JPH0490082A - Device for detecting character direction in document - Google Patents

Abstract

PURPOSE:To make it possible to accurately detect the character direction of a document even in the case of using an original mixing halftone or graphic or an original of complex column composition by finding out a unit area (circumscribed rectangle), measuring the length (run length) of a space area (color run) in the unit area and deciding the character direction by the distribution of appearance frequency. CONSTITUTION:A picture element area forming means 1 detects a picture element group consisting of mutually relating picture elements in an input image, e.g. a picture element group constituting one character and a picture element constituting one table or a graphic, as a unit area. A run length frequency distribution feature extracting means 3 finds out the distribution of run length appearance frequency in each of vertical and horizontal directions from run length found out by a run length detecting means 2 and finds out its feature, e.g. the peak value of frequency. A direction deciding means 4 compares the features (peak values) of vertical and horizontal appearance frequency found out by the means 3 with each other and decides the vertical or horizontal writing of an inputted document. Thereby, the character direction of a document can be accurately detected even in the case of using an original mixing halftone or graphic or an original of complex column composition.

Description

[Detailed description of the invention]

In other words, for simple columns containing only characters, the direction of the characters could be determined based on the peripheral distribution. However, manuscripts containing halftones, figures and tables as shown in Figure 13(b), or the same Z (c
) It was difficult to determine the character direction based on the peripheral distribution in manuscripts with complicated columns, such as the one shown in (). The present invention aims to solve these problems of the prior art. That is, the present invention provides a document character direction detection device that can accurately detect the character direction of a document, not only for a simple document containing only text, but also for a document containing halftones, figures and tables, and a complex document with columns. The purpose is to Means for solving rI1 layer] In order to achieve the above object, the present invention, as shown in FIG. run-length detection means 2 for determining run lengths in the vertical and horizontal directions of blank areas between adjacent unit areas; and run-length frequency distribution feature extraction for extracting features of appearance frequency distribution of run lengths in the vertical and horizontal directions; The present invention includes a means 3 and a direction determining means 4 for determining the direction of characters in a document by comparing the characteristics of the appearance frequency distribution of the extracted vertical and horizontal run lengths.

[Effect]

Figure 9 is an example of a document image, where the document includes characters, halftones,
It is composed of figures, etc. If we focus on the characters here,
Characters form lines, and in the case of horizontal writing, vertical characters (lines)
The horizontal character spacing is narrower than the spacing. Conversely, in the case of vertical writing, the vertical character spacing is narrower than the horizontal character (line) spacing. The present invention focuses on this characteristic and detects the direction of characters in a document based on the measurement of character spacing. The pixel region forming means I
A related pixel group, for example, a pixel group constituting one character, a pixel group constituting one table or figure, is detected as a unit area. As a method for extracting a unit area from a pixel group, there is, for example, a circumscribing rectangle process in which a circumscribed rectangle of a pixel group is obtained and this is used as a unit area. FIG. 10 shows an example of the result of converting a pixel group into a unit area (circumscribing rectangle). Next, the run length detecting means 2 determines the vertical and horizontal run lengths of blank areas between adjacent unit areas. The run length frequency distribution feature extracting means 3 obtains the distribution of appearance frequency of the run lengths in each of the vertical and horizontal directions from the run lengths obtained by the run length detecting means 2, and obtains the characteristic, for example, the peak value of the frequency. The direction determining means 4 compares the vertical and horizontal appearance frequency features (peak values) obtained by the run length frequency distribution feature extracting means 3, and determines whether the input document is written vertically or horizontally. Figures 10 and 11 show examples of the frequency distribution for horizontal and vertical writing, and as is clear from these figures, there is a peak in the frequency of run lengths of white runs in the horizontal and vertical directions. The magnitude relationship of the values is
The situation is reversed for horizontally written documents and vertically written documents, and therefore, the direction determining means 4 can make a determination by examining the above-mentioned magnitude relationship. According to the present invention, it is possible to accurately detect the character direction of a document, not only for a simple document containing only text, but also for a document containing halftones, figures and tables, and a complex document with columns.

【Example】

Hereinafter, the present invention will be explained in detail with reference to embodiments shown in the drawings. FIG. 2 shows the overall configuration of a document image processing apparatus according to an embodiment of the present invention. This document image processing device includes an image scanner 21 that scans and inputs a document image as a digital image. An image memory 22 that stores input images for processing, a character direction extraction unit 23 that detects the character direction of the document of the present invention, a CPU 24 that controls the entire device, a data bus 25, a monitor 26 that displays images and messages, and a keyboard 27. , an external storage device 28 for storing images, etc., and a printer 29 for outputting images. FIG. 3 is a diagram showing the configuration of the character direction detection section 23 of the present invention in FIG. 2. The character direction detection section 23 includes a circumscribed rectangle processing section 30, a white run detection processing section 31, a frequency distribution detection processing section 32, a peak detection processing section 33, and a character direction determination section 34. The circumscribed rectangle processing unit 30 is an example of the black pixel region forming means 1 shown in FIG. A rectangular area (
unit area). Any conventional method can be used for this rectangularization process, and one example of a conventional rectangularization process is a method in which a group of black pixels is surrounded by a rectangle by tracing the outline of an image. In this method, the minimum X, Y coordinates and maximum X, Y coordinates of the element containing the black pixel group can be determined by tracing the connected components of black pixels in the black area. This allows a group of black pixels to be surrounded by a rectangle. The present inventor has proposed a method that improves this conventional method, and the present applicant has filed an application (Japanese Patent Application No. 1-870).
No. 39). Preferably, this improved technique is utilized in the rectangularization process of the present invention. An example of the rectangularization process of the above application will be briefly explained. Here, as an example, an image called "side" which is a character will be explained. FIG. 4 shows the word "side" in pixels. Therefore, black pixels are connected from the upper left to the lower right of this image using the mask pattern shown in FIG. 5(a). If the pixel above and the pixel to the left of the pixel of interest are both black pixels, processing is performed to convert the pixel of interest into a black pixel. An example of this masking process is shown in FIG. 5(b). An image in which black pixels are connected in the right-hand direction is obtained. Next, black pixel connection processing is performed from the upper right to the lower left using the mask pattern shown in FIG. 6(a). The obtained results are shown in FIG. 6(b). An image in which black pixels are connected in the left direction is obtained in the opposite way to the previous process. Next, black pixel connection processing is performed from the lower left to the upper right using the mask pattern shown in FIG. 7(a). The obtained results are shown in FIG. 7(b). It can be seen that the entire character is almost surrounded by a rectangular area. Finally, black pixels are connected from the lower right to the upper left using the mask pattern shown in FIG. 8(a). The obtained results are shown in FIG. 8(b). By performing a series of four processes in this way, the character area can be surrounded by a rectangle. Although the processing has been explained using text as an example, the results also apply to figures, tables, and halftones. FIG. 9 shows an example of a horizontally written document image, and FIG. 10 shows an image after circumscribing rectangle processing of the image in FIG. 9. The white run detection processing section 31 corresponds to the run length detection means 2 in FIG. 1, and scans the image obtained as described above in the horizontal and vertical directions to detect white runs. This is a processing section that calculates the length. The process for determining the run length can be arbitrarily selected from conventional methods. The frequency distribution detection processing section 32 determines the distribution of the frequency of appearance of white runs based on the run lengths of the white runs obtained by the white run detection processing section 31. FIG. 11 is a graph showing the frequency distribution of runs in the case of a horizontally written document.
) indicates the run length frequency of vertical white runs. In addition, Fig. 12 is a graph showing the frequency distribution of runs in the case of a vertically written document. Indicates the run length frequency of white runs. As can be seen from these figures, the distribution of the run length of white runs in the vertical and horizontal directions is different for both vertically written documents and horizontally written documents, and the run length of white runs is different between vertically written documents and horizontally written documents. The distribution is reversed. Based on the difference in this distribution, it is possible to determine whether the text is written vertically or horizontally. In this embodiment, the characteristics of each distribution are captured by the run length of the white run having the peak value of the appearance frequency. The peak detection processing unit 33 calculates the run length of a white run having a peak frequency value by comparing the frequencies. The above frequency distribution detection processing section 32 and peak detection processing section 3
The processing unit 3 corresponds to the run length frequency distribution feature extraction means 3 in FIG. The character direction determining unit 34 corresponds to the direction determining unit 4 in FIG.
When (horizontal run length - vertical run length), horizontal writing is determined, and when (horizontal run length > vertical run length), vertical writing is determined. This example is based on white runs between characters, so
It is not affected by columns. Even in a document with complicated columns as shown in FIG. 13(C), the orientation of which cannot be determined using conventional peripheral distribution detection, can be easily determined. Furthermore, since halftones and figures and tables are wider than the normal character spacing, their appearance frequency is located outside the peak value in the appearance frequency distribution, so they do not affect the determination of vertical writing/horizontal writing. Therefore, the character direction of the document can be detected with high accuracy. In the embodiment, white runs between rectangles are calculated for all circumscribed rectangles, but rectangles that are large to a certain extent may be excluded from processing targets as a preprocess for calculating white runs. By performing this preprocessing, it is possible to avoid the influence of halftones and graphics on character direction determination, and only data that is effective for line direction detection can be collected, thereby improving measurement accuracy.

【Effect of the invention】

According to the present invention, a unit area (circumscribed rectangle) is obtained, and the length (run length) of a blank area (white run) between the unit areas is determined.
Since it measures the character direction and determines the character direction based on the appearance frequency distribution, it can accurately detect the character direction of documents, not only for simple documents with only text, but also for documents with mixed halftones, figures and tables, and complex documents with columns. can do.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the basic configuration of the present invention. FIG. 2 is a diagram showing the configuration of a document image processing apparatus according to an embodiment of the present invention. FIG. 3 is a diagram showing the configuration of the character direction detection section in FIG. 2. FIG. 4 is a diagram showing an example of a pixel pattern of a character. FIG. 5(a) is a diagram showing the mask and scanning direction when masking is performed from the upper left to the lower right, and FIG. 5(b) is a diagram showing the result of the masking process. FIG. 6(a) is a diagram showing the mask and scanning direction when masking is performed from the upper right to the lower left, and FIG. 6(b) is a diagram showing the result of the masking process. FIG. 7(a) is a diagram showing the mask and scanning direction when masking is performed from the lower left to the upper right, and FIG. 7(b) is a diagram showing the result of the masking process. FIG. 8(a) is a diagram showing the mask and scanning direction when masking is performed from the lower right to the upper left, and FIG. 8(b) is a diagram showing the result of the masking process. FIG. 9 is a diagram showing an example of a document image. FIG. 10 is a diagram showing an image after circumscribed rectangle processing. FIG. 11 is a diagram showing the frequency distribution of runs in the case of a horizontally written document, where (a) is a diagram showing the frequency distribution of horizontal white run length, and (b) is a diagram showing the frequency distribution of vertical white run length. FIG. FIG. 12 is a diagram showing the frequency distribution of runs in the case of a horizontally written document, where (a) is a diagram showing the frequency distribution of horizontal white run length, and (b) is a diagram showing the frequency distribution of vertical white run length. FIG. FIG. 13 is a diagram for explaining the method of detecting document character direction using conventional peripheral distribution detection, in which (a) is a case of a simple column of characters only, and (b) is a case of a half-tone image. (c) shows the case of a manuscript with complicated columns. (2)...Pixel region forming means, 2...Run length detection means, 3...Frequency distribution feature extraction means, 4...Direction determining means. Patent applicant Fuji Xerox Co., Ltd. Agent Patent attorney Iwatakayoyo Attorney Patent attorney 1) Hideyo Nakataka Attorney Patent attorney Yoji Onodera Figure 2 Figure 1 Figure 4 (b) Figure 6 Figure 7 (b ) Fig. 8 (b) Fig. 9 No. 10 casting (b) Fig. 13 (a) <c>

Claims (1)

[Scope of Claims] A black pixel region forming means for obtaining mutually related black pixel groups as unit regions in a binary image, and a run length for obtaining the vertical and horizontal run lengths of blank areas between adjacent unit regions. A detection means, a run length frequency distribution feature extraction means for extracting features of the appearance frequency distribution of the run lengths in the vertical and horizontal directions, and a feature of the extracted appearance frequency distribution of the run lengths in the vertical and horizontal directions are compared. 1. A document character direction detection device comprising: direction determination means for determining the direction of characters in a document.