JPH0624019B2 - Circumscribing rectangular pattern arithmetic circuit - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an arithmetic circuit for calculating a circumscribing rectangular pattern of each character pattern by sequentially performing directional expansion processing on a document image. This is performed as a pre-process for cutting out only character patterns from a general document, and one of the arcuate patterns whose circumscribed frames overlap is 1
It can be used when you want to extract as one pattern.

[Conventional technology]

As a method of separating a character pattern from a document image, a method of obtaining a circumscribed rectangle of each pattern by connectivity analysis is known. The connectivity analysis takes time when the pattern is complicated, but even if the circumscribed rectangles have overlapping regions, they can be separated as separate patterns. However, in the case of a character pattern, it may be desired to handle it as one pattern even if it is composed of a plurality of patterns, which is inconvenient in such a case.

FIG. 8 shows an example of a document image. Here, 11 is a document image, 12 is a character pattern, and 13 is a circumscribed rectangle. That is, when trying to obtain the circumscribed rectangle 13 as shown in the figure by the connectivity analysis, "character", "cut", "outside",
There are two character patterns of "contact" and "shape" respectively, 2
Since each pattern is composed of two, three, four patterns, it is necessary to perform appropriate post-processing to integrate the patterns.

On the other hand, as a method other than the connectivity analysis, a clipping method using a projection method is well known. In this case, the order in which the vertical and horizontal projections are taken differs between vertically written and horizontally written documents. Then, after the line is cut out, projection is performed to cut out the character for each line.

[Problems to be solved by the invention]

However, in the case of the projection method, there is a problem in that it is not possible to cut out a character from a tilted document, which is good when the tilt of the document is small as shown in FIG. Further, it is difficult to cut out a document when photographs and figures are mixed.

Therefore, an object of the present invention is to make it possible to obtain a circumscribing rectangular pattern of a character pattern at high speed even in an inclined document, a document in which photographs and figures are mixed, and the like.

[Means for solving problems]

A storage unit that captures a document image and stores the document image data obtained by binarizing each pixel, and a reading unit that reads the document image data by executing horizontal scanning from four directions that are a combination of upper and lower sides and left and right sides. In addition, a predetermined window of 2 × 2 pixels is formed for each pixel of interest in each scanning direction to be read out, and when the other pixels except the pixel of interest in the window are black pixels sequentially in accordance with the scanning direction, Anisotropic expansion processing of document image data by sequentially performing the means for determining the value as a black pixel and the processing for determining the value of the pixel of interest by using the determined image value at the next window position for each scanning direction. And calculating means for calculating a circumscribing rectangular pattern.

[Action]

By using four types of windows (2 x 2 pixel size) and sequentially expanding the document image in four directions, the circumscribed rectangular pattern can be calculated at high speed simply by reading and writing four times in the memory. Is what you are trying to do. Furthermore, the position and size of the circumscribed rectangle can be obtained by applying the corner point filter to the image memory that has undergone the expansion process and detecting the corner point coordinates of the four corners.

〔Example〕

FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, 1 is an address generation circuit, 2 is a document image memory,
3A and 3B are multiplexers, 4A and 4B are line memories, 5A and 5B are latch circuits, 6 is an AND gate, 7 is an OR gate, and 8 is a buffer.

In this case, in short, the circumscribed rectangular pattern 13 is calculated by subjecting the character pattern 12 as shown in FIG. 2 to anisotropic expansion processing. Here, the anisotropic expansion processing means that the conditions C _{1 to} C ₄ shown in FIG. 3, that is, the black pixels BL shown by hatching are shown in FIGS.
Only when there are four types of arrangements such as (C) and (D), it is defined as a process of converting the pixel of interest CP into a black pixel. At this time, the pixels DC other than the black pixels BL may be white pixels or black pixels. When the 3 × 3 condition shown in the figure is executed in parallel for the document image, the number of times of convergence differs depending on the character pattern. Therefore, here, these four conditions C _{1 to} C
₄ , the image memory 2 is read out in the fourth order.
It is assumed that the _four scanning modes S _{1 to} S ₄ shown in the figure are performed, and the new value of the pixel of interest is used in the expansion processing of the adjacent pixel to perform the sequential pixel calculation. The calculation processing of the circumscribing rectangular pattern is completed by only the four types of scanning shown in FIG.
As described above, the condition C ₁ is the scan S ₁ and the condition C ₂ is the scan S 1.
_The condition C ₃ corresponds to the scan S ₃ , and the condition C ₄ corresponds to the scan S ₄ , so that the condition can be determined in the 2 × 2 window including the pixel of interest.

A more specific description will be given with reference to FIG.

The address generation circuit 1 generates horizontal (X) and vertical direction addresses A _X and A _Y for addressing the document image memory 2 that stores, for example, binarized document image data. As a result, the image data P of the addresses A _X and A _Y
D is read out, and the multiplexer 3A selects the line memory 4A or 4B corresponding to the A _Y address. For example, when the address A _Y is an odd number, the image data PD is written in the line memory 4A, and when the address A _Y is an even number, the image data PD is written in the line memory 4B. Then, reading the contents of the line memory 4A or 4B corresponding to the previous address of the address A _Y, is selected by the multiplexer 3B, and outputs the image data f _C of said line. At this time, the value f _D of the immediately preceding pixel is latched in the latch circuit 5A. Here, the line memories 4A and 4B are the number of horizontal pixels H of the document image.
It is a memory of the same size as _max . Further, the image data f _A stores the previous value as the value f _B in the latch circuit 5B, and the values (f _A , f _B , f _C , and
f _D ) forms the 2 × 2 window shown in FIG. The figure shows the arrangement for the C ₁ condition. And the values f _B , f
The logical product of _C and f _D is obtained by the AND element 6, and the logical sum with the target image f _A is obtained by the OR element 7. By this, C
_{When 1} condition is satisfied, f _A becomes the value “1”, and the OR element 7
Is latched by the latch circuit 5B as the output value f _A ′ of Also, this new value f _A ′ is stored in the image memory 2 through the buffer 8. The anisotropic expansion process is executed by repeating the above process for all screens. When the anisotropic expansion processing is executed in the four kinds of directions shown in FIGS. 4 (a), (b), (c), and (d), the circumscribed rectangular pattern 13 of FIG. 2 remains in the memory 2. .

A sixth processing example in which the character image 11 shown in FIG. 2 is expanded by the circuit shown in FIG. 1 is shown. That is, as shown in FIG. 9A, when the scan S ₁ is expanded using the condition C ₁ , the portion of the black pixel F _{1 in} the memory 2 is changed from the white pixel to the black pixel. Furthermore, the scanning _S 2 as shown in FIG (B),
When expanded under the condition C ₂ , the area of the black pixel F _{2 in} the memory 2 is expanded. Further, when the scan S ₃ and the condition C ₃ are used for expansion as shown in FIG. 7C, the area F _{3 of the} memory 2 becomes a black pixel. Finally, as shown in FIG. 4D, when the scan S ₄ and the condition C ₄ are expanded, the area F ₄ in the memory 2 becomes a black pixel, and the circumscribed rectangular pattern 13 is obtained.

For the circumscribed rectangular pattern 13 thus obtained, for example, by using the corner point filters (templates) TL _{1 to} TL ₄ as shown in FIG. 7, one corner point of the rectangular pattern, for example, FIG. A _{1 to} A ₄ can be easily obtained.

As described above, according to the present embodiment, the circumscribing rectangle can be obtained at high speed by scanning the entire surface of each window using four types of 2 × 2 windows. In other words, even if the document is tilted or mixed with figures and photographs, if the rectangle circumscribing the characters stands up, there is an advantage that the rectangle can be obtained accurately and at high speed. Moreover, the position and size of the rectangle can be easily obtained by using the corner point filter.

〔The invention's effect〕

According to the present invention, the judgment conditions of the expansion process are limited to four types, and the anisotropic expansion process is realized by changing the scanning of the image memory according to the conditions, and the circumscribing rectangle is circumscribed by four times of scanning. The pattern can be calculated at high speed. In addition, since the position and size of the rectangle can be calculated by using the corner point filter from the circumscribed rectangle pattern, it is possible to calculate the circumscribed rectangle at high speed even in a document with a slanted document or a mixture of figures. Become.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of the present invention, FIG. 2 is an explanatory diagram for explaining an example of a character pattern and a circumscribed rectangle, and FIG. 3 is an adjacent pixel condition for making a target pixel a black pixel. FIG. 4 is an explanatory diagram for explaining, FIG. 4 is an explanatory diagram for explaining the scanning direction of the document image corresponding to each condition of FIG. 3, and FIG. 5 is 2 in the case of a certain adjacent pixel condition (C ₁ ). FIG. 6 is an explanatory diagram for explaining a × 2 window, FIG. 6 is an explanatory diagram for explaining an execution process of anisotropic expansion processing according to the present invention, and FIG. 7 is an explanatory diagram for explaining a corner point filter (template). 8 and 9 are explanatory diagrams for explaining an example of a document image. Explanation of code 1 ... Address generation circuit, 2 ... Document image memory, 3
A, 3B ... Multiplexer, 4A, 4B ... Line memory, 5A, 5B ... Latch circuit, 6 ... AND gate, 7 ... OR gate, 8 ... Buffer, 11 ... Document image, 12 ... Character pattern , 13 ... circumscribed rectangle.

Claims (1)

[Claims] 1. A document image obtained by capturing a document image and storing document image data obtained by binarizing each pixel, and performing horizontal scanning from four directions, which are a combination of vertical and horizontal directions. Read-out means for reading out data and 2 × 2 for each pixel of interest for each scanning direction for reading out
A method for forming a predetermined window of pixels and sequentially determining the value of the pixel of interest to be a black pixel when the other pixels except the pixel of interest in the window are black pixels sequentially according to the scanning direction, and the determined image value Arithmetic processing means for realizing the anisotropic expansion processing of the document image data by sequentially performing the processing for determining the value of the pixel of interest using the next window position for each scanning direction, and calculating the circumscribed rectangular pattern. A circumscribing rectangular pattern operation circuit characterized by the following.