JPS60210069A - Document information coding method - Google Patents

Abstract

PURPOSE:To code document picture information with high efficiency by dividing data for each series of black picture elements continuous to each other, extracting the data section of a prescribed size to recognize characters and to obtain character codes and at the same time converting other pitcure data into run length codes. CONSTITUTION:The document picture information is scanned based on the position of an upper left picture element and a square window defined in X and Y axes to obtain black picture elements. Then a notice picture element is set at the black picture element. The black picture elements adjacent to the notice picture element are obtained, and the window is shifted along continuous black picture element groups, i.e., the contour of a pattern. Then the minimum coordinates Xmin and Ymin and the maximum coordinates Xmax and Ymax are obtained among coordinates X and Y of the black picture element along which the window has a round. Then a square pattern area ARC passing through picture elements D (Xmin, Ymin) and (Xmax, Ymax). The horizontal and vertical lengths WX and WY of the area ARC are compared with their reference values respectively. When the extracted pattern is equal to a character pattern, the characters are recognized with the dictionary data and coded. While the extracted pattern equal to a graphic pattern is kept as it is, and other picture data are converted into run length codes.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to the encoding of document information that can be employed in a device, such as a facsimile, which reads and encodes document information including figures and characters.

■Prior Art Documents transmitted by facsimile usually contain text information and graphic information. In a typical facsimile, an image is scanned, data of each minute pixel making up the image is sequentially read, and the collection of data is directly converted into a run-length code and transmitted. When the density of information contained in a document is low, such as when it contains only graphical information,
Run-length encoding significantly reduces the number of bits to be transmitted. However, in the case of a document containing character information, the information density is high, and the number of bits to be transmitted does not decrease much even if run-length encoding is performed.

Since the types of character information itself are limited, if the character can be identified, it can be expressed as a single piece of information with a small number of bits (16 bits for Kanji). Therefore, in the same way as normal character recognition, document information is cut out for each predetermined line, then divided into rectangular small areas, and character recognition is performed on each small area by pattern matching. A method has been proposed (USP-4,091,424) in which if a pattern is recognized, the character code corresponding to that character is transmitted, and if a pattern cannot be recognized, run-length coded information about that area is transmitted. has been done.

However, in the case of a document that includes large figures, one figure is divided into many areas by cutting out characters, and as a result, some of the lines that make up the figure are divided into numbers.
``!'' or hyphen 7'#l may be misrecognized and inaccurate information may be reproduced. Also, in the case of complex figures, the transmission efficiency will be lower than usual.

■Purpose The present invention efficiently encodes document information including character information and graphic information into information with a small number of bits.
The purpose is to prevent incorrect encoding due to misrecognition.

■In order to efficiently encode the constituent document information using a small number of bits, it is best to convert the character information contained in the document into a character code and output it. However, if patterns (characters) are cut out using the same method as conventional character recognition, the above-mentioned juxtaposition will occur.

This type of juxtaposition occurs because a pattern that should originally be one is divided into multiple regions. Therefore, for example, if the data is divided into a series of consecutive black pixels (or white pixels), one pattern data will not be divided into a plurality of pieces. Furthermore, in the case of characters, the size of each pattern generally falls within a predetermined range. Therefore, it is sufficient to extract only those of a predetermined size from the divided pattern data and perform character recognition on them. For patterns in which characters can be recognized, if the character code is output and the pattern data is deleted from the image memory that stores the pattern data, the amount of effective information in the image data remaining in the image memory will be reduced. If the image data is run-length coded as in the conventional method, efficient coding can be achieved.

Hereinafter, the present invention will be described in detail with reference to the drawings.

A case will be described in which a document image as shown in FIG. 1 is read. Referring to FIG. 1, this image includes a graphic pattern PTI and character patterns PT2 and PT3. Furthermore, since these patterns are close to each other, character patterns PT2 and PT3 cannot be successfully extracted using the character extraction method used in conventional optical character reading (OCR). , binary information corresponding to black and white obtained by optically reading this information is stored in advance in a buffer memory, and the contents of the memory are processed as follows.

Starting from the upper left pixel position of the document image information, a 3×3 matrix register or window as shown in FIG. 1 is sequentially scanned in the X-axis direction and the Y-axis direction to find a black pixel.

When a black pixel is found, the pixel of interest is adjusted to that position. Next, a black pixel adjacent to the pixel of interest is found, and the window is moved along the outline of a group of consecutive black pixels, that is, one pattern. When the window goes around the outline of the pattern, the smallest of the coordinates X and Y of the black pixels passing through it is Xm
Determine 1n and Ymin and the maximum Xmax and Ymax.

One pixel D (Xmin, Ymi
n) and another pixel D (Xmax, Ymax
) is defined as one pattern area ARC. As shown in Fig. 1, the horizontal length WX and vertical length WY of the pattern area ARC are relatively small when it is a character, but when it includes a figure, the horizontal and vertical lengths are relatively small. At least one becomes larger.

Therefore, the sizes WX and WY are compared with predetermined reference values RefX and RefY, respectively, to determine whether the extracted pattern area ARC is a character pattern or a graphic pattern.

If the pattern area ARC is determined to be a figure, nothing is done and the next pattern is searched, but if it is determined to be a character, the pixel data of this area is compared with dictionary data prepared in advance and the pattern area ARC is searched for. (other methods may be used) and perform character recognition processing. As a result, if the recognition is successful, the obtained character code and the upper left coordinate data of the character pattern, that is, Xm1n and Y
Generate a character code data group including min.

As shown in Figure 1, multiple character patterns (PT2゜PT
3) are lined up (determined by comparing X min and Y min of each pattern), these are regarded as a series of characters and included in the same character code data group. In this case, the coordinates of the second and subsequent characters are expressed as relative coordinates dXY with respect to the coordinates of the previous character (for the second character, X ll1in , Y min ).

In addition, when a character is recognized, the pattern area AR
The data in the buffer memory corresponding to C is made the own pixel data. In other words, the black pixel data that existed up to that point is erased. Therefore, if this process is repeated, the character pattern components will disappear from the data in the buffer memory.

A schematic flow of the above processing is shown in FIG. 2a.

Further, an example of the structure of the character code data group is shown in FIG. 3a. Referring to FIG. 3a, CC is an identification code for identifying that a character code data group exists after that, DN is the number of characters existing in this data group, CHI, CH2, . . . CHn is the code of each character (for Kanji characters, each consists of 16 bits), and C/R is a carriage return code indicating the end of this data group.

When the above processing is performed, the image data contained in the buffer memory will consist of only graphic components and slightly unrecognizable character pattern components, so the amount of information contained in this data is actually small, and it cannot be run as is. Even if it is length encoded, the amount of data can be reduced quite efficiently. However, in this embodiment, in order to further reduce the amount of data, line recognition is performed as follows.

A case will be described in which a linear pixel group as shown in FIG. 4 is processed. In the same way as above, move the window horizontally and vertically to find the beginning of the remaining black pixel (D in Figure 4).
Find SX, 5Y)). Black pixels adjacent to the pixel of interest are prioritized as shown in FIG. That is, in this example, the black pixel located at the upper right of the pixel of interest has the highest priority, so if the pixel at this position is a black pixel, it is ignored even if other pixels are black pixels.

In the starting point pixel D (SX, SV), the black pixel located at the center right thereof has the highest priority. Therefore, as indicated by the arrow shown in FIG. 4, the pixel of interest is moved to the position of a black pixel with a higher priority. Depending on the direction of movement at this time,
Direction data (any one of 1 to 8) as shown in FIG. 5 is generated. The direction data generated here is the counter CN
It is stored in the direction memory address associated with the value of T.

Next, the contents of the counter CNT are incremented by 1, and the position of the pixel of interest is (sx+i,
sy) in case of (SX, 5Y-1).

(sx, sy) and (SX, SY+1) coordinates),
Update black pixel data to white pixel data. When this process is repeated, in the case of the linear pattern shown in FIG. 4, only some hatched pixels remain as black pixels, and all other black pixels are erased.

When the pixel of interest reaches the final pixel of the linear pattern, the pixel of interest itself is also erased, and line code data is generated with reference to the start point data SX, SY, the counter CNT, and the contents of the direction memory. That is, as shown in FIG. 3b, the code CL for identifying that the following data is a series of line code data, the X coordinate of the starting point (SX), the Y coordinate of the starting point (SY)? A series of code data consisting of a data number (CNT), each direction data, and a carriage return code C/R indicating the end of data is generated.

Line recognition is not performed when the length of the pattern is short or when the pattern is complex. For other patterns, the above process is repeated to convert them into line code data. When this process is finished, the amount of pixel data remaining on the buffer memory becomes very small. This remaining data is run-length encoded as is done in general facsimile machines, to generate a run-length code as shown in FIG. 3b. The code CR shown in Figure 3b is used to identify that the data that follows is a run-length code, EOL is a synchronization code to distinguish between scanning lines, and EOF is used to indicate that all data has been sent. This is the data end code indicating that the

An outline of the processing procedure after line recognition is shown in FIG. 2b.

■Effects As described above, according to the present invention, it is possible to accurately cut out a character pattern and code it into a character code. Moreover, since the character pattern is deleted from the remaining patterns, the remaining data can be efficiently coded into a small number of bits. sell.

[Brief explanation of drawings]

FIG. 1 is a plan view of a portion of document image information. FIGS. 2a and 2b are flowcharts showing a procedure for encoding image information in an embodiment of the present invention. Figures 3a and 3b are block diagrams showing examples of encoded data arrangements. FIG. 4 is a plan view showing a part of the pattern of the document image. FIG. 5 is a plan view showing the correspondence between the types of pixels cursed by windows used in encoding processing and processing.

Claims (3)

[Claims] (1) Store document information in a predetermined memory, examine the stored information, find a group of consecutive effective pixels, consider them as independent pattern data, and at least The length of one is determined and each pattern data is divided into at least two types, one type of pattern data is subjected to character recognition processing, and when a character is recognized, the code of that character is output. At the same time, the document information encoding method erases the pattern data corresponding to the character from the document information stored in the memory, and then encodes the remaining document information by any means. (2) For the remaining document information, determine a specific pixel of interest, move the pixel of interest to an effective pixel adjacent to it, and output the moving direction as coded information, and at least transfer the information of the pixel of interest to the document information in memory. delete from,
A document information encoding method according to claim (1). (3) The document information encoding method according to claim (1), wherein the remaining document information is output as a run-length code.