JPH01231186A - Character recognizing system - Google Patents

Abstract

PURPOSE:To execute a classification of a character which cannot be discriminated by only a shape, a detection of a line pitch, and a comparison, an integration and a classification based on a line pitch and a center line position by extracting a center line from one character line, evaluating a height position of a character, based thereon and classifying the character. CONSTITUTION:A document image which has been inputted from an input means 1 passes through a character line extracting part 2 and a character segmenting part 3, and a series of character circumscribed rectangles and a character pattern are extracted. A center line extracting part 4 extracts a center line from this series of character circumscribed rectangles, and a classifying part 5 uses the obtained center line and classifies each character, based on a position of its circumscribed rectangle. A pattern recognizing part 6 coordinates the character pattern and a pattern collation dictionary 7 and outputs a result of collation. An interpreting part 8 evaluates both a result of classification and a result of collation and outputs a result of recognition, and an output means 9 outputs a result of recognition.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Field of Industrial Application) The present invention relates to a character recognition method for cutting out a character string from a document image and recognizing each character.

(Prior Art) In the recognition of European characters, for example, the distinction between a comma and an apostrophe, or the distinction between the meanings of "2" in X2 and X2, may not be determined based only on the shape of the characters.

The distinction between these characters can be easily determined by paying attention to the differences in the positions in which the characters appear, but it is necessary to provide a positional standard for evaluation. Furthermore, in order to deal with a variety of documents, it is desirable to extract these standards from the documents and use them.

Therefore, the problem was how to extract criteria from documents.

Conventionally, as a method of classifying characters in a document by focusing on their positions, for example, as in JP-A-62-187988, four purlines are set in a character line, and the characters are classified while correcting the purlines. There were ways to do this, but they were not sufficient.

Additionally, in document recognition, being able to compare, integrate, and classify related character lines is useful for understanding the structure of a document. To do this, it was necessary to extract parameters for comparison from the character lines.

(Problems to be Solved by the Invention) As described above, the conventional techniques include character distinction, which cannot be determined by shape alone, and problem 1 of extracting criteria for distinction from a document. I had a problem extracting from a row.

An object of the present invention is to extract the center line from a line of characters, evaluate the height position of the characters based on this, and classify the characters, thereby solving all of the above problems.

[Structure of the invention]

(Means for Solving the Problems) In order to achieve the above object, the present invention calculates the center line using a function obtained by calculating in the row direction a numerical value or function defined by the position and range of the character pattern. A character recognition method is provided which has a function of extracting each character in the character line and a function of classifying each character in the character line based on the amount of deviation in the horizontal and vertical directions with respect to the obtained center line.

(Function) According to the present invention, it is possible to stably extract the center line for every 5 character lines,
Characters can be classified using the extracted center line. According to this classification, the difference in the position of characters becomes clear, so it is possible to determine the distinction between characters, which is difficult to do by evaluating only the shape. Furthermore, by evaluating the position and interval of the extracted center lines, it becomes possible to extract the pitch of one line and to compare, integrate, and classify the lines.

(Example) An example of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing the overall configuration of a character recognition device according to the present invention. A document image inputted from the input means 1 passes through a character line extracting section 2 and a character cutting section 3, where a series of character circumscribing rectangles and character patterns are extracted. The center line extraction unit 4 extracts a center line from this series of character circumscribed rectangles. The classification unit 5 uses the obtained center line to classify each character based on the position of its circumscribed rectangle. The pattern recognition unit 6 associates the character pattern with the pattern matching dictionary 7 and outputs the matching result. The interpreter 8 evaluates both the classification results and the matching results and outputs the recognition results. The output means 9 outputs the recognition result.

Next, the center line extraction section 4 will be explained in detail. FIG. 2 is a flowchart of one processing system of the center line extraction section 4 in FIG. The function of each block in FIG. 2 will be explained below.6 Function definition 10 defines a function corresponding to the position of each character in a character line and the size of the character in the direction perpendicular to the line. In this embodiment, this function is a triangular function as shown in FIG. This triangle is an isosceles triangle whose base is the side on the X axis (direction perpendicular to the character line) shown in FIG. 3, and whose height is a fixed value H.

In histogram creation step 11, a histogram is created by adding the function defined in step 10 in the row direction over one character line.

In peak detection 12, the highest peak is detected from the obtained histogram. This peak gives the location of the centerline.

Next, the function of the classification section 5 shown in FIG. 1 will be explained. In this embodiment, characters are classified into categories ω to ■ shown in FIGS. 4(a) and 4(b). The classification section shown in FIG. 1 classifies characters based on the amount of deviation from the center line in the direction perpendicular to the line. Figure 5 is a diagram showing the definition of the above deviation amount. Find the difference between Ll and L2 in Figure 1, take this as L, and classify the character according to the magnitude of the absolute value of L as a character with a small value or a medium character.・Large letter 3
Sort by street. These three classification divisions correspond to the classification of characters into three groups, [■], [■■], and [ni)■] using the category number 1 in FIG. At this time, the characters belonging to the second and third groups are further classified according to the sign of the above values. In other words, L=L1
-L2, if L is positive, the character is determined to be ■ or (to) in the category 1 number in Figure 4, and if L is negative, the character is determined to be ■ or (to). It is determined that ■.

FIG. 6 shows the conditions for the above classification. The numbers in parentheses in FIG. 0 are the category numbers in FIG.

Also thl, th2. th3 is a threshold value. Note that the interpreter 8 in FIG. 1 selects, as the recognition result, the character that matches the classification result and has the maximum likelihood among the candidate characters with associated likelihoods obtained by pattern matching.

Further, as a modification of this embodiment, the following example will be given.

In the above embodiment, character classification may be performed prior to pattern matching to limit the category to which the character belongs, and during pattern matching, only a pattern matching dictionary consisting of only characters belonging to the category may be referred to. good.

In the above embodiment, not only the histogram obtained by adding isosceles triangles defined by the character position and the existing area in the row direction, but also the histogram of the center position of the character circumscribing rectangle, for example, may be used.

Further, processing such as blurring the histogram may be performed.

In the above embodiment, the center line may be extracted after removing small characters or correcting skew in advance.

In the classification section shown in FIG. 1 of the above embodiment, a plurality of categories may be output by adding a likelihood, etc., without uniquely determining the category of characters. In this case, the interpreter evaluates the likelihood of all the output categories and the output pattern matching results to determine the final result.

In the above embodiment, processing is performed by having knowledge about the categories into which a certain character in a certain font can be classified, and comparing this knowledge with the classification results of characters that belong to multiple classification categories due to differences in fonts, even if the same character is the same character. It is possible to estimate the range of types of character fonts in the text, and to adapt subsequent processing to the range of types of character fonts.

In the embodiments described above, similar processing may be performed using not only one line of information but also multiple lines of information.

Comparison, integration, and classification of one line with line pitch detection based on the center line extracted by the above embodiment can also be performed. That is, the distance between the center lines of adjacent rows in the direction perpendicular to the row direction is determined to obtain the distance between adjacent rows (row pitch). Lines with similar line pitches can be combined as the same article. Furthermore, rows that are adjacent in the row direction and whose center lines are at the same position can be combined into the same row.

In short, the present invention can be modified and used in various ways without departing from the gist thereof.

〔Effect of the invention〕

With this invention, the classification of characters that cannot be distinguished by shape alone,
It is possible to detect line pitch and compare, integrate, and classify lines based on line pitch and centerline position.

[BRIEF DESCRIPTION OF THE DRAWINGS] FIG. 1 is a block diagram of a character recognition device according to the present invention, FIG. 2 is a flowchart of center line extraction processing in an embodiment of the character recognition device shown in FIG. 1, and FIG. FIG. 4 is a diagram showing a function defined by the existence position and existence area of a character circumscribing rectangle in one embodiment of the character recognition device shown in FIG. 1, and FIG. 4 is a character classification in one embodiment of the character recognition device shown in FIG. 1. 5 is a diagram showing the definition of the amount of deviation in one embodiment of the character recognition device shown in FIG. 1, and FIG. 6 is a diagram showing the classification conditions in one embodiment of the character recognition device shown in FIG. 1. FIG. 1... Input means, 2... Character line extraction unit,
3... Character cutting section, 4... Center line extraction section,
5. 1 classification section, 6... pattern recognition section. 7... Dictionary for pattern matching, 8... Interpretation unit, 9... Output means. Agent Patent Attorney Noriyuki Chika Yudo Mitsuyuki Matsuyama Figure 1 Figure 2 Figure 3 Iku! [Riyo) a/+ that (〃Degory (G number (α)) (F) (21 (3) (41 (5) Category increase Y number Cb) Figure 4

Claims (2)

[Claims] (1) Cut out a character line from an image containing a character string, define a numerical value or function based on the location or area of each character in the character line, and calculate this defined numerical value or function for each character in the row direction. A character recognition method characterized in that the center line of the character line is extracted using a function obtained by (2) Characters according to claim 1, wherein the extracted center line is used to classify the characters according to the position of each character in the character line and the amount of deviation in the line perpendicular direction from the center line. Recognition method.