JPS61296481A - Document reader - Google Patents

Abstract

PURPOSE:To improve the accuracy of attribute decision and to process graphic areas or list areas other than characters by detecting an image area on the basis of a compressed and stored image and comparing the detected image area with a feature previously prepared in each attribute of a character string, a list, a graphic, or the like to decide the attribute. CONSTITUTION:A document image stored in an image information storing part 1 is divided into small areas and a compressed image setting up '1' as the value of picture elements when the number of picture elements '1' is >=5 and setting up '0' when the number of picture element <5 is stored in a compressed image storing part 2. A feature image setting up the value of picture elements to '1' when picture element '1' are connected in the eight directions from the left end to the right end of each small area and setting up other picture elements to '0' is stored in a picture element feature storing part 3. A compressed image processing part 4 scans the compressed image in the character string direction, and when the continuousness of picture elements '0' is less than a prescribed value, forms a conversion image obtained by rewriting the value of the picture elements into '1' and an area detecting part 5 detects an area where picture elements '1' are connected in four directions. An attribute deciding part 16 compares the detected result with the feature previously prepared in each attribute such as a character string, a list, a graphic or the like to decide the attribute. An area processing part 17 separates a character pattern and outputs the pattern to recognize it.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a document reading device that reads characters from a document containing a mixture of characters and non-text information such as diagrams, photographs, etc. 1 is a diagram showing a configuration of a conventional document reading device that cuts out and reads only turns of a document reading device that can perform processing according to the attributes of areas such as tables and figures. In the figure, ■
2 is an image information storage section as an image storage means for storing a binary image (document image) obtained by optically scanning a document and photoelectrically converting it; 2 obtained by compressing the document image by dividing it into small regions and making each small region correspond to one pixel.
A compressed image storage section 3 serves as an image compression storage means for storing a value image (compressed image), and 3 a pixel feature storage section 3 serves as a feature extraction means for extracting and storing the features of each pixel of the compressed image from the corresponding document image. Section 4 scans the above compressed image in the direction of the character string, checks how many consecutive pixels have a value of "0", and sets the value of these pixels to "1" when the value is below a preset darkness value. A compressed image processing unit 5 converts the image (converted image) obtained by the compressed image processing unit 4 into “
A region detecting unit as a region detecting means detects a region where pixels of 1” are connected as an image region, and 6 stores the shape and pixel characteristics storage unit 3 for each image region detected by the region detecting unit 5. A character string detection unit detects a character string by determining whether or not the detected area is a character string area based on the ratio of pixel features stored in the area, and 7 indicates that the area detected by the area detection unit 5 is a character string area. If it is determined that the character string area is, a character string image storage unit 8 cuts out and stores an image corresponding to the character string area from the document image stored in the image information storage unit 1;
is a character recognition unit that cuts out each character from the character string image stored in the character string image storage unit 7, recognizes each character, and outputs a character code.

FIG. 4 is a diagram showing output images of each processing section in FIG. 3. In the figure, 9 is a document image stored in the image information storage unit 1, 10 is the document image @, 9 is compressed,
In the compressed image 11 stored in the compressed image storage unit 2, in a small area of the document image 9 corresponding to each pixel of the compressed image 10, the value of the pixel where a horizontal straight line exists is set to "1", and the value of the pixel that does not exist is set to "1". A pixel characteristic image with a value of "0", 12 is a converted image processed by the compressed image processing unit 4, and 13 is a region detection unit 5
An area determined to be a character string by the character string detection unit 6 (character string area) based on the shape of the image area detected and the proportion of linear features.
, 14 is an image of a character string cut out by the character string image storage section 7, and 15 is a recognition result in the character recognition section 8.

Next, the operation will be explained.

For example, 1024X1 stored in the image information storage unit 1
Divide the document image 9 consisting of 0.024 pixels into 128 x 32 small areas consisting of 8 pixels vertically and 32 pixels horizontally, count the number of pixels with a value of "1" in each small area, and if the counted value is 5 or more. If the pixel value is "1", if the pixel value is less than 5, the pixel value is "0", and a compressed image 10 is formed and stored in the compressed image storage section 2.

On the other hand, in each small area, the value is “1” from the left end to the right end.
Forming a pixel characteristic image 11 with a pixel value of "1" when the pixels of are connected in eight directions, and "0" otherwise,
It is stored in the pixel feature storage section 3.

Next, in the compressed image processing unit 4, the compressed image 10 is scanned in the character string direction (horizontal direction), and the number of consecutive pixels with the value "O" is determined. If the number is 3 or less, these pixels are scanned. A converted image 12 with pixel values rewritten to "1" is formed.

Next, the area detection unit 5 scans the converted image 12,
An area in which pixels with a value of "1" are connected in four directions is detected, and each detected area is inspected by the character string detection unit 6 to determine whether or not it is a character string area. This is because it is known that the shape of a character string is a horizontally long rectangle and is not a horizontally straight line. The character string region 13 is detected by inspection using information in the pixel characteristic image 11 indicating the existence of a straight line.

Next, a character string image 14 corresponding to the character string area 13 detected by the character string detection unit 6 is cut out from the document image 9, and the character recognition unit 8 extracts the image 14 of the character string from the image +114 of this character string using a well-known technique. Each character pattern is cut out and recognized, and a recognition result 15 is output.

In this way, conventional document reading devices cut out only the character string area from the document image and read this area. Therefore, the character string detection unit 6 checks whether the detected region is a character string region or not based only on the shape of the detected region and the proportion of linear features.

[Problem that the invention seeks to solve]

As mentioned above, conventional document reading devices only read character string areas, so for example, for a table area, the character pattern in the table is cut out and read, and for a photo area, the document image in that area is compressed. It was not possible to perform processing according to the attributes of the area, such as output. In addition, since only the shape of the area and the ratio of linear features were used to check whether it is a character string area, it is difficult to determine whether the detected area is a figure area or a table area. There were problems in that there were few criteria and it was difficult to distinguish.

This invention was made to solve the above-mentioned problems, and its purpose is to determine the attributes (character strings, figures, tables, photographs, etc.) of the detected area, and to determine the attributes of the area. The object of the present invention is to provide a document reading device that can perform various processes.

[Means for solving problems]

A document reading device according to the present invention divides a document image stored in an image storage means into small regions each consisting of a predetermined number of pixels,
an image compression storage means for forming a compressed image by compressing each small region into one pixel and storing the compressed image; a feature extraction means for extracting a predetermined feature of an image for each of the small regions from the document image; a region detection means for detecting an image region based on pixel connectivity; and a region detection means for detecting an image region based on pixel connectivity; The apparatus is equipped with an attribute determining means that makes a determination by comparing the characteristics of the image, and an area processing means that cuts out an area of the document image corresponding to the image area and performs processing according to the attribute of the image area.

[Effect]

In this invention, the image area is detected by the area detection means based on the compressed image compressed and stored by the image compression storage means. The attribute determining means determines the attributes of the image region by comparing the outline of the image region and the features extracted by the feature extracting means with features prepared in advance for each attribute such as a character string, table, figure, etc.

Then, based on this attribute, the area processing means processes the image cut out from the document image corresponding to the image area.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is an overall configuration diagram showing an embodiment of a document reading device according to the present invention. Incidentally, the same reference numerals are used for the same or corresponding parts as those shown in FIG. 3, and the explanation thereof will be omitted. In the figure, reference numeral 16 denotes an attribute determining unit as attribute determining means for determining the attributes of areas such as character strings 9, photographs, etc. for each image area detected by the area detecting unit 5. To determine attributes, the shape of the detected image area, the detected image area is divided into a predetermined number of partial areas, and the distribution of the proportion of pixel features in each partial area is used. This is done by identifying the model. Reference numeral 17 denotes a region processing section as a region processing means for processing the detected image region according to the attribute determined by the attribute determination section 16. To process a region, for example, if the detected region is a character string, cut out the corresponding region of the document image, cut out character patterns for each character from the obtained character string image, and recognize these to create characters. Output a string of code. If the detected area is a front, a character string in the front is detected, the corresponding area of the document image is extracted, character extraction and character recognition are performed, and a string of character codes is output together with the back structure. If the detected area is a photograph or drawing, the corresponding area of the document image is cut out, the resulting image is compressed, and then output.

FIG. 2 is a diagram showing output images of each processing section in FIG. 1, and 18 is a document image stored in the image information storage section 1, which consists of a table and a diagram.

19 is a compressed image stored in the compressed image storage unit 2 by compressing the document image 18, and 2o is a compressed image of the pixel where a horizontal straight line exists in each small area of the document image 18 corresponding to each pixel of the compressed image 19. A pixel characteristic image in which the value is "1" and the values of other pixels are "0", 21 is a converted image obtained by processing the compressed image 19 by the compressed image processing unit 5, 22 is a converted image 21
The area detection unit 5 detects an area where pixels with a value of "1" are connected in four directions, and in the circumscribed quadrilateral of the detected area,
As a result of examining the distribution obtained from the vertical length, the horizontal length, the ratio of pixel features, and the ratio of pixel features in each partial region when the region is divided into four, and determining the attribute in the attribute determination section 16, 2
3 is a character pattern in which characters are extracted from a character string image obtained by detecting a character string area from an area determined to be a table area by the attribute determination unit 16 and cutting out a corresponding area from the document image 18; 24 is a character pattern; This is an output result in which the result of recognizing the pattern 23 is output together with the back structure.

Next, the operation will be explained.

In such a configuration, a document image stored in the image information storage unit 1 and consisting of, for example, 1024×1024 pixels is divided into 12 pixels consisting of 8 pixels in the vertical direction and 32 pixels in the horizontal direction.
Divide into 8x32 small areas.

Next, the number of pixels with a value of "1" is counted in each small area, and if the counted value is 5 or more, the pixel value is "1", and if it is less than 5, the pixel value is set to "0" to create the compressed image 19. and stores it in the compressed image storage section 2.

On the other hand, in each of the above small areas, when pixels with a value of "1" are connected in eight directions from the left end to the right end, the pixel value is "
1”, and in other cases, a pixel characteristic image 20 is formed in which the pixel value is “O” and is stored in the pixel characteristic storage unit 3.

Next, the compressed image processing unit 4 scans the compressed image 19 in the direction of the character string, checks how many consecutive pixels with the value "O", and when the number is less than a predetermined number, changes the value of these pixels. “1”
A converted image 21 rewritten as follows is formed.

Next, the area detection unit 5 detects an area in which pixels having a value of "1" are connected in, for example, four directions in the converted image 21 as an image area.

Next, the attribute determination unit 16 determines the vertical and horizontal lengths of the circumscribed quadrilateral in each image area detected by the area detection unit 5, and the area of the pixel characteristic image or compressed image corresponding to the detected area. In addition to the ratio of the number of pixels with a value of "1" to the number of pixels of the circumscribed quadrilateral, a similar ratio is examined in each partial region obtained by dividing the circumscribed quadrilateral into four, and its distribution is determined.

Tables and figures are longer in the vertical direction than text strings, tables have many horizontal straight lines, and the distribution of pixel features between subregions is relatively stable, whereas figures generally have horizontal straight lines. The attributes of the image region are determined by identifying it with a pre-prepared model in which the distribution of pixel features between partial regions is often biased, and a determined result 22 is obtained.

Next, in the area processing unit 17, for the area whose result determined by the attribute determination unit 5 is a table, the character string area in the table is detected using the pixel feature image or the compressed image, and the corresponding area is detected. is extracted from the document image 18, the character pattern 23 is extracted from the obtained character string image, each extracted character pattern is recognized, the result is output together with the position of the column where the character string existed, and the output result 24 is obtain.

In the above example, the presence or absence of a straight line in a small area is extracted and used as a pixel feature, but circular arcs,
Other features that can be extracted in a small area, such as a broken line, solid black, and a thick straight line, may be used, and multiple types of pixel features may be used.

Further, as for how to use the feature information in each image region, the case where the detected image region is divided has been described, but the continuity, density, etc. of pixel features may be used as the feature information without dividing. Further, although this feature information has been exemplified in the case of distinguishing between a figure and a table, it can generally be used as a criterion for determining the attributes of an area.

Further, although the area processing has been described for table areas, similar processing can be performed for other attributes as well.

〔Effect of the invention〕

As described above, the document reading device according to the present invention divides the document image stored in the image storage means into small regions each consisting of a predetermined number of pixels, compresses each small region into one pixel, and stores the compressed image. an image compression storage means for forming an image; a feature extraction means for extracting a predetermined image feature for each of the small regions from the document image; and an area detection means for detecting an image region from the compressed image based on pixel connectivity. and attribute determination means for determining the attributes of the image region by comparing the image features of the entire region or each small region extracted by the feature extraction means with image features prepared in advance for each attribute; By including an area processing means that cuts out an area of the document image corresponding to the image area and performs processing according to the attribute of the image area, it is possible to increase the accuracy of attribute determination. It is also possible to process areas and tablespaces, and has the effect of configuring a flexible system.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram showing output images from each component in FIG. 1, FIG. 3 is a block diagram of a conventional device, and FIG. 4 is a block diagram of a conventional device. FIG. 3 is a diagram showing output images from each component of the device. 1 is an image information storage unit (image storage means), 2 is a compressed image storage unit (image compression storage unit), and 3 is a pixel feature storage unit (
4 is a compressed image processing section, 5 is a region detection section (means), 16 is an attribute determination section (means), and 17 is a region processing section (means). In addition, in the figures, the same reference numerals indicate the same or corresponding parts. Soil 1 map

Claims (2)

[Claims] (1) In a document reading device that is equipped with an image storage means for storing a document containing information other than characters as an image pixel by pixel, and that cuts out and reads a character pattern from the stored document image, the document image is divided into a predetermined number of pixels. an image compression storage means for forming a compressed image by dividing the document image into small regions, and compressing and storing each small region into one pixel; a feature extraction means; a region detection means for detecting an image region from the compressed image based on pixel connectivity; and a region detection means for detecting an image region from the compressed image based on pixel connectivity; attribute determination means for determining the image by comparing the image characteristics prepared in advance for each attribute; and area processing means for cutting out a region of the document image corresponding to the image region and performing processing according to the attribute of the image region. A document reading device comprising: (2) The attribute determining means divides the image region detected by the region detecting means into a predetermined number of partial regions, and for each of the obtained partial regions, the characteristic of each small region extracted by the feature extracting means is determined. Claim 1, characterized in that the appearance frequency is determined and the attribute of the detected image area is determined based on the value.
The document reading device described in Section 1.