JP2890788B2 - Document recognition device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to an apparatus for recognizing a document image and creating an electronic document.

[Conventional technology]

With the development of character recognition technology in recent years, various OCR (Optical
Charactor Reader) has been developed.

As a method of extracting elements such as characters, diagrams, photographs, etc. constituting a paper surface without using prior information on a format, a method using a run length of both black and white pixels, a method using a window of a fixed size, There are a method using a pixel density and a method using a proximity linear density method. These methods are mainly intended to separate a gray-scale figure and a binary figure on the premise of image storage / editing or image transmission, and to separate a figure area and a character area.

On the other hand, as a method of extracting a character string on the premise of character recognition, a method of tracking connected components of black pixels has been reported. In addition, there is a method in which rough information about a format such as the position of a column is given in advance, and a paper component is extracted based on the rough information.

Kida, Masuda, "Paper Constituent Elements Extraction Method without Format Specification Information", IEICE Transactions, 83/1, Vol.J66-
In the paper described in D No. 1, p111 to p118, a field separator, that is, a linear figure forcibly dividing an area on the paper and a headline and a character string of the main body as a constituent element, a paper form is used. There has been proposed a method of automatically extracting these paper components without using any prior information on the position of the paper.

As described above, when the arrangement of the page components, that is, the layout is automatically recognized, it may be erroneously recognized. For this reason, humans have modified the automatically recognized layout.

In the conventional document recognition device, in the work of correcting the recognition result, the correction of the recognition result is performed by correcting only the layout when correcting the layout, and correcting only the character portion when correcting the character portion. . For this reason,
Even if the character recognition is mistaken due to the wrong recognition of the layout structure, the correction is completed only with the layout, and the character portion is not automatically corrected.

Explaining this with a specific example, an original image composed of a two-stage sentence as shown in FIG. ), The portion to be laid out in two stages and the middle character is subjected to layout analysis because the middle character and the character lines in the blocks at both ends are close to each other in the horizontal direction. ), The layout analysis may be performed in such a manner that the intermediate character portion is merged with the adjacent line. In this case, as shown by an area 6 surrounded by a broken line in FIG. 9D, one line on the left, the intermediate literary part, and one line on the right are regarded as one continuous line. Because the context processing is performed, the part that should be contextually processed as "I'm fine. Go tomorrow at 10 roks" is, as shown in Fig. (E), "I'm fine. Go tomorrow 100-7".
Context processing is performed by mistake. The reason why the context processing is performed in this way is that recognition of numbers may be given priority as described later.

Thus, if layout analysis is performed by mistake,
In the knowledge processing of creating a meaningful sentence by performing context processing based on the character recognition result, a normal function cannot be performed, and an incorrect recognition result as shown in FIG. I will.

In such a case, the operator manually corrects the layout.However, conventionally, if the correct layout analysis can be performed, even if the knowledge processing on the character recognition result can be obtained correctly, the operator can modify the layout. Even if you modify
The character recognition result was not automatically corrected.

[Problems to be solved by the invention]

As described above, in the conventional document recognition apparatus, even if an error in the layout analysis is corrected by the operator, the document recognition apparatus does not function to have a normal effect on the character recognition result.

The present invention has been made to solve the above problems, and has as its object to improve the efficiency of correcting a recognition result.

[Means for solving the problem]

In order to achieve the above object, a document recognition device according to the present invention includes an image input unit that inputs a document as an image, a layout analysis unit that extracts a layout of a document image input by the image input unit, A character recognition unit that recognizes a character portion based on the layout information obtained from the analysis unit and outputs candidate characters along with the recognition result, a knowledge processing unit that performs context processing from the character recognition result of the character recognition unit, A correction interface unit for displaying layout information from the layout analysis unit and a result of character recognition by the character recognition unit and enabling layout correction; and a knowledge processing unit based on the layout correction information from the correction interface unit. And a reprocessing instruction unit for instructing reprocessing.

[Action]

According to the present invention, a general document is input as a digital image from an image input unit, a layout of an original image input from the image input unit is extracted in a layout analysis unit, and a layout analysis is performed on an area determined to be a character block. On the other hand, the character recognition unit performs character recognition processing for outputting candidate characters. Next, in the knowledge processing unit, context processing is performed on the result of character recognition including the candidate character to create a meaningful sentence. Next, the correction interface unit displays the result of the character recognition unit that has passed through the layout analysis unit and the knowledge processing unit on the recognition result display unit, and the operator makes a correction through an instruction device such as a keyboard / mouse. When the layout information is corrected as a result of the correction, the reprocessing instruction unit feeds back to the knowledge processing unit based on the layout correction information, and displays a newly knowledge-processed result on the recognition result display unit. As described above, a target document is created by repeating the correction process.

〔Example〕

Hereinafter, features of the present invention will be specifically described based on embodiments shown in the drawings.

FIG. 1 is a configuration diagram of a document recognition device in the present embodiment, and FIG. 2 is a configuration diagram of a correction unit used in the document recognition device.

The document recognition apparatus includes an image input unit 1 for inputting a document as an image, a layout analysis unit 2 for extracting a layout of the input document image, and character recognition based on the obtained layout information. A character recognition unit 3 that also outputs candidate characters, a knowledge processing unit 4 that performs grammatical processing and the like from the result of character recognition, and a correction interface unit 51 that displays layout information and the result of character recognition to enable layout correction. And a reprocessing instruction unit 52 for instructing the knowledge processing unit 4 to perform reprocessing based on the correction information.

Further, the correction interface unit 51 and the reprocessing instruction unit
52 and a correction unit, and a correction interface unit 51
A keyboard / mouse 511 as an instruction device, a recognition result display unit 512 for displaying a recognition result in the character recognition unit 3,
It comprises a correction processing control unit 513 that receives the result of character recognition. The correction processing control unit 513 has a function of creating a document based on an input from the keyboard / mouse 511,
That is, it has a word processor function.

 Next, the operation of the above-described document recognition device will be described.

First, a binarized document image input by the image input unit 1 is
The analysis is performed by the layout analysis unit 2. The analysis method is a method of analysis proposed in the above-mentioned reference document "paper component extraction method not depending on format specification information", that is, a field separator (a linear figure forcibly dividing an area on the paper). ) And a heading and body character string as constituent elements, using a method of automatically extracting these paper constituent elements without using any prior information on the position on the paper, etc. Character recognition is performed by the character recognition unit 3 on the block determined to be.

A known method is used as a method of character recognition. The character recognizing unit 3 outputs a plurality of candidate characters that are likely to be the character in addition to the first recognized character as the certainty of the recognition.
The knowledge processing unit 4 performs a grammar process or the like from the recognition result output from the character recognition unit 3 and performs a process of converting it into a character string conforming to the prepared grammar.

An example of this processing will be described using the image shown in FIG. 3 as an example. Now, it is assumed that the recognition result including the candidate characters output from the character recognition unit 3 for the original character string is as shown in Table 1.

As shown in Table 1, from the recognition result including the candidate characters output from the character recognition unit 3, the character immediately before the currently focused character in the character string of the first degree of certainty is a numeral. If there is, there is a high probability that the current character is also a number. Therefore, if the first character in the certainty is not a number and the second character in the certainty is a number, the second character is output as a recognition result. Do the same for kanji, hiragana, and katakana other than numbers. By this processing, the character string “10
If "Rourke" is composed of only the character having the first place in the certainty of the recognition process, it is correctly recognized as the character string "10 Rourke".
Here, if there is “0-7” as the second-place candidate character in the likelihood of the recognition result “Rook”, the above-described knowledge processing will result in an intermediate character, that is, the line number “10”.
Together to generate a character string "100-7".
As described above, the result of the character recognition unit 3 is automatically corrected by the knowledge processing unit 4.

However, in this processing, if the character string is not correctly cut out as a processing target, that is, if the processing of the layout analysis unit 2 is not analyzed as a target layout, as shown in FIG. It becomes processing to lower. The result of the layout analysis, that is, the result of character recognition (including candidate characters) that has been subjected to knowledge processing such as the type of block, the position and size of each character, and the character recognized at which position in which block A table indicating whether or not it is present is also sent to the correction unit 5.

In the correction unit 5, first, in the correction interface unit 51, the correction processing control unit 513 receives the result of the layout analysis and the result of the character recognition that has been subjected to the knowledge processing, and displays the layout and the character recognition result on the recognition result display unit 512. .

Then, the operator makes a correction using an instruction device such as a keyboard / mouse 511 while looking at the recognition result display section 512.

At the time of display, blocks of characters only are color-coded such as blue, and blocks of figures are red, so that the type of block can be easily understood by the operator.

Further, it is also possible to color-code a block between a corrected block and an uncorrected block. The modification procedure is first modified from a layout that is easy for the operator to see. The modification changes the type of block (a block of only characters or a block of a figure, etc.), its position and size, and the like. As a result, the table held by the correction processing control unit 51 and containing information such as the position and size of the block and the characters included in the block is rewritten.

Then, when the layout modification is completed and the target layout is created, the reprocessing instruction unit 52 instructs the knowledge processing unit 4 to reprocess the changed layout portion.

Thus, if a layout analysis as shown in FIG. 3C is performed on a document having a layout as shown in FIG. 3A, the operator can correct the layout as shown in FIG. 3B. In addition, the knowledge processing section can correctly perform the knowledge processing even on the area 6 subjected to the erroneous layout analysis.

Finally, an error in the character recognition result is corrected using the word processing function of the correction processing control unit 513 for the character portion. Through these processes, an operator creates a target document.

In the above example, the embodiment in which only a document is created has been described, but the present invention can also be used for slip input.

Further, in the above-described embodiment, the reprocessing instruction is issued to the knowledge processing after all the layout corrections are completed. However, the reprocessing instruction is issued to the knowledge processing every time each layout correction is performed. It is also possible.

In the above embodiment, only the reprocessing to the knowledge processing has been described, but the reprocessing to the character recognition is also possible.

By the way, the reprocessing instruction for character recognition is issued
This is a case in which extraction of a character, which is the minimum unit, in layout analysis has failed.

More specifically, if an accurate character cutout fails for the character "Hayashi", the character is recognized as "tree" or "tree". In this case, the character recognition unit is instructed to reprocess by correcting the two rectangles surrounding the characters "tree" and "tree" to be one character.

〔The invention's effect〕

As described above, according to the present invention, the result of the layout correction is processed by the knowledge processing unit, so that the knowledge processing is completely performed, and the efficiency of the character correction operation at the time of document creation is significantly improved. It has the effect of

[Brief description of the drawings]

FIG. 1 is a block diagram of a document recognition device showing an embodiment of the present invention, FIG. 2 is a block diagram of a correction unit used in the document recognition device, and FIG. 3 is an explanatory diagram of problems in a conventional example. is there. 1: Image input unit, 2: Layout analysis unit 3: Character recognition unit, 4: Knowledge processing unit 5: Correction unit 51: Correction interface 511: Keyboard / mouse 512: Recognition result display unit, 513: Correction processing control unit 52: Reprocessing instruction section 6: incorrect layout analysis area

Claims (1)

(57) [Claims] An image input unit for inputting a document as an image;
A layout analysis unit for extracting a layout of the document image input by the image input unit; and a character recognition unit for character recognition of a character part based on the layout information obtained from the layout analysis unit and outputting candidate characters along with a recognition result. Unit, a knowledge processing unit that performs context processing based on the result of character recognition by the character recognition unit, and the layout information from the layout analysis unit and the result of character recognition by the character recognition unit are displayed to enable layout modification. And a reprocessing instruction unit for instructing the knowledge processing unit to perform reprocessing based on layout correction information from the correction interface unit.