JP2981902B2 - Character photo automatic recognition processing method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an improvement in a method of automatically processing an original image in which characters and photographs are mixed and distinguishing between characters and photographs.

SUMMARY OF THE INVENTION Dispersion processing, mesh processing and / or edge processing are performed on input data composed of a large number of pixels having different densities that constitute an original image in which characters and photographs are mixed, so that characters and photographs are accurately recognized. It is about the method.

2. Description of the Related Art When a document in which a character image and a photographic image are mixed is optically read and image information obtained by digital processing is reproduced, the reproduced image is very difficult to see. For this reason, edge processing and mesh processing have been proposed as processing methods for automatically recognizing the image area separation between characters and photographs and making the reproduced image easier to see.

[Problems to be Solved by the Invention] However, since the conventional method focuses only on the density difference between pixels, in the case of a thick character, the non-edge portion of the thick character has no density difference and is not recognized as a character. There was a problem that was recognized as.

Accordingly, an object of the present invention is to provide an automatic character / photograph recognition system for automatically recognizing image area separation between a character and a photograph and making it easy to view a reproduced image.

Means for Solving the Problems In order to achieve the above object, the present invention converts an original image in which characters and photographs are mixed into a plurality of pixels having different densities and reads the input data for each pixel. 3x pixels
3 are sequentially divided on a matrix of 3 and the average value of the density of 9 pixels including 8 pixels in the vicinity of each pixel of interest in the divided input data is obtained. Calculating the sum of the squares of the pixels and sequentially comparing the result with a threshold value to determine whether the pixel is a character pixel or a photographic pixel, and a plurality of n × n (n is a natural number) A) a mesh processing means for counting pixels determined to be photographic pixels for each mesh, comparing the counted value with a threshold value, and determining whether the photographic region is a text region for each mesh based on the result. And a character / photo automatic recognition processing method comprising:

[Operation] Since the distributed processing means is provided in addition to the mesh processing means and / or the edge processing means, the recognition accuracy for particularly thick characters is improved.

EXAMPLES The present invention will be described below with reference to examples shown in the drawings.
FIG. 1 shows a basic system conceptual diagram of an embodiment of an automatic text / photo recognition processing system according to the present invention. In the figure, 1
Is a reading unit, 2 is a central processing unit (CPU), 3 is a memory (RA
M), 4 is ROM, 5 is mesh processing unit, 6 is distributed processing unit,
Reference numeral 7 denotes an edge extraction processing unit.

The reading unit 1 includes a known optical character reading device (OCR) or the like, and converts an original image in which characters and photographs are mixed into a large number of pixels having different densities and reads the digital data as digital data.

The CPU 2 temporarily stores the read data in the memory 3 and controls the mesh processing unit 5, the distribution processing unit 6, and the edge extraction processing unit 7 so as to recognize characters and photographs as described below. Let it be processed. The ROM 4 stores a program necessary for the above control.

Thus, in the first method of the present invention, the pixel data is subjected to a dispersion process and a mesh process. In this distributed processing, a predetermined operation is performed on one pixel of interest, which is input data (1 pixel = 1 dot), while observing the situation of eight pixels in the vicinity thereof, and the result is compared with a threshold to compare the result with the threshold. It is determined whether the pixel is a character pixel or a photo pixel.

The calculation of the distributed processing is performed by the distributed processing unit 6 as follows.

9 including the pixel of interest stored on a 3 × 3 matrix
An average value of the densities of the pixels is obtained, and a sum of squares of the average value and the density of each pixel is calculated. Assuming that the sum of squares is SD, the average value is m, and each pixel density is X _i (i = 0 to 8), This SD is proportional to nine pixels.

In the input data of a character, the SD is large with respect to the edge density distribution of the character, and the SD of a character region having a thick uniform density is small. The pixel density of a photograph has a distribution of almost SD.

According to the experimental (computer simulation) study of the inventor, when the threshold value of SD is Sa, Sb, Sa <Sb, the pixels in the character region are SD <Sa or Sb <SD, and the pixels in the photographic region are Sa <Sb. ≦ SD ≦ Sb.

By such dispersion processing, thin characters, thick characters, and photographs having a gentle change in pixel density can be accurately determined.

Next, the mesh processing unit 5 divides the entire area of the input data into a plurality of meshes in order to make the discrimination information more accurate.
For example, one mesh is divided into 2 × 2 portions, and for each mesh, the number of character pixels in the four portions is counted. If the count value Sm is larger than the threshold value Sc, the number of character pixels is calculated from the character region and Sc. If it is smaller, it is determined to be a photographic area, and image area separation is performed.

FIG. 2 is a flowchart showing an algorithm for performing the above-described distributed processing and mesh processing.

In the second method of the present invention, the edge extraction processing is performed in parallel with the distribution processing of the first method described above, thereby further improving the recognition accuracy in the case of particularly thin characters.

In the edge extraction processing unit 7, 9
A matrix in which a number of pixels are stored is multiplied by a 3 × 3 weighting matrix to obtain a sum, the density of each pixel is converted, and an edge is extracted. The density of the pixel thus extracted is compared with the threshold value T to determine whether the pixel is an edge pixel (character pixel) or a non-edge pixel (photo pixel).

The CPU 2 obtains the determination results of the edge extraction processing unit 7 and the distribution processing unit 6, determines whether the pixel is a photograph pixel or a character pixel, and shifts to the mesh processing.

FIG. 3 is a flowchart showing the algorithm of the second method.

[Effects of the Invention] As described above, according to the present invention, the accuracy of image area separation of thick characters, thin characters, and characters and photographs is improved as compared with the conventional method, and image data without unnaturalness is obtained. Is obtained, and it is possible to determine a character as a character even with a thick character by the above algorithm, so that the practical effect is remarkable.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a basic system configuration of an embodiment of the present invention, and FIGS. 2 and 3 are flowcharts showing algorithms of the first and second systems of the present invention, respectively. 1 image reading unit, 2 CPU, 3 memory, 4 R
OM, 5: mesh processing unit, 6: distributed processing unit, 7:
Edge extraction processing unit.

Claims (2)

(57) [Claims] An original image in which characters and photographs are mixed is converted into a plurality of pixels having different densities, and read input data is divided into pixels of each pixel sequentially on a 3 × 3 matrix. For each pixel of interest in the divided input data, the average value of the density of nine pixels including eight pixels in the vicinity thereof is calculated, and the sum of squares of the average value and the density of each pixel is calculated.
A distributed processing unit for determining whether a pixel is a character pixel or a photograph pixel by sequentially comparing the result with a threshold value; and dividing each area of the original image into a plurality of n × n (n is a natural number) meshes;
Mesh processing means for counting pixels determined to be photographic pixels for each mesh, comparing the counted value with a threshold value, and determining a photographic area or a text area for each mesh based on the result. Character and text automatic recognition processing method. 2. An input image in which an original image in which characters and photographs are mixed is converted into a plurality of pixels having different densities, and read input data is sequentially divided for each pixel into a 3 × 3 matrix. For each pixel of interest in the divided input data, the average value of the density of nine pixels including eight pixels in the vicinity thereof is calculated, and the sum of squares of the average value and the density of each pixel is calculated.
A dispersion processing means for determining whether the pixel is a character pixel or a photographic pixel by sequentially comparing the result with a threshold value; and each pixel subjected to density conversion by multiplying the pixel storage matrix by a 3 × 3 weighting matrix to obtain a sum. Edge extraction processing means for comparing the threshold value with the threshold value to determine whether the pixel is an edge pixel or a non-edge pixel,
A determination unit for determining whether the pixel is a photograph pixel or a character pixel according to the determination result of the distribution processing unit and the edge extraction processing unit, and converting each area of the original image into a plurality of n × n (n is a natural number) meshes And a mesh processing means for counting pixels determined as photo pixels for each of the meshes, comparing the counted value with a threshold value, and determining a photo region or a text region for each mesh based on the result. An automatic text / photo recognition processing method.