JPH02116985A - Automatic input device for document - Google Patents

Abstract

PURPOSE:To extract a character string corresponding to a meaning item to improve the productivity of data input by dividing character strings extracted from a document picture by respective meaning items to output them with respect to a document having a list-form structure like a list of names. CONSTITUTION:A character candidate extracting process 3 which takes in document picture data 2 outputs the area, which is diagonally designated with a pair of coordinate values (X1,Y1) and (X2,X2) of a circumscribed rectangle of a black picture element connection area, as a character candidate area to a character recognizing process 4. The process 4 convertes picture data in the character candidate area extracted by the process 3 to a character code by character recognizing technique. A character string extracting process 5 extracts a character string in accordance with positional relations of character candidate areas. For example, when X1(K)<=X1(K+1)<=X2(K) or X1(K)<=X2(K +1)<=X2(K) (K is the index of the character candidate area) is true with respect to coordinates of two character candidate areas adjacent to each other in the case of a laterally written document, it is judged that these two areas belong to the same character string.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to an automatic document input device that encodes and outputs a document input as a single document image.

(Prior art and its problems) Conventionally, there have been document input devices that read characters in documents consisting of continuous character lines such as sentences, or in pre-specified areas such as slips, but There has not been a document input device that can input a tabular document in which the meaning items are constant depending on the position (means for solving the problem) The present invention was created in view of the above problems.

This document automatic input device encodes and outputs a document input as a character image, and proposes a document automatic input device that divides a character line into semantic items with a length of 19 based on the beginning of the character line of a manually inputted document. It is.

(Operation) In the present invention, for a document having a tabular structure such as a list,
This method divides character lines extracted from a document image into semantic items and outputs them.

(Embodiment) An embodiment of the present invention will be described with reference to the block diagram of FIG. An automatic text input device 1 to which the present invention is applied includes document image data 2 inputted from an input part of an image scanner, a copying machine, etc., a character candidate extraction process 3, a character recognition process 4, a character string extraction process 5, and a structure. It consists of an analysis process 6, document code data 7 and divided data 8 as data.

In the character candidate extraction step 3 that takes in the document image data 2,
The coordinate values (Xo, Y,) of the circumscribed rectangle of the black pixel connected area and (
The area specified diagonally by the pair (Xz, Yz) is output to the character recognition step 4 as a character candidate area.

In the character recognition step 4, the image data of the character candidate area extracted in the character candidate extraction step 3 is processed using a known character recognition method (for example, see "Pattern Recognition and Its Applications" edited by Kazuo Nakata, published by Corona Publishing).
Convert to character code by

In the character string extraction step 5, character strings are extracted based on the positional relationship of character candidate areas. For example, in the case of a horizontally written document, the coordinates of two adjacent two-character candidate areas.

X□(K)≦Xyu(K+1)≦X2(K)... (1) or Xl(K)≦X, (K+1)≦X, (K)...
When (2) holds true, it is determined that the two areas belong to the same character string. Here, represents the index of the character candidate area.

Figure 2 shows the document situation with the back structure omitted.

In the structure analysis step 6, one character string is determined to be one line, and the character string is divided into meaning items based on division data specified in advance with a length based on the beginning of the line. for example,
In the example of Figure 2, 1. is the name, ■□-1, is the address, l, -1
Divide the string so that 2 is the phone number. The divided string is 1. The document code data 7 is output as a character string whose meaning items can be distinguished by dividing them by .

Furthermore, in character string extraction, it is also possible to combine separated characters into one character by using dynamic programming5.
Also, in order to eliminate the effects of indentation and noise at the beginning of the line, 5
It is desirable to use a statistical estimate of the line start coordinate value of each character string as the line start coordinate value. This makes it possible to deal with a certain degree of inclination of the document. Furthermore, although the data is in tabular format, we have explained without ruled lines, but if there are ruled lines in the data, there is no problem if you delete the ruled lines by image processing in advance. Semantic items can also be separated by detecting their positions.

(Effects of the Invention) According to the present invention, character strings corresponding to semantic items can be extracted from a single document image for a document having a tabular structure such as a directory, so data input productivity can be significantly improved. Can be done.

[Brief explanation of the drawing]

The drawings show one embodiment of the present invention; FIG. 1 is a block diagram of the present invention, and FIG. 2 shows a document data situation.

Claims (1)

[Claims] A document automatic input device that encodes and outputs a document input as a character image, characterized in that the character line is divided into semantic items by lengths based on the beginning of the character line of the input document. .