JPS62133585A - Word segmenting system - Google Patents

Abstract

PURPOSE:To segment exactly a word from image information which has been read from an original of European language, etc., by deciding a break of a projection of width exceeding a deciding threshold value, to be a gap between words. CONSTITUTION:A vertical projection extracting circuit 20 derives a projection in the vertical direction of sentence line image information which has been inputted from a line buffer 16. Also, to a gap histogram generating circuit 22, information of a vertical projection of one sentence line is inputted from the circuit 20, and a gap histogram is generated by calculating a frequency of width of a break of the vertical projection. This histogram is inputted to a deciding threshold value determining circuit 24, detects the maximum width corresponding to a peak and sends it to a word segmenting circuit 26. The circuit 26 discriminates the break of the vertical projection of width exceeding a word gap deciding threshold value, as an inter-word gap, from the information of the vertical projection which is inputted from the circuit 20, and segments image information of a word which has been delimited by the inter-word gap from the sentence line image information.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to a word extraction method 11 for cutting out words from read image information of a Roman manuscript or the like.

[Prior art]

In document processing devices, when processing Western languages, it is not only necessary to process single characters, but also words! Processing of the ones place is often necessary.

For example, in OCR that processes Roman characters, it not only recognizes individual characters, but also corrects or prevents recognition errors by comparing words, which are collections of recognized characters, with a word dictionary. May perform word-by-word processing, such as estimating possible characters. In this case, it is necessary not only to cut out text lines and characters from the read image information of the European manuscript, but also to cut out J5.

Similar 1'L word extraction processing is essential in computer translation systems that translate European texts into Japanese texts.

Conventionally, such word extraction processing takes a vertical projection of the image information of a line of text extracted from a read image, and then calculates the cut of the projection by using a fixed Qt interword gap judgment threshold set in advance. , the projection break with a width greater than the judgment threshold [1 is regarded as a word break, and 1 sentence line image information to 1
11 words have been extracted.

However, European magazines are generally printed with proportion pitch, and the gap widths between characters and words are not constant, and if a certain threshold is used, an error in determining the gap between words may occur. As a result,
Word segmentation errors are likely to occur.

Further, even if the manuscript is not necessarily a European manuscript printed with a proportional pitch, a method using a fixed determination threshold may frequently cause word extraction errors. This is the case, for example, in the case of a European manuscript in which characters of different sizes are mixed.

In other words, the width of the inter-character caps printed in large font sizes may be larger than the width of inter-word gaps printed in small font sizes, so why not use a fixed judgment threshold as in the past? There is a high probability that a word segmentation error will occur due to the inability to correctly determine the lj-nri gap.
).

[verbal]

An object of the present invention is to provide a word extraction method for reliably extracting words from image information read from a European manuscript printed with proportional pitch printing, a European manuscript containing human-sized characters, etc. It's about doing.

〔composition〕

The word extraction method of the present invention, which was made to achieve this purpose, takes a vertical projection of the image information of a text line, creates a histogram of the widths of the projection breaks, and calculates the maximum width corresponding to the peak. According to this method, a threshold for determining an inter-word gap is determined, a projection break having a width equal to or greater than the determination threshold is determined to be a gap between words, and a word is extracted from image information of a text line.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a schematic block diagram showing the configuration of the main parts of a document processing device to which the word extraction method of the present invention is applied. In the figure, reference numeral 10 denotes a scanner for reading a document (for example, an English document), and image information of the document read by this scanner 10 is stored in an image buffer 12. The image information from this image buffer 12 is extracted by the line cutting unit 14.
are input sequentially into the text line, and the image information of the text line is cut out.

This line cutting process is performed, for example, by a general projection method in which a horizontal direction (line direction) projection is obtained and the valleys of the projection are cut out. Of course.

Gyokiri Oshishi may be performed using an improved projection method that takes projections for each partial region. The image information of the cut out sentence lines is temporarily stored in the line buffer 16, and then sent to the word cutting unit 18 in the subsequent stage.
is input.

This η1 word extraction unit 18 includes a vertical projection extraction circuit 20, a gap hisdog 911 creation circuit 22, 'l'-l! It consists of an n-gap judgment threshold determining circuit 24 and a word cutting circuit 26.

The vertical projection extraction circuit 20 is a circuit that obtains a projection of the text line image information input manually from the line buffer 16 in the vertical direction (direction perpendicular to the text line). The continuous part of this projection corresponds to the range of characters, and the break "1 (mountain part)" of the projection is the gap between characters or the gap between 111 words and 111 words.

As mentioned above, in European manuscripts with proportional pitch printing, the inter-character gaps and inter-word gaps fluctuate widely, so it is necessary to appropriately set the judgment threshold for identifying word 1jlT gaps for each line of text. There is a need to. The circuits for this purpose are a gap histogram creation circuit 22 and an inter-word gap determination threshold determination circuit 24.

The gap histogram creation circuit 22 receives information on the vertical projection of one text line from the vertical projection extraction circuit 20, counts the frequency of the width of the gap in the vertical projection, and creates a gap histogram. For example, assume that image information of a Roman manuscript as shown in FIG. 2 is input. In this figure, LL, L2. Each L3 means a line of text, each shaded area means a word,
The spaces between words! 1 means the gap between words.

If the image information of the text line of L2 is now input to the word cutter 71'j part 18, the gap histogram 1 as shown in FIG.
- grams are obtained. A in this gap histogram
The range B corresponds to the inter-character gap, and the range B corresponds to the inter-word gap.

The inter-word gap determination threshold determination circuit 24 receives information on such a gap histogram and detects the maximum width (Gm) corresponding to the peak of the gap histogram. However, the peak has a frequency value of a predetermined value (for example, 2) or more.

For example, in the case of the gap histogram shown in FIG. 3, the gap width corresponding to the peak in range B is detected as Gm.

Then, the inter-word gap judgment threshold determination circuit 24 calculates the inter-word gap judgment threshold Gt by the formula % (where Gδ is O<Gδ (constant of Gm)), and calculates the inter-word gap judgment threshold G1, The information is sent to the word extraction circuit 26.

11 The narration circuit 26 extracts ! from the vertical projection information inputted from the vertical projection extraction circuit 20. A break in the vertical projection with a width equal to or greater than an inter-ri gap determination threshold is identified as an inter-word gap, and image information of words separated by the inter-word gap is extracted from one sentence line image information.

For example, assume that image information of two text lines whose vertical projections are as shown in FIG. 4 are sequentially input to the word extraction section 18. In the figure, the hatched range is a continuous range of vertical projection, and each corresponds to one or more character strings or words. In the case of proportional pitch printing, the gap width in the vertical projection of one text line varies as shown in this example.

In such a sentence line, a vertical projection cut Ll with an arrow ↑ is determined to be an inter-word gap, and a word is cut out.

In this way, the inter-word gap judgment threshold is dynamically determined based on the gap histogram of the vertical projection of text lines, so even in the case of proportional pitch printed documents or documents with characters of different sizes, it is possible to It is possible to cut out the exact number (6) that has definitely identified the cap.

Note that the function of the word extraction unit [8 may be realized by software processing in the software 1 using a microprocessor or the like. In addition to this, the present invention can be implemented with various modifications.

〔effect〕

As is clear from the following detailed description, the present invention dynamically determines the gap judgment threshold between words based on the gap histogram of the vertical projection of text lines to cut out words. Words can be accurately extracted even in a document containing 4A characters of different sizes.

[Brief explanation of drawings]

FIG. 1 is a schematic block diagram showing only the essential configuration of a document processing device to which the word extraction method of the present invention is applied; FIG. 2 is a diagram showing an example of a text line for explaining a gap histogram; Figure 3 shows an example of a gap histogram, and Figure 4 shows an example of a gap histogram.
The figure shows a contrast between the vertical projection of a line of text and its inter-word gaps. 10...Scanner, 12...Image buffer, 1
4. Line extraction unit, 16. Line buffer, 18. Word extraction unit, 2o... 11 Pass direct projection extraction circuit, 22
. . . Gap histogram creation circuit, 24 . . . Word 1j11 gap judgment dark value determination circuit, 26 . . 111 word extraction circuit. Representative Patent Attorney Makoto Suzuki Figure 2 4 Korororo = Koshi 1 Figure Coro = Colony L2 0 Kororo Coro e-L3 Figure 3 Restriction Figure 4 ↑ ↑↑ ↑

Claims (1)

[Claims] (1) In a document processing device that inputs image information of a document and performs processing including cutting out words from the document, a vertical projection of the image information of a text line is taken, and a histogram of the width of the projection break is calculated. Create a word gap judgment threshold according to the maximum width corresponding to the peak, judge a projection break with a width equal to or greater than the judgment threshold as a gap between words, and cut out words from the image information of the text line. A word extraction method characterized by: