JP5288915B2 - Character recognition device, character recognition method, computer program, and storage medium - Google Patents

Description

  The present invention relates to a technique for recognizing characters from an image.

  There is a technology for recognizing handwritten characters in a frame for entering a plurality of characters surrounded by ruled lines. In this technique, first, a character string is divided into pixel blocks, and a pattern in which the pixel blocks are combined is regarded as a character and recognition processing is performed. Then, a character evaluation value is calculated using the recognition result, and a row of pixel block combinations that maximizes the sum of the evaluation values is searched using dynamic programming.

The following conventional techniques exist as a method for reducing the number of combinations of pixel blocks.
(1) As a technique for narrowing down a column of pixel block combinations by using regularity of character arrangement such as a chome / address part, for example, there is Patent Document 1.
(2) As a technique for narrowing down a column of pixel block combinations in consideration of limiting character types in specific parts such as place name parts, chome / address parts, there is, for example, Patent Document 2.
Japanese Patent Laid-Open No. 06-124366 Japanese Patent Laid-Open No. 08-243504

  However, even when using the conventional technique as described above, if there are many pixel clusters that are separated from each other in radicals such as bias and heel in recognition of the place name part, processing is performed using dynamic programming. May take a long time. In addition, when the number of pixel block combinations increases, the false recognition rate may increase.

In view of the above problems, a character recognition device according to the present invention is
Pixel block extraction means for extracting a plurality of pixel blocks constituting a character image from input image data;
Detecting means for detecting a continuous portion of a specific character from the plurality of pixel blocks;
In a pixel block excluding a portion where the specific character detected by the detection means is continuous among the plurality of pixel blocks, a distance between adjacent pixel blocks is within a predetermined distance determination threshold, and the adjacent When the aspect ratio of the pixel blocks to be combined is within a predetermined aspect ratio determination threshold , combining means for combining the adjacent pixel blocks,
Character candidate corresponding to the combined pixel block by character recognition processing, and character candidate acquisition means for acquiring an evaluation value corresponding to the character candidate;
Determining means for determining a character candidate that maximizes the evaluation value in the entire character image ;
Wherein the specific character that is detected by the detection means and numbers, and a separator between the between the numbers and numbers, wherein Rukoto included.

  According to the present invention, when characters are automatically read from an image, recognition speed and recognition accuracy are improved.

  Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the drawings. However, the constituent elements described in this embodiment are merely examples, and the technical scope of the present invention is determined by the scope of claims, and is limited by the following individual embodiments. is not.

(First embodiment)
FIG. 1 is a block diagram showing a schematic configuration of the character recognition apparatus according to the first embodiment.

  The CPU 101 controls the entire apparatus by executing a control program stored in the ROM 102. The ROM 102 stores computer programs executed by the CPU 101 and various parameter data. The computer program is executed by the CPU 101 to cause the character recognition device (computer) to function as execution means for executing each process shown in the flowcharts described below. In the present embodiment, processing corresponding to each step of the flowchart to be described later is realized by software using the CPU 101. However, part or all of the processing is realized by hardware such as an electronic circuit. It doesn't matter. The character recognition device of the present invention may be realized using a general-purpose personal computer or may be realized as a device dedicated to character recognition.

  The RAM 103 stores images and various information. The RAM 103 functions as a work area for the CPU and a temporary save area for data.

  The external storage device 104 stores various data such as a dictionary. The external storage device 104 is composed of, for example, a hard disk or a CD-ROM. A computer program for causing a computer to execute the control method of the character recognition device may be stored in a computer-readable external storage medium or may be supplied via a network. The display 105 is composed of, for example, an LCD or CRT.

  The input device 106 may be an interface for connecting an image input device such as a scanner or a digital camera, or may be an image input device itself such as a scanner.

  A network interface (I / F) 107 communicates with external devices connected to the network, for example, a server, an external storage device, an image input device, and the like, and reads and writes programs and data. The network is typically a so-called communication network such as the Internet, a LAN, a WAN, or a telephone line as long as it can transmit and receive data. Further, the display 105 and the input device 106 may be connected via a network interface 108.

  Such a character recognition device is realized, for example, in the system shown in FIG. FIG. 2 is a diagram illustrating a configuration example of a computer system that can be employed in the first embodiment. A computer apparatus 201 receives image data optically read by the scanner 202 and executes processing for character recognition.

  Next, the character recognition process will be described with reference to FIGS. In the present embodiment, an address described in image data optically read by a scanner or the like is recognized.

  FIG. 3 is a diagram illustrating an example of a handwritten address. In the case of a Japanese address, a place name portion 301 and a chome / address portion 302 exist. In FIG. 3, “Shimouma, Setagaya-ku, Tokyo” is the place name part, and the rest is the chome / address part.

  FIG. 4 is a flowchart showing character recognition processing in the character recognition device of the first embodiment. This process is executed under the overall control of the CPU 101.

  In step S401, a plurality of pixel blocks constituting a character image are extracted from the input image data. The row extraction may be performed by taking a projection in the row direction and taking a rectangle in the row direction (row rectangle) so that the portion where the projection exists is the height. Extraction of a pixel block finds a pixel that forms a character from the line rectangle, traces its outline, and returns to the original pixel, and defines it as one pixel block. Similarly, all the pixel blocks in the row rectangle are taken out, and in the horizontal writing mode, the pixel blocks existing above and below are combined. Alternatively, the projection may be taken in the vertical direction, and the pixel block may be extracted so that the portion where the projection exists is the width. Thereafter, the isolated small pixel block of a part of the character is targeted, and the pixel blocks existing close to the isolated small pixel block are combined.

  In step S402, a continuous portion (chome / address portion) of a specific character is detected from the pixel block. Details of step S402 are shown in FIG.

  In step S501 in FIG. 5, a delimiter is detected. The delimiter is a character that exists between numbers, such as “-” (hyphen), “no”, “no”, “chome”, “address”, “no.” And Details of step S501 are shown in FIG.

  In step S601 in FIG. 6, a delimiter candidate is detected based on shape, position information, and the like. For example, in the case of a hyphen, it can be specified by information on the position of the pixel block in the aspect ratio or row height (height of one row). Looking at examples of address entry, many people use hyphens as delimiters, and it is expected that candidates for delimiters can be detected using only the shape and position information of characters, leading to faster processing.

  In step S602, steps S603 and S604 are executed until the delimiter candidate detected in step S601 exists.

  In step S603, it is determined whether both neighbors of the delimiter candidate are numbers. If both sides of the delimiter candidate are numbers, the process proceeds to step S604.

  In step S604, a delimiter candidate is determined as a delimiter.

  In step S605, it is confirmed whether or not a plurality of delimiters are detected in steps S601 to S604. If they are detected, the delimiter detection process is terminated. If not detected, the process proceeds to step S606.

  In step S606, a delimiter candidate is detected using OCR. Separation symbol candidates are detected by repeatedly recognizing one character for a pixel block.

  In step S607, steps S608 and S609 are executed until the delimiter candidate detected in step S606 exists.

  In step S608, it is determined whether both sides of the delimiter candidate are numbers. If both neighbors of the delimiter candidate are numbers, the process proceeds to step S609.

  In step S609, the delimiter candidate is determined as a delimiter.

  When step S609 ends, the process returns to step S502 of FIG.

  In step S502 of FIG. 5, it is determined whether a delimiter is detected. If a delimiter is detected, the process proceeds to step S503. If no delimiter is detected, the chome / address part detection process is terminated.

  In step S503, a chome / address portion is selected. Details of step S503 are shown in FIG.

  In step S701 of FIG. 7, in the case of horizontal writing, the processing of steps S702 to S704 is executed until the boundary of the chome / address portion is determined to the left from the leftmost delimiter.

  In step S702, one character is recognized on the left side of the leftmost delimiter.

  In step S703, it is determined whether the result of single character recognition is a number. If it is a number, the process proceeds to step S704, and if it is not a number, the process proceeds to step S705.

  In step S704, the boundary of the chome / address part is expanded by one character, and the process returns to step S701. Thereafter, in step S702, one character on the left side is further recognized, and the processes in steps S703 and S704 are performed.

  In step S705, in the case of horizontal writing, the processes in steps S706 to S708 are executed until the boundary between the chome / address part is determined to the right from the rightmost delimiter.

  In step S706, one character is recognized right next to the rightmost delimiter.

  In step S707, it is determined whether the result of single character recognition is a number. If it is a number, the process proceeds to step S708, and if it is not a number, the chome / address portion selection process is terminated.

  In step S708, the boundary of the chome / address part is widened by one character, and the process returns to step S705. Thereafter, in step S706, one character on the right is recognized, and the processes in steps S707 and S708 are performed.

  An example in which the chome / address portion is selected in step S503 is shown in FIG.

  Reference numeral 801 denotes an example in which “chome”, “address”, and “issue” are used as delimiters. Since no number exists on the right side of “No.”, “No.” is not designated as the chome / address part. However, there is no problem because it does not affect the radical coupling process performed in step S404.

  Reference numeral 802 is an example in which “no” is used as a delimiter. In this case, the whole is made into a chome / address portion through the processing.

  Reference numeral 803 is an example in which “no” is used as a delimiter. In this case, the whole is made into a chome / address portion through the processing.

  804 is an example in which “chome” and “-” (hyphen) are mixed in the separator. In this case, the whole is made into a chome / address portion through the processing.

  In the example 805, “chome” and “no” are mixed in the separator. In this case, the whole is made into a chome / address portion through the processing.

  806 is an example in which “chome” and “no” are mixed in the separator. In this case, the whole is made into a chome / address portion through the processing.

  When the selection process (step S503) of the chome / address portion in FIG. 5 is completed, the process proceeds to step S403 in FIG.

  In step S403 in FIG. 4, it is determined whether a chome / address portion has been detected. If the detection is confirmed, the process proceeds to step S404, and if the chome / address portion is not detected, the process proceeds to step S405.

  In step S404, pixel blocks other than the chome / address portion detected in step S402 are targeted, and the pixel blocks separated into partials and ridges are combined. Among the plurality of pixel blocks, the pixel blocks satisfying a predetermined threshold in terms of the distance between the adjacent pixel blocks and the aspect ratio are combined. For example, the height of the pixel block is two-thirds or more of the row height, the distance between the pixel blocks is within a certain threshold (distance determination threshold), and the aspect ratio of the combined pixel clusters is constant. If it is within the threshold (aspect ratio determination threshold), the pixel blocks are combined. If the conditions of the bias and the harshness are to be strict, it may be added to the condition that the pixel block is such that the aspect ratio of the pixel block is 0.4 to 0.6 in the horizontal direction. As a result, it is possible to connect the radicals existing in the place name part while preventing the combination of long and narrow characters such as numbers existing in the chome / address part.

  In step S405, a lattice structure is generated by combining the pixel blocks obtained in the steps so far. For example, if the larger one of 1.2 times the estimated character size or 1.2 times the line height is selected and the combined width of the pixel blocks is within a threshold (character size determination threshold), the combination is added to the lattice structure. Try to add.

  In step S406, character recognition processing is performed on each individual character pattern candidate of the lattice structure created in step S405. A character candidate acquisition process corresponding to the combined pixel block is executed by the character recognition process, and an evaluation value corresponding to the character candidate is acquired. As a result of the character recognition process, a plurality of character candidates can be acquired. Character candidates are obtained as, for example, “first candidate” (evaluation value x1), “second candidate” (evaluation value x2),... “Nth candidate” (evaluation value xn), and the evaluation value is maximized. The combination of characters is determined as the optimal path of the lattice structure. It is preferable to use the Viterbi algorithm as the optimum route determination algorithm. Note that the gist of the present invention is not limited to the use of this algorithm, and it is needless to say that any other algorithm that achieves the same function can be similarly applied. Also, it is possible to store the evaluation values of the candidate characters constituting the lattice structure in the external storage device 104.

  FIG. 9 is a diagram illustrating the result of the character recognition processing obtained in the steps so far.

  Reference numeral 901 denotes the state of the pixel block after pixel block extraction. “City” is divided into two parts. In this case, a pixel block is extracted by searching for an outline instead of a projection, and thus a case such as a “valley” is separated.

  Reference numeral 902 indicates the state of the pixel block after detection of the chome / address portion. “0-00-00” is the chome / address section.

  Reference numeral 903 denotes the state of the pixel block after the process of combining the radicals. “City” is in a combined state. In this case, since the circumscribed rectangles of the character blocks constituting the “valley” overlap, they are excluded from the radical combination target.

  Reference numeral 904 denotes a state of a character string after generating a lattice structure and determining an optimum route. Here, it can be seen that the characters “valley” are correctly detected.

  As described above, according to the present embodiment, character recognition with excellent recognition speed and recognition accuracy is possible when characters are automatically read from an image.

Characters by reducing the combination pattern (number of routes) for character candidate determination when determining the optimal route by combining the radicals existing in the place name portion while preventing the combination of numbers existing in the chome / address portion The recognition process can be speeded up. In addition, a reduction in the false recognition rate is expected by reducing the number of routes.
(Other embodiments)
Note that it is needless to say that the object of the present invention can also be achieved by supplying a computer-readable storage medium that records a software program for realizing the functions of the above-described embodiments to a system or apparatus. Needless to say, this can also be achieved by the computer (or CPU or MPU) of the system or apparatus reading and executing the program stored in the storage medium.

  In this case, the program itself read from the storage medium realizes the functions of the above-described embodiments, and the storage medium storing the program constitutes the present invention.

  As a storage medium for supplying the program, for example, a flexible disk, a hard disk, an optical disk, a magneto-optical disk, a CD-ROM, a CD-R, a nonvolatile memory card, a ROM, or the like can be used.

  Further, the functions of the above-described embodiments are realized by executing the program read by the computer. In addition, it is also included that an OS (operating system) or the like running on a computer performs part or all of actual processing based on a program instruction, and the above-described embodiment is realized by the processing. Needless to say.

It is a block diagram which shows schematic structure of the character recognition apparatus which concerns on embodiment of this invention. It is a system configuration figure showing a schematic structure of a character recognition device concerning an embodiment of the present invention. It is a figure which shows an example of a handwritten address. It is a flowchart explaining the flow of the character recognition process which concerns on embodiment of this invention. It is a flowchart explaining the flow of the detection process of the chome / address part which concerns on embodiment of this invention. It is a flowchart explaining the flow of the separator detection process which concerns on embodiment of this invention. It is a flowchart explaining the flow of the selection process of the chome / address part which concerns on embodiment of this invention. It is a figure which shows the example which selected the chome and the address part. It is a figure which illustrates the result of a character recognition process.

Explanation of symbols

101 CPU
102 ROM
103 RAM
104 External storage device 105 Display 106 Input device

Claims (6)

Pixel block extraction means for extracting a plurality of pixel blocks constituting a character image from input image data;
Detecting means for detecting a continuous portion of a specific character from the plurality of pixel blocks;
In a pixel block excluding a portion where the specific character detected by the detection means is continuous among the plurality of pixel blocks, a distance between adjacent pixel blocks is within a predetermined distance determination threshold, and the adjacent When the aspect ratio of the pixel blocks to be combined is within a predetermined aspect ratio determination threshold , combining means for combining the adjacent pixel blocks,
Character candidate corresponding to the combined pixel block by character recognition processing, and character candidate acquisition means for acquiring an evaluation value corresponding to the character candidate;
Determining means for determining a character candidate that maximizes the evaluation value in the entire character image ;
Wherein the specific character that is detected by the detection means, the character recognition device for a separator, characterized Rukoto includes separating and numbers between the numbers and figures. Said detecting means, and aspect ratio of the pixel block, based on the information of the position of the pixel block is located, the character recognition apparatus according to claim 1, characterized in that for detecting the candidates of the separator. The character recognition device according to claim 1 , wherein the detection unit determines, as the delimiter, a delimiter candidate that has a numeral on both sides of the delimiter. A character recognition method in character recognition device,
A pixel block extraction means for extracting a plurality of pixel blocks constituting a character image from the input image data;
A detection step in which the detection means of the character recognition device detects a continuous portion of a specific character from the plurality of pixel blocks;
In the pixel block excluding a portion where the specific character detected in the detection step is continuous among the plurality of pixel blocks, the combining unit of the character recognition device determines a distance between adjacent pixel blocks in advance. When the aspect ratio is within a predetermined aspect ratio determination threshold within the threshold value and the adjacent pixel blocks are combined, a combining step of combining the adjacent pixel blocks;
Character candidate acquisition means of the character recognition device, a character candidate acquisition step of acquiring an evaluation value corresponding to the character candidate corresponding to the combined pixel block and the character candidate by character recognition processing;
The determination unit of the character recognition apparatus, the whole of the character image, have a, a determination step of determining a character candidate which the evaluation value is maximized,
Wherein the specific character, the character recognition process for a number, a delimiter to separate between the numbers and numbers, wherein Rukoto contains detected by said detecting step. The computer program for functioning a computer as each means of the character recognition apparatus of any one of Claims 1 thru | or 3 . A computer-readable storage medium storing the computer program according to claim 5 .

Images