EP1133751A1

EP1133751A1 - Method for determining the position of text lines in text recognition tasks

Info

Publication number: EP1133751A1
Application number: EP99950528A
Authority: EP
Inventors: Walter Aigner; Bertram Zagler
Original assignee: Siemens AG
Current assignee: Siemens AG
Priority date: 1998-09-30
Filing date: 1999-09-16
Publication date: 2001-09-19
Also published as: MXPA01003116A; WO2000019358A1; CA2348633A1

Abstract

The invention relates to a method for determining the position of text lines in text recognition tasks, wherein the distribution of brightness of a detected image segment is determined along a vertical and said distribution of brightness is then filtered, wherein the maximum and minimum values of the function thus obtained are determined, on the basis of which extreme threshold values are calculated that are used for differentiating between text lines and line spacing. The method is particularly advantageous for electronic hand-held read devices.

Description

Procedure for determining the position of lines of text in text recognition tasks

The invention relates to a method for determining the position of lines of text in text recognition tasks.

With the automatic recognition of texts, i.e. When converting the graphic information of a document into text characters that can be further processed using electronic word processing programs, a precise prerequisite for a successful recognition process is the exact determination of the position and size of the individual characters. Knowledge of the position and dimensions of the text lines is a prerequisite for this.

In the case of hand-held reading devices, there is also a non-linear course of the text lines in the captured image section. Here it is necessary to determine the course of a line of text.

The invention is therefore based on the object of specifying a method with which the position and course of lines of text can be determined.

This is done according to the invention with a method of the type mentioned at the outset, in which the brightness distribution of a captured image section along the vertical is determined and this brightness distribution is filtered, the maximum value and minimum value of the function obtained in this way are determined and threshold values are calculated on the basis of these extremes , which serve as the basis for the distinction between text line and cell space. The method according to the invention enables the line position and height to be determined with little effort and thus very quickly, so that this

Information is obtained immediately, which enables a quick recognition process. This is particularly the case with Readers of importance, which have only limited electronic memory available and therefore require a rapid conversion of the memory-intensive image information into text information while the image is being captured.

It is advantageous to design the method in such a way that a cell gap is determined when the function has a combination of a maximum with a minimum, in which the minimum has a value smaller than the functional minimum + number of pixels over the width of the image cutout / 15 + 2 * number of pixels over the width of the image / 15 * maximum function / number of pixels over the width of the image portion and the decrease in the function values after the maximum is a value larger (maximum function - minimum function) / 2 has.

In practice, this design has been preserved through very good results.

Another embodiment of the method is also advantageous in such a way that to determine the left

Edge of a line, the brightness distribution of a captured image section along the horizontal is determined and the function obtained in this way represents the beginning of the line by a sudden increase in the function value. This makes it easy to determine the beginning of the line with little effort. Furthermore, for the determination of the position of the text lines it can be ensured that only images that actually contain text lines and a user error such as e.g. placing the reader pen too far to the left of the beginning of the line does not affect cell detection.

It is advantageous if, after the position of a line has been determined for the first time, the further course of the same is determined by evaluating the information about the recognized text characters. By evaluating the result of the Character classification allows a particularly precise determination of the cell course.

The invention is explained in more detail with reference to figures. The following are examples:

Fig.l a text excerpt as it is typically captured by a hand-held reading device, and the hystogram and

2 shows the filtered hystogram with the entered parameters for assessing the image.

The sequence of the method according to the invention is as follows: A line histogram is determined for the captured image section. The values of all pixels of this line (0 for white and 1 for black) are added up to each line. The result is a function f (y) with

Wιdth- \ f (y) = Y iBlackPixeϊ) _where

; = 0

y The row index of the picture means Width Width (number of columns) of the picture section.

This function is typical when a text is available

Course as it is shown as an example in Fig. 1.

In a further step, filtering takes place in accordance with

with y index in the line histogram

G Weighting according to an exponential smoothing curve i Index of the smoothing curve During the filtering process, the absolute maximum value Va l uemax ie d-_e number of black pixels (black pixels) of the darkest line and the absolute minimum value u uin ie the number of black pixels (black pixels) of the brightest line are determined.

From these two values, parameters for the assessment of the image are derived. These are:

Pit limit = (Val uemax - Val uemin) / 2 but at least the number of pixels across the width of the image section / 30

Mini margin = value min. + Number of pixels across the width of the image / 15, but at most 2 * number of pixels across the width of the image / 15

Minimum threshold = mini margin + (2 * number of pixels over the width of the image / 15 * (value max / number of pixels over the width of the image)) but at most 3 * number of pixels over the width of the Section of image / 15

On the basis of the function f '(y) and the determined threshold values, as are shown by way of example in FIG. 2, the captured image is now assessed with regard to the presence of text lines and line spaces.

For this purpose, the course of the curve is examined to determine whether there are values that are lower than the parameter minimum threshold. If this is the case, the affected area is qualified as a valid minimum and thus as a possible space between cells.

However, there is only an actual cell gap if there is an adjacent maximum with a certain one The presence of a line of text is displayed. These valid maxima are defined by a subsequent decrease in the curve value by an amount> pit limit.

The coincidence of a valid maximum with a valid minimum marks the transition from a text cell to a space between cells. The parameter mini margin is used to precisely determine this transition.

The point at which the curve crosses this threshold between a valid maximum and a valid minimum is defined as the line edge.

A column histogram is used to determine the left edge of a row

Heιght- \ f (x) = ∑ (BlαckPixel)

x Column index of the image section

Height. ... image height

created, in words the color information of the pixels η of the column of the captured image section is added up. The left edge of the text is (if there is at least one line) defined by a sudden increase in the function value f (x).

The tracking of the lines, i.e. The information about the further course of the lines, which is particularly important in the case of hand-held reading devices due to the fluctuations that occur with them, is based on the position of the recognized characters.

For this purpose, the recognized characters are divided into the following large groups: Small characters (e.g. 'a') 0, 7 * character height

Large characters (eg ^Λ A ',' g ') line height Oversized characters (e.g. ^x [ ^λ , 'j') Zeιlenhohe + 0, 3 * Zeιlenhohe

(Descenders)

Special characters: the characters cannot be clearly assigned by the size

A distinction is made between the following character groups to determine the new bottom line of text lines:

Baselme character (eg A ', ^Λ .'): The lower edge of the character corresponds to the lower edge of the text line, regardless of the size of the character;

Short characters (eg ^Λ g ','['): the lower edge of the character corresponds to the lower limit, regardless of the size of the character; Special characters: these characters can be related to their

Do not assign the bottom edge clearly.

Based on these assignments and a probability value G obtained in the course of the classification process regarding the correct classification of the character, the new line height is now determined as follows:

G = probability * CYC MAX WEIGHT

CYC MAX EXTRPAR-1

2 OldHeιght [i] + NewHeight * G Height ι = 0

CYC MAX EXTRPAR + G

G Weighting from the current

Character derived line high probability probability of correct

Character classification (range of values between 0 and 1)

CYC_MAX_WEIGHT Maximum weighting of the new

Character position (e.g. 5)

Height Tracked line height (capital letter height) CYC_MAX_EXTRPAR size of the ring buffer for the

Averaging (e.g.: 3)

OldHeight [] ring buffer

NewHeight Cell height derived from current character (capital letter height)

I index in the ring buffer

The course of the bottom line of the text line is determined according to:

λ

G = probability + - * CYC MAX WEIGHT

CYC_MAX_WEIGHT7

Oldlncrease + Newlncrease * G

Increase = \ + G

Increase * DeltaX + 50 Base = NewBase + ^■

100

G Weighting of the new character position

Probability probability of correct

Character classification CYC_MAX_WEIGHT Maximum weighting of the new

Character position (e.g. 5)

Increase tracked current slope of the

Baselme in%

Oldlncrease Previous slope of the Easeline in% Newlncrease Slope of the Baselme calculated from the position of the current character in%

Base Tracked base position (rounded to integer value)

NewBase Baseline position calculated from the position of the current character

DeltaX X distance in the image of the two

Centers of the most recently extracted characters

The 'Increase' is limited by the plausibility limit CYC MAX ZEILENVERSATZ (in the pocket reader: 15 °).

Claims

claims

1) Method for determining the position of text lines in text recognition tasks, characterized in that the brightness distribution of a captured image section along the vertical is determined and this brightness distribution is filtered, that maximum values and minimum values of the function obtained in this way are determined and threshold values are calculated on the basis of these extremes will serve as the basis for the distinction between text line and line space.

2) Method according to claim 1, characterized in that a line space is determined when the function has a combination of a maximum with a minimum, in which the minimum is a value less than

Functional minimum + number of pixels over the width of the image section / 15 + 2 * number of image points over the width of the image section / 15 * functional maximum / number of pixels over the width of the image section and the drop in the function values after has a value greater than the maximum (functional maximum - functional minimum) / 2.

3) Method according to claim 1 or 2, characterized in that to determine the left edge of a line

Brightness distribution of a captured image section along the horizontal is determined and the function obtained in this way represents the beginning of the line by a sudden increase in the function value.

4) Method according to one of claims 1 to 3, characterized in that after the first determination of the position of a line, the further course of the same is determined by evaluating the information about the recognized text characters.