JPS6325391B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a two-dimensional character area extraction device, particularly an optical character reading device (OCR) that reads characters on a paper surface.
(device), etc., the inclination of the paper surface is corrected even for free-format paper surfaces for which no prior information such as character position, size, etc. This invention relates to a character area extracting device that can automatically extract character areas using.

When reading characters using OCR, it is first necessary to correct the tilt of the input paper and cut out the characters to be read. Conventionally, as methods for correcting the inclination of paper, there have been methods such as printing guide marks on the paper and correcting the paper using the guide marks as a reference, and methods that detect the paper frame and correct the paper using it as a reference. However, in the former case, guide marks must be printed on the paper in advance,
Furthermore, in the latter case, there is a drawback in that it is not possible to correct the skew of a copy that is skewed during cutting or copying. In addition, conventionally, when it comes to cutting out characters, the position and size of the characters on the paper are determined in advance, and each character is cut out based on guide marks or the distance from the edge of the paper. This method had the disadvantage that it was not possible to extract character areas at all for documents created using free formatting.

In order to solve these drawbacks, the present invention corrects the skew even for documents in arbitrary formats that do not have any prior information for skew correction or character extraction, and then corrects the skew of the text areas included in the document. The purpose is to perform extraction using the density value accumulation results, and will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram of the present invention, in which 1 is a quantization circuit that quantizes an input signal, 2 is a two-dimensional image memory that stores the quantized image, 3 is a tilt correction device, and 4 is a region dividing device. , 5 is a field separator detection device, 6 is an area addition/division device, 7 is an area division result storage device, 8 is a non-blank area detection device, 9
1 is a non-blank region rectangularization device, and 10 is a region labeling device. Each device will be explained in detail below.

The quantization circuit 1 scans an original image using a photoelectric conversion device or the like and quantizes the obtained electrical signal into a plurality of levels, and the quantized data is stored as is in the two-dimensional image memory 2.

FIG. 2 explains the principle of tilt correction performed by the tilt correction device 3. a is an example of image data stored in the two-dimensional image memory 2. Here, b ₁ represents a heading character included in the screen, and b ₂ represents one character string of the main text. The two-dimensional image memory 2 is raster-scanned in multiple directions to obtain the cumulative density value, which is the sum of the density values of pixels on each scanning line, and the results are shown in a histogram as shown in Figures 2e and f. be. e is a histogram obtained when raster scanning is performed in the direction c, and f is a histogram obtained when raster scanning is performed in the direction d. When the original screen is quantized into binary values of 0 and 1, the cumulative density value is equal to the number of black pixels on the scanning line. As is clear from FIG. 2, when the direction in which the characters are lined up and the scanning direction match, a peak clearly appears on the histogram (see f), and this property is used to correct the inclination. Specifically, for example, the difference (e′, f′) in density cumulative values in mutually adjacent scanning lines
, and use the sum of the absolute values of the cumulative density value differences, which is the sum of the entire scan or a portion of the scan. For the data in the two-dimensional image memory, the sum of the absolute values of cumulative density value differences is determined for raster scanning from a plurality of directions, and the direction of scanning that takes the largest value among them is determined as the main direction of the screen. That is, θ ₀ such that max {θ} { _o-1 〓 ⁱ⁼¹ |f(l _i+1 )−f(l _i )|} is determined. Here, f(l _i ) is the cumulative density value of the pixel on the i-th scanning line. Correction of the tilt can be realized by adding rotation by −θ ₀ to the two-dimensional image data.

An example of the circuit configuration is shown in FIG. During raster scanning, the density value of a pixel is added to the register 11, and when one raster scan is completed, the difference absolute value circuit 13 is used to add the density value of the pixel to the register 11 and the buffer memory 12, which holds the density cumulative value from one scan before. The absolute value of the difference is taken and added to the contents of the register 14, and at the same time the contents of the register 11 are written to the buffer memory 12, and the contents of the register 11 are cleared. When raster scanning of the entire screen is completed, the contents of the register 14 are stored in the memory 15 and the register 14 is cleared. Reference numeral 16 denotes a tilt correction circuit that looks at the contents of the memory 15 and corrects the tilt of the two-dimensional image data. When the screen is rotated clockwise by θ degrees, the coordinates (x, y) of the tilt-corrected data corresponding to the coordinates (x, y) of the original data are determined as follows: x′=xcosθ−ysinθ y′=xsinθ+ycosθ.

FIG. 4 describes the principle of region division performed using the region division device 4 in FIG. 1.
The area dividing device 4 uses the cumulative density value of the tilt corrected screen obtained by the tilt correcting device 3. It is most effective to use cumulative density values obtained by raster scanning the screen after tilt correction in the horizontal and vertical directions. The density cumulative value reflects the position of the character string existing within the screen. That is, when the density cumulative value is expressed in a histogram, peaks appear on the histogram corresponding to character strings, so the character strings are cut out using the coordinates of the rising and falling positions of the bar-shaped histogram (FIG. 4). g ₁ to g ₆ and h ₁ to _{h 6} are area dividing lines. FIG. 5 is an example of a peak in a density histogram that occurs when a character string exists within a scanned screen. Here, if parts i ₁ and i ₂ can be detected, it is possible to extract one character string. Here, in order to stably detect the i ₁ and i ₂ parts, Fig. 5 is used.
Since it is necessary to remove small irregularities such as those shown in i ₁ to _{i 5} , the following modifications are made to the histogram. When the cumulative density value in the i-th scan is f(l _i ), if f(l _i ) < f( _i+1 ), then f(l _i ) < f(l _i+a ) and f(l _i ) ≥ Find the smallest positive integer a that satisfies f(l _i+a+1 ), and if the value of a is smaller than a certain value, the
The density cumulative value up to the a scan is replaced with the density cumulative value in the i-th scan. This process is then performed for the density cumulative values in all scanning lines. By this process, fine irregularities can be removed without dulling the rising and falling portions of i ₁ and i ₂ in FIG. 5 (FIG. 6). To fill in narrow valleys on the histogram, the histogram may be processed in the opposite way. That is, if f(l _i )>f(l _i+1 ), then f(l _i )>f(l _i+1 ), then f(l _i )<f(l _i+a ′) and f(l _i )≦f(l _i+a ′
₊₁ ), and if the value of a' is smaller than a certain value, the
The density cumulative value up to the a scan is replaced with the density cumulative value in the i-th scan. In order to determine the position of the area dividing line, for example, the sign of the difference between the density cumulative values f(l _i ) and f(l _i+1 ) can be used. In FIG. 6, the range m is indicated by the sign (+). If the difference between the left end cumulative density value f( _ls ) and the right end cumulative density value f( _le ) is greater than or equal to a certain value θ', it is considered that there is a rising portion of the histogram within that range. Furthermore, determine an appropriate value θ″, f(l _s+1 )−f(l _s )＜θ″ 〓 f(l _s+j )−f(l _s+j-1 )＜θ″ f(l _s+ Find a positive integer j that satisfies _j+1 )−f(l _+j )≧θ″, and set the position between l _s+j and l _s+j-1 to draw the area dividing line n ₁ . Similar processing is performed for the falling portion of the histogram. The results obtained by the region dividing device 4 are stored in the region dividing result storage device 7.

The field separator detection device 5 is a device that detects the position, length, etc. of the field separator existing in the two-dimensional image memory 2. Field separators are a general term for shapes used to separate screens, and are usually linear, and most of them are horizontal or vertical to the screen (Fig. 7, Q ₁ , Q ₂ , R). Therefore, it is possible to extract a field separator by detecting a portion where pixels having a certain density value or more are continuously present in the horizontal and vertical directions. If the quantization level of each pixel is 3 or higher, an appropriate threshold value is given, and a location where a certain number or more of pixels with a density value higher than the threshold value continuously exists is detected as a field separator. do. In principle, it is possible to detect a straight field separator using this method, but depending on the printing condition or type of field separator, the field separator may be blurred or may not be an exact straight line. In order to stably detect the field separator, the data is blurred. When detecting field separators in the horizontal direction, it is effective to add blur in the vertical direction. For example, if the density value of pixel (i, j) exceeds the threshold for field separator detection, pixels (i-1, j), (i+
1, j) are also counted. Furthermore, a bridging operation is performed to connect field separators that are within a certain distance.

The area additional division device 6 reads the contents of the area division result storage device 7 and uses the field separator detection device 5.
It is checked whether the field separator detected in is already detected by the area dividing device 4, and if it has not been detected, a new area is divided. Specifically, if there is a region dividing line within a certain distance α from the field separator in the width direction, it is assumed that the field separator has been detected by the device 4. If this is not the case, the contents of the device 7 are modified by adding a dividing line at a distance α from the detected field separator in the smallest area that completely includes the field separator.

In FIG. 7, field separator R is considered to be detected by area dividing lines t ₅ and t ₆ , but field separators Q ₁ and Q ₂ are not detected. Therefore, the area w ₁ that includes the entire Q ₁ ,
At w ₂ and w ₃ , new dividing lines u ₁ and u ₂ are drawn at a distance of α from the field separator. The same applies to _Q2 .

The non-blank area detection device 8 detects the pixels included in each area divided by the area division device 4 and the area addition division device 6 and whose results are stored in the area division result storage device 7. The sum of the density values is determined, and if the ratio of the value to the area is less than a certain value, the area is determined to be a blank area, and if not, the area is determined to be a non-blank area.

The non-blank area rectangularization device 9 converts the connected non-blank areas detected by the non-blank area detection device 8 into one unified area, and converts the non-blank areas detected by the non-blank area detection device 8 into a minimum rectangular rectangle that includes the entire area. A rectangular area is cut out and the result is passed to the area division result storage device 7 (FIG. 8). In Figure 8a, y ₁₁ ~ y ₁₅ ,
y ₂₁ , y ₂₅ , y ₃₁ , y ₃₄ , y ₄₁ , y ₄₃ to y ₄₄ are blank areas,
y ₂₂ to _{y 24} , y ₃₂ to y ₃₃ , and y ₄₂ are non-blank areas. Here, the non-blank areas are combined into one, and the minimum rectangular area containing all the non-blank areas is shown in FIG.
The result is stored in the device 7. The region registered as a non-blank region in the two-dimensional image memory 2 by the device 7 is further divided using the region dividing device 4, and the region that could not be divided in the first division is divided. In Fig. 7, the area x ₁ ,
x ₂ and x ₃ are integrated as a non-blank area and divided by a new dividing line v ₁ . The same applies to the dividing line _v2 . Repeat this operation until no more dividing lines can be drawn.

The region labeling device 10 outputs the coordinates, size, and category of each non-blank region stored in the region division result storage device 7 in the form of a table.

As explained above, according to the present invention, when inputting a document using OCR etc., even if there is no skew correction mark printed in the document, or when copying, etc., the original is skewed and copied. Even if the object to be read is a cutout, or a free-format object in which the position and size of characters are not determined in advance, density values are accumulated during raster scanning. In this way, character areas can be cut out relatively easily. Although the present invention is largely effective for printed matter, it is also effective for handwritten documents, and is also effective for extracting character areas from sheets of paper that include tables, graphs, etc. other than characters.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an embodiment of the device of the present invention, FIG. 2 is a principle explanatory diagram illustrating the method of angle correction, and FIG. 3 is a circuit diagram of an embodiment of the tilt correction device shown in FIG. 1. Figure 4 is an explanatory diagram showing the principle of area division.
FIG. 5 is an example of a density cumulative value histogram, FIG. 6 is an explanatory diagram illustrating a method for determining the position of area dividing lines by modifying the histogram, and FIG. 7 is related to area division and character area extraction. Explanatory diagram, No. 8
The figure shows an explanatory diagram for explaining the operation of the non-blank area rectangularization device shown in FIG. In the figure, 1 is a quantization circuit, 2 is an image memory, 3 is a tilt correction device, 4 is an area division device, 5 is a field separator detection device, 6 is an area addition division device, 7 is an area division result storage device, 8 9 is a non-blank area detection device, 9 is a non-blank area rectangularization device, and 10 is a non-blank area detection device.
represents a region labeling device.

Claims (1)

[Claims] 1. A quantization circuit that decomposes a two-dimensional screen into pixels and quantizes an electrical signal output according to the density of each pixel, and a two-dimensional image memory that stores the quantized data. In preparation, in a two-dimensional character area extraction device that performs raster scanning on a two-dimensional image memory to extract a character area, raster scanning is performed on the two-dimensional image memory from multiple directions, and pixels on a scanning line are extracted. a tilt correction device that detects the tilt of an input screen by accumulating quantized density values and determining the scanning direction in which the density value accumulation results change most rapidly, and corrects the tilt based on the detection result; a device that performs region segmentation by detecting a portion where the density value cumulative result obtained by raster scanning the two-dimensional image memory resulting from the correction from multiple directions changes rapidly; a device for storing results obtained using a region dividing device; a field separator detecting device for detecting a field separator from a continuous amount of pixels having a predetermined value or more in the two-dimensional image memory; A two-dimensional character area, characterized in that it is equipped with an area additional dividing device that corrects the already obtained area dividing result based on the field separator detected by the device, and extracts a character area. Extraction device.