JPH11316801A - Character read method and address read method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently read a handwritten address character string on a postal matter or the like. SOLUTION: Characters are temporarily segmented from picture information of an address character line (206) to generate a table 219 of hypothesis character patterns, and character discrimination is performed for hypothesis character patterns (207), and especially, peripheral information (the length, the width, the aspect ratio, spaces before and after patterns, etc.), are extracted with respect to characters in the block part of the address character string (212), and this information is used to segment characters with a high precision (215). Thus, the character segmentation precision is improved to improve the recognition precision of the address character string. The precision of character segmentation of the handwritten address character string and that of a character string written in the vertical direction with Chinese numerical characters are improved.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character reading method, and more particularly to a character reading method suitable for reading an address character string described in a mail or the like.

[0002]

2. Description of the Related Art As a method of automatically reading an address of a postal matter, a prefecture name of an address written on a mail,
Various methods have been devised for automatically and appropriately reading place names such as city names, town names, and the like using an electronic device.
In Japanese Patent No. -64882, an address is recognized by performing different character extraction processing on a character string from the head of an address to a town name and a character string on a street address. As a method of detecting an address description area on a mail, Japanese Patent Application Laid-Open No.
No. 1389 is a method of detecting a postal code position and detecting an address writing area from this position. As a method of extracting a character line from the address description area, a patent application publication 6
No. 0-41396, which is a method of detecting the height of a block pattern of a written character or the like and detecting a pattern having the same height as a line. Japanese Patent Application Laid-Open No. 63-18785 discloses a method for judging the vertical and horizontal writing of the extracted character line. This method determines the horizontal and vertical peripheral lengths of a character group in an address area, and compares these peripheral lengths. To determine the line direction of the character group.

A technique for extracting individual characters from a character line is described in IEICE Transactions (D) J68-D, No. 1
2. As shown on pages 2123 to 2131, character extraction is determined based on character width, character pitch, character spacing, and the like. In addition, IEICE (D) J6
8-D, no. 4, page 765 to page 772, there is also a method of determining character cutout using information on character identification results. But this is also a separator,
In particular, it is insufficient to resolve the ambiguity of character segmentation in a series of vertical Chinese characters.

[0004]

As for the above-mentioned prior art, particularly for individual character extraction from a character line, a character pattern such as a character width, a character pitch, a character interval, or the like can be used. In the method of determining character segmentation using character similarity, in the case of a handwritten character string, feature amounts such as character width, character pitch, character length, etc. are not constant,
Because of the ambiguity, it is difficult to determine character extraction in a handwritten character string based on such information, and it has been difficult to accurately extract characters. In addition, even when the character cutout is determined by integrating the feature amounts such as the length and width of the pattern, the aspect ratio, the gap before and after the pattern, and the similarity of the candidate character obtained from the character identification result, the optimum length and width depending on the character type is determined. The length, the aspect ratio, the character spacing, and the like are different, and are insufficient for use as information for character cutout. For example, as shown in the address description surface 101 in FIG. 1, in the case of the Chinese numeral "3-2-1" of the address description portion 102 (hereinafter abbreviated as the block description portion) 102, the length and width of the pattern and the aspect ratio are determined by the character type. , The character segmentation becomes ambiguous, and there are various interpretations of recognition results. Therefore, as shown in 103, various recognition results (six patterns) generated due to the ambiguity of character segmentation of the block description part are expected. The recognition result 104 is a correct recognition result.

Accordingly, it is an object of the present invention to solve this problem and to realize a character reading method for accurately identifying each character from a character line. It is another object of the present invention to realize a method for accurately reading an address character including a town area description part and a block description part.

[0006]

In order to achieve the above object, a character reading method according to the present invention provides a connected component (for example, a continuation of black pixels) in a character line to be identified which is electronically imaged by an image input device. Stroke)
Cut out a plurality of characters that may be characters (this cut-out pattern is hereinafter abbreviated as a hypothetical character pattern).
A second step of performing character identification on the hypothetical character pattern by referring to a character identification dictionary and obtaining a character identification result (recognition candidate character and similarity thereof) as auxiliary information of the hypothetical character pattern; In addition to the information, the third step of obtaining the hypothetical character pattern peripheral information and the extraction dictionary having character peripheral information with the recognition candidate character obtained in the second step as a key, and in the third step, A fourth step of obtaining the certainty factor of the peripheral information of the extracted hypothetical character pattern and weighting the hypothetical character pattern; a fifth step of determining a character cutout using the weight of the hypothetical character pattern; The characters to be identified by performing word matching using a character identification dictionary based on a set of character identification results generated from the character cutout results determined by Performs recognition of broadcast having a sixth step of reading the character line.

When reading a character string of an address in which a town area and a block are written by using the above-described character / character reading method, image information including the character string of the address is converted into an electric signal, and the electric signal is read. A character line area of the character string of the address is cut out from the image information, and a plurality of hypothetical character patterns are generated by combining the connected components in the character line cut out in the first step.

A character identification dictionary is used to identify each of the plurality of hypothetical character patterns, and information on candidate characters similar to the hypothetical character patterns is obtained. The town area of the address is read by referring to the town area dictionary using the information of the candidate characters, and the head position of the block is detected. The town area dictionary stores all possible town names.

When the head position of the block is detected, pattern peripheral information of the hypothetical character pattern of the block is obtained, and a certainty factor of the pattern peripheral information is obtained by referring to a character extraction dictionary based on the pattern peripheral information. Considering the certainty factor, re-determining the character cutout of the hypothetical character pattern for the character of the block, and referring to the block dictionary using information on candidate characters similar to the hypothetical character pattern obtained in the re-determination. Then, the character reading of the above-mentioned block is determined. The block dictionary stores character information of all blocks that can be considered as blocks.

As described above, according to the character reading method of the present invention, the extraction of a character line to be identified is reflected on the character extraction based on the similarity obtained by character identification, and the aspect ratio, aspect ratio and pattern By integrating the method of reflecting the feature amount of the interval length in character segmentation, the disadvantages of each other's method are compensated for and the effective information is optimally used to recognize the character string, and character width, character pitch, character spacing, etc. Even if is irregular, it enables accurate character extraction.

In character extraction in recognition of a handwritten character string, it is difficult to estimate a common character width, character pitch, and character interval length for all characters, and to extract characters based on the estimated values. In order to solve this problem, prepare a character extraction dictionary for character extraction hypothesis testing, use the character type of the character identification result as a key, refer to the dictionary, compare it with the character pattern feature, and calculate the confidence factor. To reflect on character extraction.

In this character extraction dictionary, a likelihood distribution is calculated in advance as a certainty factor of a pattern with respect to each feature amount and stored in the dictionary. However, there is a problem that calculating this confidence requires a great deal of effort, time, and empirical knowledge.
According to the present invention, it is possible to generate an optimal weighted certainty factor for each character type only by displaying the character extraction hypothesis on the screen and selecting the correct character extraction hypothesis.

[0013]

FIG. 2 is a processing flowchart showing an embodiment of a character reading method according to the present invention. In this embodiment, the character reading method is applied to automatic reading of an address of an address. In the reading process of the address consisting of the town area and the block, the address writing surface 201 of a mail or the like is captured as an image input by the imaging means (scanner unit) (20).
2) The information of the address description area including only the address information is extracted from the image input (203), and a character line is cut out from the address description information (204). The image information from which the character line has been cut out is subjected to vertical / horizontal writing determination to determine whether the character string in the character line is vertical or horizontal writing (205), and the vertical / horizontal writing processing mode is switched using the determination result. (22
1).

Next, character extraction processing is performed at a place where a character may be constituted by a combination of character black pixel blocks (connected components) in a character line. Hereinafter, this character extraction processing is abbreviated as character extraction hypothesis generation (206), and the character pattern extracted by the extraction hypothesis generation is abbreviated as a hypothesis character pattern. A plurality of hypothetical character patterns include an inappropriate pattern in addition to a correct character pattern to be read. The generated hypothetical character pattern is registered in the pattern table 219. The hypothetical character pattern and pattern table 219 will be described in detail with reference to FIGS.

Character identification is performed on all the hypothetical character patterns registered in the pattern table 219 using the character identification dictionary 208 (207). The character identification is performed by recognizing information such as a plurality of recognition candidate characters similar to each hypothesis character pattern, the degree of similarity of each recognition candidate to the hypothesis character pattern, the position of the hypothesis character pattern in the character line, and the number of connected components. Request as attribute information. The recognition candidate character and the attribute information are written in the pattern table 219 for each hypothetical character pattern. Based on the pattern table 219 in which the recognition candidate characters and the attribute information are written, a proper character cutout is selected from the hypothetical character patterns, and character cutout is determined (209). A set of recognition candidate character strings is created based on the recognition candidate character strings corresponding to the hypothetical character pattern selected by the character extraction determination (2).
10). In this set of recognition candidate character strings, character types up to lower-order candidate characters are registered for each hypothetical character pattern. This registered character type is called a lattice.

Using this lattice, town area matching is performed using the town area dictionary 220 (211), and a character string appropriate as a character in the town area of the address is generated. The town area dictionary 220 describes all possible town area names. In addition, the reading of the character describing the town area is completed by the town area collation 211, and the last character of the character describing the town area is determined, and the block start position information that the next is the block start position is obtained. .

The above-mentioned block head position information is obtained, and information on the vertical length, horizontal length, aspect ratio, pattern interval length, and number of connected components (referred to as pattern peripheral information) of only the block portion in the pattern table 219 is extracted. Yes (21
2). The degree of certainty for the extracted pattern peripheral information is calculated using the character extraction dictionary 214 (213). The calculated certainty factor is written in the pattern table 219 as the attribute of the corresponding hypothetical character pattern. The hypothesis character pattern in the pattern table 219 is reselected using the above-mentioned certainty factor, that is, the character extraction re-determination is performed which differs from the hypothesis character pattern selected at the time of the character extraction determination 209 only in the block (215). . After the character cutout redetermination 215 is performed, the lattice is regenerated based on the information (216). Using the regenerated lattice as an input, block matching is performed using the block dictionary 222 (217), and block characters are read. The block dictionary prepares all block characters that can be used as blocks. Thus, the reading of the address is completed together with the characters of the town name read in the town area comparison (211) (address recognition result 218).

Hereinafter, details of each unit of the processing of FIG. 2 and an apparatus for performing the processing will be described. The image input 202 and the vertical / horizontal switching 221 are the same as those in the prior art.
One part is substantially the aforementioned IEICE (D) J68-D,
No. 4, pages 765 to 772 can be applied.

FIG. 3 shows the configuration of an apparatus for implementing the above address reading method. In the figure, reference numeral 301 denotes an image input 20
2, a sorter unit for sorting mails based on the address reading result, that is, the sorting information of the address recognition result 218 in FIG.
This is a recognition device that performs processes 202 to 222 in FIG. The scanner unit 301 and the recognition device 312 are connected by an input / output cable 304. Sorter unit 303 and recognition device 3
12 is connected by an input / output cable 305.

The recognition device 312 includes a bus 313 for connecting various parts inside the recognition device, an interface unit 306 for communicating with the scanner unit 301, and an arithmetic processing unit for controlling the entire recognition device 312 and performing address recognition processing. 30
7, an interface unit 308 for communicating with the sorter unit 303, and a keyboard 30 for performing operations at the time of startup and the like
9. CRT 31 for displaying the execution status as needed
0, a memory 311 for storing tables, programs, dictionaries, etc. necessary for reading addresses.

FIG. 4 is a view for explaining the processing of character line extraction 204 from the image input 202 of FIG. Reference numeral 407 denotes an image of the destination area cut out from the image input 202 by the destination area cutout 203. The histogram in which the black pixels in the address area 407 are projected on an axis 408 parallel to the y-axis 402 is 4
03, 404, and 405. By detecting the values of these histograms, the upper and lower y-coordinates of the character line such as the dotted line 406 are extracted, and the character line is extracted (204).

FIG. 5 is a diagram for explaining the processing of the vertical / horizontal writing determination 205 in FIG. Reference numeral 501 denotes an image of the extracted horizontally written character string. Numerals 502 and 503 denote character patterns at the beginning and end of the character line.
03 505 and 509
The character patterns 506 and 510 are rotated by 90 degrees in the pattern rotation processes 504 and 511. Character recognition is performed on these character patterns (507). The similarity of the result is compared (508), and vertical / horizontal writing determination 205 is performed based on the result. Based on this determination result, input switching between vertical writing feature extraction and horizontal writing feature extraction is performed (FIG. 2).
221).

In the case of using the layout information in the image, the possibility of erroneous determination is high when an image including an address character string other than the prescribed layout is input. However, according to the present embodiment, Since the discrimination is performed using the character recognition itself, the vertical and horizontal writing can be discriminated stably. If the difference between the first and last characters of the address string and the similarity between the character in the same direction and the character rotated 90 degrees is not large, cut out one character from the beginning and end of the address string. A similar discrimination method is performed by shifting them one by one. As a result, characters that are unlikely to cause a difference in similarity even when rotated by 90 degrees such as “field” or “mountain” are avoided, and vertical / horizontal writing determination using characters suitable for determination is performed, thereby improving accuracy. .

FIG. 6 is a diagram for explaining a hypothetical character pattern in association with an input image. Using the handwritten address character string 601 to be identified as an input image, the character extraction hypothesis generation 20 of FIG.
6, a boundary line that can be cut out as one character (dotted line 6)
03-i (i = 1, 2,..., N)) is used to cut out the pattern of the input image. (0),..., 602-i (i
= 1, 2,..., N) are called nodes. A line 604 connecting the nodes is called an arc, and patterns 605, 606... 611, etc. corresponding to the arc 604 are hypothetical character patterns. For example, the character pattern “Hyo” to be read can be divided into a hypothesis character pattern of a pattern “Hoka” 606 and a pattern “Hachi” 607 outside the pattern “Hyo” 605 as the possibility of the character cutout hypothesis. Also, the pattern "OK"
Similarly, in addition to the pattern “OK” 610, the pattern “One” 6
09 and a pattern 611. As described above, a plurality of hypothetical character patterns exist between the nodes connected by the arc.

FIG. 7 shows the data structure of the pattern table 219. Reference numeral 701 denotes a pointer indicating a location in a memory where image information of a pattern cut out as a hypothetical character pattern is stored. 702 stores the degree of certainty on the arc corresponding to this hypothetical character pattern (representing the weight, which depends on the distance between the start point and the end point of the node, and the longer the distance, the greater the weight). Reference numeral 703 stores the number of connected components in the hypothetical character pattern (for example, the character pattern “3” has three connected components, and “eight” has two connected components). The coordinates (for example, the coordinates of the upper left and lower right points of the block surrounding the hypothetical character pattern “hill” in FIG. 6) are stored. Reference numeral 705 stores the number of the node before the arc, and reference numeral 706 stores the number of the node after the arc. By referring to this node number, a network of a hypothetical character pattern can be expressed.
Reference numeral 707 stores a plurality of candidate character strings obtained when the hypothetical character pattern is subjected to character identification 207 with reference to the character identification dictionary 208. Reference numeral 708 stores a degree of similarity indicating the degree to which each candidate character is similar to the hypothetical character pattern. It is.

The method of calculating the similarity is not particularly specified, and a conventionally known method can be used. A bold line 709 indicates the range of one temporary character pattern table, and this range corresponds to one arc. For example, when the temporary character pattern is “soldier”, it corresponds to one arc 604-1.
Therefore, the previous node number 705 is (0) and the rear node number 706 is

FIG. 8 is a view for explaining the character cutout determination processing 209 of FIG. The figure shows a hypothetical character pattern string uniquely determined by performing the character extraction determination processing 209 based on the data of the pattern table 219. In the character cutout determination process 209, the similarity of candidate characters obtained when character identification is performed on all the hypothetical character patterns in the pattern table is registered as the certainty factor 702 of each arc. For a plurality of routes based on the combination of the arcs following the number 8, the sum of the certainty factors of the arcs of each route is obtained, and the hypothetical character pattern sequence in the route having the highest total certainty factor is determined as character extraction.
FIG. 8 shows that the root with the highest total confidence is node (0) −
−−−−−.

When FIG. 8 is compared with FIG. 6, in FIG. 8, there is no arc 604 from the node number (0) of FIG. 6, and no arc 604 from the node number. ing. As described above, by evaluating all the hypothetical character patterns in the pattern table 219 by character identification, the certainty of the character-like pattern is increased, and character cutout is determined (FIG. 2).
209) is performed.

FIG. 9 is a diagram for explaining the recognition result of the town area portion obtained by the town area collation processing 211 with the input of the character identification result for the uniquely determined character segmentation and the start position of the town block portion. Reference numeral 601 denotes an image of an address character string to be read, dotted line 902 denotes a boundary determined by character extraction, and 707 denotes a candidate character string of a character identification result of a hypothetical character pattern corresponding to character extraction. The character selected as a result of performing the town area matching (211 in FIG. 2) based on the candidate character string 707 is a character 903 surrounded by a circle, and each selected character “Holder”, “Storage”. To generate a character string 910 of the town area recognition result. A dotted line 905 indicates a range in the input image 601 where the character string obtained by the town area matching corresponds. The start position 911 of the block is determined by the town area matching process 211.

In the figure, 906, 907, 908,
Reference numeral 909 denotes a hypothetical character pattern in the block, and 912...
Reference numeral 916 denotes a candidate character string of the character identification result for the hypothetical character pattern in the block. These are obtained by the processing up to lattice generation 211 in FIG. The block starting position 911 and thereafter are often described with kanji numerals or arithmetic numerals, and the character segmentation process of the block portion is different from the character segmentation process of the town area portion described with kanji.
In other words, when the same method as the character extraction of the town area is applied to the block, the pattern “three” is changed to the hypothetical character pattern 90.
6, 907, and the hypothetical character pattern “two” is often separated as 908, 909. Also, the characters used are limited.

FIG. 10 is a flowchart showing the process of re-determining the character segmentation of the block. That is, it is a processing flow diagram of the pattern peripheral information extraction 212 or the character cutout re-determination processing 215 of FIG. An example of a specific character pattern will be described later in detail with reference to FIGS. Block top detection 1013 is performed using the block top position information 911 obtained from the town area comparison 211 as input, and recognition candidate characters for the hypothetical character pattern in the block are extracted from the pattern table 219 as character type information (1002). In the case of this embodiment, the candidate character having the highest similarity among the candidate character strings resulting from the character identification 207 is set as the character type information. The extraction dictionary 214 is referred to using the extracted character type information as a key. At the time of character type extraction 1002, peripheral information of a hypothetical character pattern corresponding to the character type is extracted (212). The above peripheral information is the length, width, aspect ratio,
This is data on the pattern interval length and the number of connected components.

The extraction dictionary 214 is referred to by using the character type as a key to extract the portrait, landscape,
The likelihood ratio of the aspect ratio, the pattern interval length, and the number of connected components to peripheral information is obtained. Here, the likelihood ratio for each value of the peripheral information is registered as a certainty factor in the dictionary 214, and the likelihood ratio for each peripheral information value is calculated as the certainty factor (1005, 1006, 1007, 1008). . The calculated certainty factors are integrated (1010). An event that was correct and H as character types cutout hypotheses character pattern is character _{identification, e 1, e 2, e} 3, ... as the feature amount of the peripheral information of e _n pattern, P | a (e H) H
Assuming that the occurrence probability of e is the case, the likelihood ratio L (e _k | H) with respect to a certain feature value e _k is obtained by the following equation.

[0033]

[0034]

(Equation 1)

Here, let P (H) be the occurrence probability of H,
Using a plurality of likelihood ratios determined by equation (1), e ₁ ,
e ₂ , e ₃ . . . , The probability of occurrence of H when e _n is obtained P
_{(H | e 1, e 2} , e 3, ..., e n) Ba when obtaining the
If yes rule is used, the following equation is obtained.

[0036]

(Equation 2)

Using formula (2), certainty integration processing 101
In the case of 0, the likelihood ratios calculated in each of 1005, 1006, 1007, and 1008 are integrated using Bayes rule. Next, a value obtained by multiplying the integrated certainty by the similarity obtained by character identification is reflected on the arc and weighted (1011). The optimal route from the first node to the last node is searched based on the hypothetical character pattern data weighted to the arc as described above, and a character extraction path is determined (1012). The result is used in block matching (217 in FIG. 2).

FIG. 11 is a diagram for explaining a hypothetical character pattern corresponding to a block in an address character string. The street section 1102 of the input image of the address character string is extracted as a character extraction hypothesis generation 206 and a boundary line 1104 in FIG. 2 and recorded in the pattern table 219 as a hypothesis character pattern. The data structure in the pattern table is shown in FIGS.
This is the same as that described above. A dotted line 1103 indicates the correspondence between the cutout boundary line 1104 and the node number 1105 (numbers 51, 52,. For example, the hypothetical character pattern “three” 1106 (arc 1107)
), A candidate character string 1108 of “3”, “3”, “5” is obtained. When the hypothetical character pattern “3 |” 1109 (corresponding to arc 1111) is subjected to character identification processing, candidate character strings 1110 of “show”, “not”, and “body” are obtained.

Also, Regarding the weighting of the arc corresponding to each hypothesis character pattern, the weighting described in FIG. 10 is performed. For example, the above character type "show""not""body" 1
Since 110 is an inappropriate character that does not occur as a character in the block, the weight of the arc 1111 corresponding to the hypothetical character pattern 1109 is reduced. By this weighting process, arcs that are clearly inappropriate as a block portion are deleted. Other hypothetical character patterns (eg, "two",
For "1", "1 |", "|", etc., the corresponding arcs are weighted, and inappropriate arcs are deleted.

FIG. 12 is a diagram for explaining arc weighting processing 1011 for the hypothetical character pattern in the block section in FIG. The hypothetical character pattern “3” 1106 will be described in detail as an example. First, peripheral information is extracted from the hypothetical character pattern "3" (212). Here, the peripheral information refers to values of the vertical length, horizontal length, aspect ratio, number of connected components, and character interval length of the pattern, and the candidate characters “3,
The first place candidate character “3” of “3, 5” 1108 is also referred to here. Block 1213 summarizes this information.
Are shown as 1214 and 1215 in FIG.

In character type extraction 1002, information 1214 of character type “three” is sent to character extraction dictionary 214. The extraction dictionary 214 stores data 1205 for arc weighting that can be referred to using the character type “3” as a key, and the index in the extraction dictionary 214 using the character type obtained in the character type extraction 1002 as a key. "three"
Search for. When searching for the character type "3", the indexed data 120 that matches the extracted character type "3"
5 to vertical 1206, horizontal 1207, aspect ratio 12
08, the number of connected components 1209, and the likelihood ratio corresponding to the character length 1210, and the confidence of each peripheral information is obtained (12
04), the likelihood ratios (confidence 1,
2, 3, 4, and 5 are integrated (1010), and the weight 1101 is used to weight the arc 1107 corresponding to the hypothetical character pattern “three” 1106. As described above, the peripheral information of the pattern is converted to the arc 1 of the hypothetical character pattern 1106.
In step 107, optimal weighting is performed according to the character type.

FIG. 13 is a diagram showing the result of the weighting in terms of the thickness of the arc line. Reference numeral 1301 denotes an arc that is weighted in consideration of the peripheral information corresponding to the hypothetical character pattern “three” 1106. Numeral 1303 denotes an arc weighted in consideration of the peripheral information corresponding to the hypothetical character pattern "|". The weight also becomes extremely large when the peripheral information is added. The numeral 1302 connects the node numbers 52 and 54 with a hypothetical character pattern arc very similar to the character pattern of “2” formed by combining the lower two connected components of the pattern “3”. This hypothetical character pattern "two"
When the character is identified, as shown in FIG.
Candidate characters “3” and “2” are obtained, and among them, the character type “2” also has a large similarity. Therefore, there is a risk that this hypothesis is determined to be valid as character extraction. . However, considering surrounding information, arc 13
Since the pattern corresponding to 02 has a narrow gap with the pattern one immediately above, when the character extraction dictionary 214 is referred to, the confidence of the pattern on the arc 1302 is low, and the weight of the arc 1302 is compared with the case where only the similarity is used. And then lower. For this reason, the weighting on the arc 1301 is larger, and the total of the weights of the route passing through the arc 1302 is relatively smaller.

FIG. 14 shows a hypothetical character pattern string selected by re-determining the character segmentation of the block. That is, when the arc corresponding to the hypothetical character pattern is weighted by weighting the arc (1011 in FIG. 10) and the path is determined so that the sum of the weights is maximized, the arc 1301 of the hypothetical character pattern “three” is determined. The arc 1303 of the hypothetical character pattern "|" is selected, and the route is selected. In this manner, in the character segmentation redetermination of the block portion, the hypothetical character patterns “3”, “|”, “2”,
The arcs corresponding to "|" and "one" are selected to form a path. Using the candidate character string attached to the selected arc pattern, a lattice of the block is generated (216).

FIG. 15 shows the result of recognition of the entire address character string in the above embodiment. That is, the recognition result of the entire address character string is shown by combining the town area collation result with the character string obtained by performing character extraction for the city block, lattice generation, and city area collation processing. In the input address character string image 601, a dotted line 905 indicates a range of a town area portion, and 910 indicates a town area collation result shown in FIG. Dotted line 1510 indicates the cutout boundary of character cutout redetermination shown in FIG.
A character string 1506 indicates a candidate character string of a character identification result for each character of the character cutout result. Dotted lines 1507 and 1509 indicate the range of the block, and a character string 1508 indicates the block recognition or reading result obtained by the block verification 217 from the candidate character string of the character identification result. A character string 1504 indicates a recognition result of the entire destination character string generated by combining the town area comparison result 1502 and the block comparison result 1508. As described above, the character segmentation is retried only for the block portion and is combined with the town area comparison result, thereby improving the recognition accuracy of the entire address character string.

FIG. 16 shows an example in which the input destination character string, the pattern table of the character extraction hypothesis, and the character identification result are displayed on the screen. 1600 on the screen of the display device 310 in FIG.
Includes an input address character string image 1601, pattern table nodes 1602-i (i = 1, 2,..., 8), arcs 1603-1 and 1603-4 connecting nodes, and adjacent nodes. Connecting arc, hypothetical character pattern 1604
−j (j = 1, 2,..., 10), hypothetical character pattern 1604
Candidate character string 160 obtained by character identification for −j
5 is displayed. In this manner, by displaying the screen on the screen 1600 of the display device 310, it becomes easy to intuitively grasp the progress of character segmentation and character identification processing during address character string recognition processing, and to maintain and function the apparatus. Useful for expansion. In addition, when creating and updating the character extraction dictionary 214, it is necessary to collect peripheral information of a pattern extracted based on a hypothesis.

Referring to the equation (1), the likelihood ratio stored in the character extraction dictionary indicates the distribution of the peripheral information of the pattern of the hypothesis where the character extraction was correct, and the hypothesis that the character extraction was incorrect. Is divided by the distribution of the peripheral information of the pattern. Therefore, when a certain arc is designated by the pointer on the screen display as shown in FIG. 16 and information such as each parameter of the peripheral information and the character identification result can be output, the peripheral information of the pattern is easily classified as the correct answer. Can be collected. The tool displaying the screen is also effective for collecting a character pattern required for generating and updating a character identification dictionary for character identification processing, and can reduce the amount of work.

[0047]

As described above, according to the present invention, the accuracy of character extraction, that is, character reading, is improved. Especially when reading the address of a handwritten address on mail,
When the processing of the street name description section and the block description section are made different, and the present invention is applied to the character reading of the block description section, the characters used for the block are limited, so the character extraction and the dictionary may be small in size, and therefore, Processing time can be reduced. It is possible to efficiently and accurately read an address including the street name description section and the block description section.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating an example of an input image in which ambiguity in character segmentation occurs in the related art.

FIG. 2 is a processing flowchart showing an embodiment of a character reading method according to the present invention.

FIG. 3 is a configuration diagram of one embodiment of a character identification device that performs a character identification method according to the present invention.

FIG. 4 is a diagram illustrating a process of character line extraction 204 of FIG. 2;

FIG. 5 is a diagram illustrating a process of vertical / horizontal writing determination 205 in FIG. 2;

FIG. 6 is a diagram illustrating the character segmentation hypothesis generation 206 of FIG. 2 in association with an input image.

FIG. 7 is a diagram showing a data structure of a pattern table 219 in FIG.

8 is a diagram conceptually showing a sequence of hypothetical character patterns uniquely determined by the character cutout determination process 209 in FIG.

FIG. 9 shows the lattice generation 210 and the town area collation 211 of FIG.
It is a figure explaining the processing of.

FIG. 10 is a diagram illustrating a process 215 of character segmentation and redetermination for Chinese numerals and arithmetic numerals only in the block portion of FIG. 2;

FIG. 11 is a diagram illustrating character identification indicating correspondence of a hypothetical character pattern to an input image of a block section.

12 is a diagram illustrating a process of calculating the degree of certainty for the hypothetical character pattern in FIG. 10;

FIG. 13 is a diagram showing a result obtained by calculating the certainty factor of a pattern and reflecting it on the weighting of an arc with respect to a character segmentation hypothesis in a block area.

FIG. 14 is a diagram illustrating a character cutout selected when the character cutout of the block is re-determined.

FIG. 15 is a diagram showing the recognition result of the entire address character string by combining the recognition results of the town area part and the block part of the address character string.

FIG. 16 shows maintenance of an address recognition device according to the present invention;
FIG. 7 is a diagram showing an example of a screen display of a tool used for function extension and dictionary generation / update.

[Explanation of symbols]

206: character extraction hypothesis generation, 209: character extraction determination, 210: lattice generation, 211: town area collation, 2
12: Pattern peripheral information extraction 213: Pattern certainty factor calculation, 214: Character extraction dictionary, 215: Character extraction redetermination, 216: Lattice regeneration.

Continuing from the front page (72) Inventor Hiromichi Fujisawa 1-280 Higashi Koikekubo, Kokubunji-shi, Tokyo Inside the Hitachi, Ltd. Central Research Laboratory (72) Inventor Hisao Ogata 1-280 Higashi Koikekubo, Kokubunji-shi, Tokyo Hitachi, Ltd. Inside the research institute (72) Yoshihiro Shima 1-280 Higashi Koikekubo, Kokubunji-shi, Tokyo Inside the Central Research Laboratory, Hitachi, Ltd. (72) Inventor Shigeru Watanabe 1 Ikegami, Haruoka-cho, Owariasahi-shi, Aichi Pref. Inside

Claims (7)

[Claims] A first step of converting image information including character line information into an electric signal; extracting a character line region of the image information converted into the electric signal to form a character in the character line region; Second to cut out a combination of connected components as multiple hypothetical character patterns
A third step of identifying a character by referring to a character identification dictionary for the hypothetical character pattern to obtain a plurality of candidate characters for recognition, a fourth step of extracting peripheral information of the hypothetical character pattern and the third step, With reference to the extraction dictionary having the peripheral information of the character using the recognition candidate character obtained in the step as a key, the confidence of the peripheral information of the hypothetical character pattern extracted in the fourth step is obtained, and the hypothetical character pattern is weighted. A fifth step, a sixth step of determining character extraction using the weighting of the hypothetical character pattern, and a character identification dictionary based on a set of character identification results generated from the character extraction results determined in the sixth step. 7. A character reading method comprising the steps of: performing a word collation to recognize the character information to be identified; and reading the character line. Method. 2. The character reading method according to claim 1, wherein the peripheral information of the hypothetical character pattern is information of a vertical length, a horizontal length, an aspect ratio, a pattern interval length, and the number of connected components of the hypothetical character pattern. The cut-out dictionary stores likelihood distribution information for each of the characters in the portrait, landscape, aspect ratio, pattern interval length, and number of connected components. Reference to the character cut-out dictionary in the fifth step is a candidate for the character identification result. A method of reading an address, wherein the character extraction dictionary is referred to by using a character type of a character as a key. 3. The method according to claim 1, wherein the fifth step includes a generation / update step of generating / updating the character extraction dictionary, wherein the generation / update step includes a step of generating a character string to be read on a screen of a display device. An input image, the hypothetical character pattern, a display step of displaying candidate characters of a character identification result for the hypothetical character pattern, and an updating step of updating a character extraction dictionary by using a screen display of the display device. And a step of graphically displaying a character string recognition processing state on a screen. 4. A first step of converting image information including character line information having a street name description portion and a block description portion into an electric signal, extracting a character line region of the image information converted into the electric signal, Second to cut out a combination of connected components constituting the characters in the character line area as a plurality of hypothetical character patterns
And a third step of identifying a character by referring to a character identification dictionary for the hypothetical character pattern to obtain a plurality of recognition candidate characters and the recognition candidate character. A fourth step of recognizing the character line of the street name description portion from the plurality of hypothetical character patterns and detecting the top position information of the block description portion, and a block description portion based on the top position information obtained in the fourth step. A fifth step of extracting pattern peripheral information of a hypothetical character corresponding to a candidate character for recognition of the hypothetical character pattern of
Fifth, weighting a hypothetical character pattern for obtaining a certainty factor of the pattern peripheral information of the hypothetical character pattern with reference to the extraction dictionary in which the likelihood corresponding to the pattern peripheral information is described using the recognition candidate character as a key. A sixth step of extracting a character of a block described by the step and the weighting, and a second step of performing word matching with a block dictionary based on a set of character identification results generated from the sixth step and reading the block character line to be identified. An address reading method comprising seven steps. 5. The method according to claim 4, wherein the second step is a step of extracting characters from the beginning and end of the extracted character line and identifying vertical and horizontal writing by identifying the extracted character pattern. Address reading method characterized by having. 6. The method according to claim 4, wherein the pattern peripheral information includes a vertical length, a horizontal length, an aspect ratio, a pattern interval length,
The character extraction dictionary has information on the number of connected components, and stores the vertical length, the horizontal length, the aspect ratio, the pattern interval length, and the likelihood distribution information for the number of connected components for each character used in the block description. An address reading method using the characters of the character extraction dictionary using the recognition candidate characters of the step as keys. 7. The method according to claim 4, wherein the fifth step includes a generation / update step of generating / updating the character extraction dictionary, wherein the generation / update step includes an address character string to be read on a screen of a display device. A display step of displaying the input image, the hypothetical character pattern, and candidate characters of the character identification result for the hypothetical character pattern, and an updating step of updating a character extraction dictionary using the screen display of the display device. And a step of graphically displaying, on a screen, a recognition processing state of the address character string, the address reading method.