JPS6348219B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to a method for detecting only character lines in a document image. In general, in image input devices such as facsimile and OCR that scan and sample original images pixel by pixel, when the input image is a document image, only the lines where the text is written are read. By skipping other parts, it is possible to efficiently read the input image. As a prerequisite, it is first necessary to detect character lines in a document image, but for this purpose, in the past, the density of each pixel was Rows are detected by projecting values. Figure 1a shows a general document image.If you read it pixel by pixel and integrate the density level (number of black pixels) for each scanning line along the vertical axis, the distribution state will be as shown in the figure. As shown in Fig. b, character lines in the document image are detected. However, in such conventional methods, if the document image is scanned while it is tilted, the density level frequency distribution for each scanning line will be dispersed in the vertical axis direction, and this will be as shown in Figure 1b. As shown, there is a problem in that it no longer corresponds to each character line in the original document image, and as a result, it becomes extremely difficult to detect the character lines. The present invention has been made in consideration of these points, and provides a document image in which character lines can be easily and accurately detected even when the document image is read and scanned in a slightly tilted state. This method provides a character line detection method. Hereinafter, one embodiment of the present invention will be described in detail with reference to the accompanying drawings. Now, as shown in FIG. 2a, if the reading scan is performed in a tilted state of each character line l1, l2, l3, ... of the document image A, in such a state, in the present invention, , First, project each character line l1, l2, l3... from the vertical direction onto the coordinate axis X parallel to the main scanning line (or it may be on the coordinate axis perpendicular to the main scanning line), and determine the projection direction. By calculating the integrated value of the density level for each pixel in , a density level frequency distribution characteristic _fX as shown in the figure is obtained. Next, the density level frequency distribution characteristic f _X is clipped at a certain threshold value T _X to detect the width W of the area where the document image exists. Next, for each band-shaped area of w from both ends of the image area width W,
As shown in Figure b, on the left and right coordinate axes L and R (or on the coordinate axes parallel to the main scanning line), the integrated value f _L , f _R is determined, and by clipping the determined concentration level frequency distribution characteristics f _L , f _R at fixed threshold values T _L , _TR , respectively, the midpoint coordinates L ₁ , L ₂ , L are obtained. ₃ ,…, L _n and R ₁ ,
Detection of R ₂ , R ₃ , ..., _Ro is performed respectively.
At that time, each character line l1, l2,
Since l3, . . . are not necessarily completed at the right end, the numbers of the left and right midpoint coordinates do not necessarily match. Next, in the present invention, the average line spacing ΔL is determined according to the following formula, and an arbitrary midpoint coordinate L _k of 1≦k≦m is determined.
Set R _k ′ ( _k′k ) corresponding to
R _k ′ and

The density levels in the image are integrated along each line segment of [formula] and the result is S _k
and

Find [formula]. ΔL=(L _n −L ₁ )/(m−1) …(1) Here, the integration result S _k of the concentration level along each line segment
and

When [Formula] satisfies the following condition, it is found that the slope of each character line l1, l2, l3, . . . in the document image A is the same as the slope between L _k and R _k '. However, T ₁ is a threshold value set as appropriate. At this time, if the condition of equation (2) is not satisfied, the values of k and k' are sequentially changed and the set is searched until the condition is satisfied. In this way, the midpoint coordinates L ₁ , L ₂ , L ₃ , ..., L _n on the coordinate axis L perpendicular to the main scanning direction
By identifying the area where characters exist in the document image, it is now possible to detect each character line.Then, based on the detection results of each character line in the document image, it is possible to skip between lines in the case of facsimile, and to detect optical characters. In the case of a reading device (OCR), each device is made to perform the desired processing as appropriate, such as cutting out each character along a line. FIG. 3 shows an example of a configuration for concretely implementing the character line detection method for document images according to the present invention described above, and its operation will be explained below. First, image signals sequentially read pixel by pixel by the scanner SCN are sent to the comparator CMP.
There, the pixel information is compared with a preset threshold level and binary quantized, and the quantized white or black pixel information is sent to the page memory PM where it is temporarily stored collectively. Next, the CPU issues a read command in the column direction (direction perpendicular to the rows) to the first memory drive device DMA1 and an adder ADD1.
An addition command is issued to each of the pages, thereby causing the first memory drive device DMA1 to add the page memory.
The adder ADD1 adds up the number of black pixels among the pixel information in the column direction sequentially read out from the PM, and from _the addition result _f The image area width W is determined by . Next, the CPU issues a read command in the row direction (direction perpendicular to the coordinate axis L) in the strip area w at the left end of the image area width W to the first memory drive device DMA1, and an addition command to the adder ADD1. The number of black pixels of each pixel information in the row direction read from the page memory PM is integrated by an adder ADD1 to obtain the density level frequency distribution f _L . At the same time,
The CPU sends a read command to the second memory drive device DMA2 in the row direction (direction perpendicular to the coordinate axis R) in the strip area w at the right end of the image area width W, and an adder.
Addition commands are issued to each ADD2, which causes the concentration level frequency distribution f _R to be calculated in the same way as in the previous case.
is required. Finally, the CPU executes calculations and logical judgments according to the predetermined algorithm based on the integration results f _L and f _R of each of the adders ADD1 and ADD2, and each character line and its slope in the document image are calculated. It will be detected. As described above, in the method for detecting character lines in document images according to the present invention, the density level of each pixel is integrated in the sub-scanning (or main scanning) direction based on image information that has been binary quantized for each pixel. From the density level frequency distribution, determine the band-shaped areas w at both ends of the image area where the distribution fluctuates, and for each band-shaped area w, calculate the density level frequency distribution of each pixel in the main scanning (or sub-scanning) direction. Calculate each existing coordinate value L _i , R _j and line spacing ΔL, and calculate the sum of density levels of each pixel on the line connecting L _i and R _j and the line between the lines parallel to it according to each calculation result. Each character line and its slope is detected by determining S _k and S _k +ΔL/2 and determining whether the deviation value of the sum of density levels on each line is greater than or equal to a certain threshold. This has the excellent advantage of being able to easily and reliably detect character lines even when a document image is read in a tilted state, and being highly versatile.

[Brief explanation of the drawing]

FIG. 1a is a diagram showing an example of a general document image,
FIG. 2B is a diagram showing the density level frequency distribution state for each main scanning line, and FIG. 2A is a diagram showing the density level frequency distribution state in the sub-scanning direction when the document image is read in an inclined state. Figure b shows the density level frequency distribution state and the inclination of character lines in each main scanning direction of the strip area w at both ends of the image area, and Figure 3 shows a concrete implementation of the character line detection method for document images according to the present invention. FIG. A...Document image, l...Character line, W...Image area, w...Strip area at both ends of the image, _fX , _fL , _fR ...
Concentration level frequency distribution, SCN...Scanner, CMP
... Comparator, PM ... Page memory, DMA ...
Memory drive device, ADD...adder.

Claims (1)

[Claims] 1 Scan and sample the image pixel by pixel, 2
A first step of finding an image area from a density level frequency distribution in which density level values of each pixel are integrated in a direction parallel or perpendicular to the main scanning line, based on image information read by value quantization. For a band-shaped area where the distribution state changes on both sides of the image area, the density level frequency distribution of each pixel is calculated by integrating the density level values of each pixel in a direction perpendicular to the integration direction in the first step. The second step calculates the left and right coordinate values and the line spacing, and calculates the sum of the density levels of each pixel on the line connecting the left and right coordinate values and on the line between the lines parallel to it. A method for detecting character lines in a document image comprising a third step in which character lines and their inclinations are detected according to the deviation value of the sum of density levels on each line.