JPS61107876A - Picture procession device - Google Patents

Abstract

PURPOSE:To apply data-compression with high efficiency to a binary picture signal to store and further decode by adding picture data for every area of the storing means, adding area classification and the size of the minimum area of areas and storing them. CONSTITUTION:The picture read from a reading device 1 by a solid-state image pick-up element is made binary in a binary processing and a coordinates arranging part 2. If the picture is inputted in inclination in order to match the character row with address space coordinates in a page memory 5, the picture is rotated, coordinates are shaped and the picture is stored in the memory 5. At a mesh dividing coding processing part 3, a half tone part or a pattern and line picture area is MH-coded by a dot dimension, and the compressed coded data are stored in a data storing means 4. On the other hand, at a compositing part 8 based upon the size of the mesh and the coding data, a printing type font stored in a font ROM9 is read, successively reproduced and stored in a space of a line memory 7 as a printing type row and outputted as a visible image at an output device 6.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to an image processing device for storing block coding of a binarized image signal and further decoding it.

[Prior art] After reading the image with a reading device such as a copying machine and converting it into two values,
For example, when saving a file to an optical disk device, etc., it is desirable to encode and compress the data. However, since the image signal is redundant and has different degrees of intensity depending on the image quality, a single image contains a mixture of text, photographs, graphics, etc. If data is compressed using only one encoding method for the entire image area as in the past, the compression efficiency will be low.

[Objective] The present invention has been made in view of the drawbacks of the conventional example described above, and its object is to provide an image processing device that compresses and stores binarized image signals with high efficiency, and further decodes the data. It's there.

[Embodiment] The outline of the present invention is to divide an image area containing a character string of a certain size into meshes in a binary image signal, and to determine the size of the mesh such that each character can be stored in the mesh. Character recognition is performed for each mesh, and the recognized characters are encoded.

Furthermore, the present invention is characterized in that characters, graphics, and photographic regions that cannot be stored in the mesh, that is, different sizes, are separated and conventional pixel-based encoding is applied to these image regions.

Based on the above features, embodiments of the present invention will be specifically described below with reference to the drawings.

FIG. 1 is a block diagram of an image processing apparatus according to an embodiment.

Reference numeral 1 denotes an image reading section using a solid-state image sensor such as a CCD. The read image is binarized in step 2, and in order to align the character string with the address space coordinates in the page memory, if the image is input at an angle, it is rotated and coordinates shaped. Store in. 3 is a portion to be named a mesh division encoding processing section which is characteristic of the present invention. The encoded data is stored in data storage means 4.

On the other hand, the decoding unit 8 reads out the type font stored in the phono ROM 9 based on the size of the mesh and the encoded data, sequentially reproduces and stores it as a type string in the space of the line memory 7, and outputs a visible image on the output device 6. Output as .

Next, an outline of the image processing in the mesh divided image processing unit 3 of this embodiment will be explained step by step based on the flowchart of FIG.

Step 20>... Determination of mesh size One page of image data D (x, y
), calculate the histogram of the number of black dots in both the X and 7 directions. However, x and y are appropriate orthogonal coordinate axes within the page memory 5.

When obtaining an X-direction histogram, the number of black dots at all y-coordinate values for a certain X-coordinate value is counted, and this is performed for all X-coordinate values to create an X-direction histogram. When creating a histogram in the X direction, the number of black dots at all X coordinate values for a certain y coordinate value is determined.

When the above method is applied to a character string of a sentence as shown in FIG. 3, a histogram in the X direction as shown in FIG. 4(a) and a histogram in the X direction as shown in FIG. 4(b) are obtained. The "troughs" in the histograms in Figures 4(a) and (b) are considered to be spaces between characters and lines, respectively. 3.
As shown in the figure, when the text has approximately constant font size, the histogram has periodicity as shown in FIGS. 4(a) and 4(b). However, if characters of different sizes are mixed, or if figures or the like are included, the periodicity of the histogram shape will be lost.

Generally, about 80% of the total number of characters on one page are the same size. Therefore, if the total distribution of the number of black dots in each direction shown in FIG.
+ "lz)+(X3+y3)s"... is obtained. Therefore, (X2-xt)I (X3-X2)l
(XA -Xs)s...(XA -Xn-t)
-1 and (yz -yt)l (y3-7z),
(ya-73).

(ys-ya L...(V4 yn-i)*
Figure 5 (a) is obtained by calculating and creating a histogram.
), (b) are obtained by Mx, My, the mesh size obtained in step 20 is X1 pixel in the X direction, and X1 pixel in the X direction. If you divide a character string with a mesh of this size, most characters will be included in each mesh, one at a time.

Furthermore, it is also possible to determine the font size with precision (X+-X
't), (X2 X'z) = (X-+.

x 'w ) and (yt -y '1) r (yz
-y'2)...(y to -y'4), convert it into a histogram, and find its maximum value.

Let the character area thus determined be Mx'XMy' in FIG.

Figure 6 shows how the text in Figure 3 has been divided into meshes of size MxXM7 (also showing Mx'XM7' recognized as one character area). The mesh according to the invention can contain not only one character but also a blank part of the background. As will be described later, the compression rate of the encoding method of the present invention is greatly improved because the encoding is performed including the blank areas and characters.

The encoding method described above can be expected to have an extremely high compression rate for documents with uniform font sizes, but for general documents, the font sizes are uniform as shown in Figure 3. Since there are only a few cases in which data is displayed, and most of the time it includes graphics and photographic areas, it is not possible to expect an improvement in the compression rate even if the entire page is divided and encoded using the mesh described above.

Therefore, in the next step 21, the algorithm for detecting areas to which the mesh cannot be applied will be explained in detail.

<Step 21> Determination of unsuitable image areas by mesh division The following are examples of image areas in which cars are inappropriate to be divided by mesh.

■Characters (text) of different sizes ■Graphics, photo areas ■Character areas with non-white background (background cross) ■Proportional printing manuscript Step 21 is where the images of This is a method for identifying them when they are mixed together, and will be explained below.

If the character string in Figure 6 and the characters with different sizes (Figure 7) are mixed, the mesh division × obtained by the method in step 20 above is used! When applied to character strings of different sizes, they may be divided as shown in FIG.

For example, regarding mesh M-1, some characters are included even in the blank area below the mesh (space between lines). Therefore, by specifying the character area within the mesh and checking the presence or absence of black pixels outside the character area, image areas containing characters of different sizes can be identified. At this time, to determine the character area within the mesh, a mesh composed of the smaller of Mx and My, that is, since Mx>My in FIG. 6, HxXMx may be determined as the character area within the mesh, In addition, to obtain even more accurate results, instead of simply obtaining the character spacing as mentioned above, you can directly calculate the character area quality.
The presence or absence of black pixels outside the character area may be checked from '×1'.

By the way, even if the characters have different thicknesses and sizes, there may be cases where there is no black dot in the above character area, as shown in 14 in the w47 diagram. However, the adjacent mesh 3 can be clearly identified as being incompatible with the mesh. That is, in step 21, compatibility is determined for each mesh, and in the next step 22, non-conforming meshes are determined two-dimensionally, and non-conforming regions can be determined.

It is also possible to determine non-conforming areas by the above-described processing for the figure (2), the photo area, and the area (2) having image information in the background.

However, in FIG. 6, if there is a straight line parallel to the X axis in the character area of the adjacent mesh, the above process cannot determine the non-conforming area. Therefore, Mx, M7
It is determined whether there are several black dots on the longer mesh, that is, in this example, the Y-axis direction. If there are, the mesh area divided along the axis is determined to be unsuitable. Therefore, the straight line described above can be identified.

Step 22> Separation of encoded regions In this step, one image is divided into two parts based on the incompatibility determined in step 21, depending on the encoding method.

■MH (Modified HuHs+an method),
MR (Modified R, E, A, D method), etc., area to be coded in pixel dot dimension ■ Area to be character coded based on the mesh division mentioned above, that is, most of one page (including white area) according to the present invention. ) is divided into meshes including one character depending on the character size, so it is possible to use the code encoding described below, but for halftone parts, figures, and line drawing areas, it is not possible to apply the existing dot-dimensional encoding. is desirable.

For example, if one page is divided into meshes as shown in Figure 8 of Kinshi, and it is assumed that there are meshes (denoted by ■) that are determined to be nonconforming in step 21, then the coding regions can be separated as follows, for example. .

The mesh rows that are continuous in the X direction are called the Yl mesh line and the Y2 mesh line.
Mesh line...P...Y 28 Named mesh line. If there is even one non-conforming mesh in each mesh line, the mesh line is encoded in the dot dimension in the X direction.0 In this embodiment, MH code .

Therefore, in FIG. 8, Y2 to Y4, Y6,
Y14 to Y18 Y2[i's mesh line is MH encoded, and all others are mesh divided character encoded according to the present invention.

Note that in the mesh division according to the present invention, areas where no black dots are present are treated as compatible meshes, so that the compression ratio can be improved. In addition, in order to improve the separation accuracy mentioned above, two rows of compatible mesh lines in the y direction sandwiching the non-conforming mesh line mentioned above are treated as non-conforming mesh lines.
We propose to perform H encoding.

In step 21> and step 22>, one page is divided into MxXMy, and one character is stored in the mesh. Since the determination has been completed, MH encoding is performed on the non-conforming mesh in step 23.

- For characters within a mesh, character recognition is performed for each mesh.

Various methods have already been proposed for this type of recognition method, and basically any method can be applied. In this embodiment, D P (D7namicPatte
rn) Using the matching method. The DP matching method is a pattern matching method based on dynamic programming, and when calculating the distance between an input cover pattern and a registered dictionary pattern, it is This is a matching method that minimizes the distance.

The dictionary patterns used are Common Kanji 2000 and other fonts, and each recognized character is encoded into a 2-byte 2x ASCII code, for example. Step 24>
. . . Storage of data The storage in the data storage means is performed page by page, and one page is further divided into records for each mesh line. Parameters for each page include mesh size Hz, )ly, and parameters for each mesh line include a code type flag added to the beginning of each mesh line to indicate whether or not the encoding is applied. In this embodiment, the code type flag indicates that the mesh to which pixel dot encoding is applied is M.
In order to apply H encoding, two types of encoded data that are switched for each mesh line may be used.

Next, the control flow in FIG. 2 will be explained in more detail based on the flowchart in FIG. 9.

In step 100, mesh sizes Mx and My are determined by the method described above.

Step 102> The image signal for one page in the page memory 5 is divided into NxXMy meshes.

Step 104> One line of mesh array having a width of 1 in the y direction is extracted.

Step 106> Image signals are extracted one mesh at a time from one line extracted in step 104.

Step 108> Determine whether there is a black dot outside the character area of the mesh.

Characters of different sizes and images such as photographs are distinguished from characters of a standard size. There is a black dot (YES)
If so, it is determined in step 114 that it is non-conforming.

Step 110> If the determination in step 108 is that there is no black dot outside the character area (No), step 110 is further performed.
It is checked whether a black dot exists on the Y-axis, and it is determined whether a straight line parallel to the X-axis direction exists in the mesh. If it exists, it is determined that it is non-conforming (step 114).

If there is no black dot in any of the above cases,
The mesh is determined to be suitable (step 112).

Step 116> In step 118, it is determined whether all the meshes in the mesh row having a width of NY have been determined to be conforming/unconforming. If the determination of all meshes has not been completed yet, the process returns to step 106 and the above flow is repeated.

Step 118> When the determination is completed for all the meshes of the one mesh line, the determination result of conformity/nonconformity is checked in step 118. If there is even one mesh that is incompatible, M and H encoding is performed (step 126), and the encoding type flag and Tersinati are set to indicate that M and H encoding has been performed.
ng Code and Make-up Code (Step 128). Steps 120 to 124> If all meshes within one mesh are determined to be compatible, perform mesh division character encoding (Step 120).
), the characters in the mesh are recognized as described above or according to P matching and converted into 2-byte ASCII code.

Steps 130 to 134> Encoding type code and Ter+*i of each mesh line
nating Code, etc., and if the mesh line is the first line of one page, Mx and My are added as data and stored in the data storage means.

Steps 136 to 140> Steps 104 and subsequent steps are repeated until all pages are completed.

Decryption is done as follows. A record for each mesh line of each page is read from the data storage means 4, and the mesh size data Mx, M stored in the data of the first line.
Based on y, for example, in the case of this embodiment, a line memory for the number of pixels in the X direction (X87) is prepared, and mesh lines subjected to MH encoding are decoded line by line. Converts the corresponding character from a font ROM prepared from two-byte character knee into a size that can be stored within the mesh size and reduces it to the dot level. Furthermore, all areas outside the character within the mesh are decoded as white.

The above process is repeated for each mesh line, and 1
Decrypt the page.

As explained above, this embodiment focuses on the fact that most of the characters in a single document are uniform in size, and performs character recognition and encoding using a mesh that includes space between lines and the characters. However, as a modified example, even if it is determined that dot-dimensional encoding should be applied after separating the encoding regions in step 22, if that region only contains characters of different sizes, , by further performing a second mesh division on that area and performing character recognition again, second mesh division character encoding is possible and even more efficient encoding can be realized.

Further, the present invention can be applied to a proportional character document by recognizing the size of the characters and then allocating them in a mesh and re-editing them.

[Effects] As explained above, according to the image processing device of the present invention, image data with an increased data compression rate can be efficiently stored,
Also, during decoding, high-speed decoding becomes possible by knowing the area type and area size.

Furthermore, the present invention can also be applied to character cutting technology using OCR, and it is possible to accurately and selectively cut out characters in papers of various unknown formats, from newspapers to magazines.
This in turn leads to an improvement in the recognition rate. Further, the data read in pixel units can be extracted by other extraction methods, or can be displayed on a display as unreadable characters.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment according to the present invention, FIG. 2 is a schematic diagram of a processing flow of the embodiment, and FIG. 3 is a diagram visually representing an input document in a page memory. Figures 4 (a) and (b) are histograms in the x and y directions, respectively. Figures 5 (a) and (b) are diagrams showing how to determine the mesh size in the x and y directions, respectively. The figure shows the input document divided by the determined mesh. Figure 7 shows the same mesh applied to characters of different sizes. Figure 8 shows the conformance/inconformity of the mesh-divided image signal of one page. A diagram of an example of the results determined. FIG. 9 is a control flowchart of the control section. In the figure, 5: page memory, 3: mesh division encoding processing section, 4: data storage means, 9: font ROM. Fig. 1 Fig. 2 Fig. 3 Fig. 4 (b) Salvation Fig. 5 (0) Degree Ei

Claims (2)

[Claims] (1) A dividing means for dividing a read image of a document into at least two types of areas according to the image tone thereof, an encoding means for encoding each area based on a different encoding method, and an encoded storage means for storing image data, and the storage means stores, in addition to the image data for each area, the type of the area and the size of the smallest area in the area. Device. (2) The image data stored in the storage means is read out, and the read image data is switched and decoded for each region based on the encoding method. Image processing device.