (S17), and if an undivided region is present (S18), the process is repeated from step S12.

When it is determined in step S16 that the projected one-dimensional data is composed of a plurality of distributions, the threshold value t is determined using a threshold value determination method based on discriminant analysis (S19). .. Due to the main axis S and the threshold value t on the main axis, the plane U as shown in FIG.
Is determined, and pixels corresponding to one of the distributions are registered as a mask image based on this plane U (S2
0). Then, for the unmasked image, R,
The process is repeated from the procedure (S13) of creating the G and B histograms.

Thus, in step S18, the process is repeated until there is no undivided area. And
When this color image division processing is completed, for example, as shown in FIG.
The color image having the configuration as shown in FIG. 10 is a set of binary images with divided labels as shown in FIG.

Next, the character-likeness determination processing in step S2 is specifically executed by the processing procedure shown in the flowchart of FIG. 4, for example. That is, first, the value of the divided label counter L is set to "1" (S21), and thereafter, as to the divided binary image of interest, projection data is obtained in the horizontal direction as shown in FIG. 11 (S22). The data shape of the projected one-dimensional data is measured as shown in FIG. 12 (S23). Then, by comparing the width W, the interval B, the height H, and the sharpness H / dx, which are the measurement results of the shape, with the reference value, the character-likeness is determined (S2).
4). For example, since characters are almost always printed in lines, character-likeness can be determined from the width W and the height H of the horizontal projection data.

When the character-likeness is determined, it is then determined whether or not it forms a character string (S25). If it forms a character string, it is determined that the character is included. The image is stored as an output image (S26). And
By proceeding to steps S27 and S28, this character-likeness determination is performed on all the divided binary images.

The steps S22, S23, and S24 for determining the character-likeness are performed by other methods, for example, the run-length frequency distribution shown in FIG. 14 (see FIG. 1).
Even if it is a method of determining character likeness by analyzing 5), the gist of the present invention is not changed at all.

When the character likeness is thus determined, the character extraction processing is completed by outputting the character-extracted binary image from the data output circuit 12 in step S3.

As described above, the above processing is to extract a character by using the color information and the global spatial information as shown in the conceptual diagram of FIG. That is, in FIG.
A full-color image as shown in FIG. 13A is divided into regions based on the distribution of R, G, B spaces, and a plurality of binary images as shown in FIG. 13B are obtained. Next, for each of the divided binary images, data in which character-likeness is extracted as shown in FIG. Then, with respect to the data in which character-likeness is extracted, for example, height, width, and sharpness are measured to determine character-likeness, and as a result, a binary image determined to be character-like as shown in FIG. 13D. It only outputs. It should be noted that the present invention is not limited to the above-mentioned embodiment, and it goes without saying that various modifications can be made without departing from the spirit of the present invention.

[0025]

As described above in detail, according to the present invention, it is possible to accurately extract only characters not only from a document represented by a black and white binary image but also from a read object such as a full-color document. A device can be provided.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a character extraction device according to an embodiment of the present invention.

FIG. 2 is a flowchart shown to explain a character extraction processing procedure.

FIG. 3 is a flowchart shown to explain a color image division processing procedure.

FIG. 4 is a flowchart shown to explain a processing procedure for character likelihood determination.

FIG. 5 is a diagram conceptually showing R, G, and B histograms.

FIG. 6 is a diagram conceptually showing a main axis in an R, G, B histogram distribution.

FIG. 7 is a diagram conceptually showing the projection of R, G, and B histograms on the principal axis.

FIG. 8 is a plane composed of a main axis and a threshold value on the main axis;
The figure which shows notionally that image division is performed.

FIG. 9 is a diagram showing an example of an object to be read.

FIG. 10 is a diagram showing an example in which an object to be read is divided into a plurality of binary images by color image division.

FIG. 11 is a diagram conceptually showing projection data in the horizontal direction of an image, which is performed to determine the character likeness from a divided binary image.

FIG. 12 is a diagram conceptually showing a feature amount captured from projection data.

FIG. 13 is a diagram conceptually showing the entire character extraction processing.

FIG. 14 is a diagram for explaining another method of determining character-likeness.

FIG. 15 is a diagram for explaining another method of determining character likeness.

[Explanation of symbols]

P ... Object to be read, 1 ... Color line sensor, 2 ... Color image input circuit, 3, 4, 5 ... Frame memory, 6
...... Memory control circuit, 9 ... CPU, 10 ... Program memory, 11 ... Working memory.

Claims (3)

[Claims] 1. An information collecting means for collecting information on each color component from an object to be read in which information expressed by a plurality of colors is mixed, and color information and spatial information collected by the information collecting means A character extracting device comprising: a character extracting unit that determines a character area from the object to be read. 2. An information collecting unit that collects information about each color component from an object to be read in which information represented by a plurality of colors is mixed, and an image that divides an area based on the color information collected by the information collecting unit. A character extracting device comprising: a dividing unit; and a character image determining unit that determines a character likeness of a region divided by the image dividing unit. 3. An information collecting unit that collects information about each color component from an object to be read in which information represented by a plurality of colors is mixed, and an image that divides an area based on the color information collected by the information collecting unit. Dividing means, accumulating means for accumulating the extracted pixels of the image information in the horizontal direction in the area divided by the image dividing means, feature extracting means for extracting the features of the accumulated information accumulated by the accumulating means, and this feature extracting A character extraction device, comprising: a determination unit that determines character likelihood from a feature extracted by the unit.