JP5964078B2 - Character recognition device, character recognition method and program - Google Patents

Description

  The present invention relates to a character recognition device, a character recognition method, and a program.

  A visually impaired person obtains information from surroundings mainly by information such as Braille or voice. However, text information including important information in daily life, such as product containers and stickers, often lacks such information.

  As a system for creating speech information from character information that does not include Braille or speech information, there are devices described in Patent Literature 1 and Non-Patent Literature 1, for example. The system described in Patent Document 1 cuts out an area surrounded by a color having a certain characteristic from an image used as input information to the apparatus, and extracts characters existing in the area. The extracted character is converted into a character code and read out by the speech synthesizer.

  The apparatus described in Non-Patent Document 1 traces characters on an input image with a fingertip, recognizes characters in the traced portion, and reads them out.

JP-A-4-24885

Masataka Hamada, Akio Nakamura, “Examination of fingertip designation character extraction”, 17th Image Sensing Symposium, Yokohama, June 2011, IS3-08-1 to IS3-08-4

  However, in Patent Document 1, characters can not be extracted unless the material is pre-marked for the device, and such a mark can be pre-marked like a generally distributed product or a sticker. Character information cannot be recognized if it is difficult.

  In the character recognition device described in Non-Patent Document 1, an area connected by performing dilation processing on the entire binarized original image is recognized as a character area. However, if dilation processing is performed on the entire original image, noise in the background area that is not a character increases, so that it may be mistakenly recognized as a character area and accuracy may be reduced.

  In view of the above problems, an object of the present invention is to provide a general-purpose character recognition device, a character recognition method, and a program for recognizing a character traced with a fingertip with higher accuracy.

According to the present invention,
Character candidate recognition means for recognizing a character candidate area that is an area estimated as one character from the binarized image;
For each of the character candidate areas recognized by the character candidate recognition means, an expansion process for expanding an area having a color estimated to be a character is performed, and an area connected between the character candidate areas subjected to the expansion process is a character string. A character string area recognition means for recognizing as an area;
There is provided a character recognition device having character string extraction means for extracting a character string from the image based on the character string region.

According to the present invention,
Computer
Recognizing a character candidate area, which is an area estimated as one character, from a binarized image;
For each of the character candidate areas, an expansion process is performed to expand an area having a color estimated to be a character, and an area connected between the character candidate areas subjected to the expansion process is recognized as a character string area.
A character recognition method for extracting a character string from the image based on the character string region is provided.

According to the present invention,
Computer
Means for recognizing a character candidate area which is an area presumed to be one character from the binarized image;
Means for expanding an area having a color estimated to be a character for each character candidate area, and recognizing an area connected between the character candidate areas subjected to the expansion process as a character string area;
A program for causing a character string to be extracted from the image based on the character string area is provided.

  ADVANTAGE OF THE INVENTION According to this invention, the general purpose character recognition apparatus, the character recognition method, and program which can recognize the character of the part traced with the fingertip with higher precision are provided.

It is a block diagram which shows the structure of the character recognition apparatus which concerns on the 1st Embodiment of this invention. It is a flowchart which shows the flow of a process of the character recognition apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the example of the rectangle produced | generated. It is a figure which shows the example of an arrangement | positioning of a rectangle. It is a figure explaining the expansion method of a rectangle. It is a figure explaining the flow which recognizes a character candidate area | region. It is a figure which shows the example of the information which a character candidate recognition part transmits to a character string area | region part. It is a figure which shows the example of the character string area | region which a character string area | region recognition part transmits. It is a figure which shows the example of the character string which a character string extraction part extracts. It is a figure which shows the flow which determines the color estimated as a character.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In all the drawings, the same reference numerals are given to the same components, and the description will be omitted as appropriate.

(First embodiment)
FIG. 1 is a block diagram showing a configuration of a character recognition apparatus 10 according to the first embodiment of the present invention. The character recognition device 10 includes a character candidate recognition unit 102, a character string region recognition unit 104, and a character string extraction unit 106.

  The character candidate recognition unit 102 recognizes a character candidate area that is an area estimated to be one character.

  The character string area recognition unit 104 performs an expansion process for expanding an area having a color estimated to be a character, for each character candidate area recognized by the character candidate recognition unit 102, and the character candidate areas subjected to the expansion process are connected to each other. This area is recognized as the character string area 112.

  The character string extraction unit 106 extracts characters from the image based on the character string region 112.

  Note that each component of the character recognition device 10 shown in each figure is not a hardware unit configuration but a functional unit block. Each component of the character recognition apparatus 10 is centered on an arbitrary computer CPU, memory, a program for realizing the components shown in the figure loaded in the memory, a storage medium such as a hard disk for storing the program, and a network connection interface. It is realized by any combination of hardware and software. There are various modifications of the implementation method and apparatus.

  The processing flow in this embodiment will be described with reference to FIGS.

  First, the character candidate recognition unit 102 acquires a binarized image (hereinafter referred to as a binarized image) (S102). Then, the character recognition unit 102 performs a character candidate area recognition process from the binarized image (S104).

  First, as shown in FIG. 3, the character candidate recognition unit 102 generates a rectangle 110 for specifying a character candidate region on the binarized image based on the index 108 in the image. Here, the relationship between the rectangle 110 and the characters on the image is as shown in FIGS. 4B and 4C, for example, when the character candidate area is within the rectangle 110 as shown in FIG. Thus, the case where the character candidate area does not fit within the rectangle 110 can be considered. In the case of FIG. 4A, the character candidate recognition unit 102 may use the rectangle 110 as the character candidate area as it is, but in the case of FIGS. 4B and 4C, as shown in FIG. The rectangle 110 is enlarged to specify a character candidate area. The flow of processing for enlarging the rectangle 110 will be described below.

  As shown in FIG. 5, the character candidate recognition unit 102 scans the outer periphery of the rectangle 110 and confirms whether or not the pixel in contact with the outer periphery of the rectangle 110 has a color estimated as a character. The color estimated as a character can be determined by storing information indicating which color of the binarized image is estimated as a character in a storage unit (not shown). When there is a pixel of a color estimated to be a character, the character candidate recognition unit 102 enlarges the rectangle 110 by a certain amount. Here, the character candidate recognition unit 102 enlarges the width in the horizontal direction when scanning the vertical direction of the outer periphery, and expands the width in the vertical direction when scanning the horizontal direction of the outer periphery. For example, at point A in FIG. 5, the character candidate recognition unit 102 enlarges the horizontal width of the rectangle 110 by a certain amount in the left direction. The character candidate recognition unit 102 repeats this process until pixels of a color estimated to be a character do not exist on the outer periphery of the rectangle 110. Then, the area within the enlarged rectangle 110 ′ is recognized as a character candidate area. Note that there is a restriction on the total amount of enlargement of the rectangle 110, and the character candidate recognition unit 102 performs control so that the rectangle 110 is not further enlarged when the amount of enlargement reaches the restricted amount. Thus, control is performed so that the rectangle 110 is not enlarged without limitation. Further, the character candidate recognition unit 102 performs control so that the rectangle 110 is not expanded to a region below the coordinate value indicated by the index 108 in order to prevent the index 108 from being included in the rectangle.

  The character candidate recognizing unit 102 identifies the plurality of character candidate regions by repeatedly executing the above-described processing based on the coordinates of the position through which the index 108 is stored, which is stored in a storage unit (not shown). Finally, the character candidate recognition unit 102 specifies a plurality of character candidate regions as shown in FIG. 6, for example. The character candidate recognition unit 102 transmits all the specified character candidate regions to the character string region recognition unit 104. FIG. 7 is a diagram illustrating an example of information transmitted to the character string region recognition unit 104.

  The character string region recognition unit 104 recognizes the character string region 112 based on the information received from the character candidate recognition unit 102 (S106). First, the character string region recognizing unit 104 performs an expansion process for expanding a color region estimated to be a character in each recognized character candidate region. Then, the character string region recognition unit 104 identifies the character string region 112 by removing noise from among several connected regions formed by the character candidate regions. The character string area recognition unit 104 determines whether the connected area is noise based on the shape of the character candidate area corresponding to the connected area. The character string region recognizing unit 104 determines the vertical and horizontal symmetry of the character candidate region corresponding to the connection region, the binarized color density ratio, the complexity and size of the shape formed by the binarized color, etc. Based on the information, the noise is determined. In this manner, as shown in FIG. 8, for example, noise is removed from each character candidate region connected by the expansion process from the connected region, and the character string region 112 is recognized. The character string area recognition unit 104 sends, for example, the coordinate value of the character string area 112 or the image (mask information) shown in FIG. 8 subjected to the expansion process to the character string extraction unit 106 as information for extracting the character string. Send.

  If the original binarized image is tilted, the character candidate regions may be connected in a tilted state. Therefore, the character string region recognizing unit 104 determines whether to connect the connected regions before removing the noise. The tilt angle may be obtained by calculating the inertial main axis, and the tilt may be corrected by affine transformation or the like. By correcting the inclination, the accuracy with which the character string area recognition unit 104 discriminates the character string from the noise is improved.

  The character string extraction unit 106 extracts a character string from the binarized image based on the character string region 112 received from the character string region recognition unit 104 (S108). For example, when the coordinate value of the character string area 112 is received, the character string extraction unit 106 extracts an area corresponding to the coordinate value as a character string. Alternatively, when mask information is received, the corresponding area is extracted as a character string by the logical product of the binarized image and the mask information. FIG. 9 is a diagram illustrating an example of a character string extracted by the character string extraction unit 106.

  As described above, in the present embodiment, the character candidate area estimated as one character is recognized based on the coordinate value passed by the index, the character candidate area is expanded, and the character string area 112 is selected from the connected areas. Take the configuration to recognize. With this configuration, it is possible to accurately extract a character string while suppressing background noise.

  In this embodiment, for example, the fingertip itself can be used as an index, or a material that reflects light of a specific wavelength can be attached to the fingertip, and the material can be used as an index.

(Second Embodiment)
This embodiment has a configuration in which the character information recognized in the first embodiment is read out as speech.

  In the present embodiment, the character recognition device 10 of the first embodiment further includes an image input unit and a voice synthesis unit (not shown). The image input unit acquires a color image from an input device such as a camera. The speech synthesis unit uses the character candidate recognition unit 102, the character string region recognition unit 104, and the character string extraction unit 106 to convert the character string extracted from the surface of the product package into text and read it out as speech.

  The image input unit acquires a color image to be processed as an original image from, for example, a small camera. Then, the image input unit converts the original image into a binary image composed of “a color estimated as a character” and “a color other than a color estimated as a character”. FIG. 10 is a diagram illustrating a flow of determining a color estimated as a character. The image input unit generates a histogram of an image obtained by performing a smoothing process on the original image, and determines a histogram separation position on the basis of a maximum value, a minimum value, or the like. Then, a color space based on the cut position is created. The image input unit generates a rectangle 110 based on the coordinate value of the index 108 and projects the color existing in the rectangle 110 onto the previous color space. Then, the image input unit sets the color corresponding to the square with the largest number of votes in the color space as a candidate for “color estimated to be a character”. Then, the image input unit binarizes the candidate “color estimated to be a character”, and the density of “color estimated to be a character” in the rectangle 110 or the color estimated to be a character. The “color estimated to be a character” of the rectangle 110 is determined using the number of regions as a criterion. When the color with the largest number of votes does not satisfy the judgment criterion, the image input unit performs the same judgment with the color corresponding to the square with the next largest number of votes. Then, the image input unit creates an image binarized with “a color estimated as a character” and “a color other than a color estimated as a character”, and transmits the image to the character candidate recognition unit 102.

  Since the process flow from when the character candidate recognition unit 102 receives the binarized image to when the character candidate is extracted is the same as that in the first embodiment, the description is omitted. The character extraction unit 106 transmits the extracted character string to the speech synthesis unit.

  The voice synthesis unit converts the character string received from the character extraction unit 106 into text using an OCR (Optical Character Reader) or the like, and reads it out as voice information using existing reading software or the like.

  As described above, the present embodiment further includes the image input unit and the voice synthesis unit. With this configuration, the character information in the image acquired by the image input unit can be recognized as in the first embodiment, and can be provided as voice information by the voice synthesis unit. With this configuration, it is possible to provide a device that supports the life of the visually impaired, for example, reading a product name on a product package.

  As mentioned above, although embodiment of this invention was described with reference to drawings, these are the illustrations of this invention, Various structures other than the above are also employable.

DESCRIPTION OF SYMBOLS 10 Character recognition apparatus 102 Character candidate recognition part 104 Character string area | region recognition part 106 Character string extraction part 108 Index 110,110 'rectangle 112 Character string area | region

Claims (12)

Character candidate recognition means for detecting a position of an index in the binarized image and recognizing a character candidate area that is an area estimated as one character based on the position of the index ;
For each of the character candidate areas recognized by the character candidate recognition means, an expansion process for expanding an area having a color estimated to be a character is performed, and an area connected between the character candidate areas subjected to the expansion process is a character string. A character string area recognition means for recognizing as an area;
A character recognition device comprising character string extraction means for extracting a character string from the image based on the character string region. The character recognition device according to claim 1,
The character string extracting means includes
A character that holds a two-dimensional coordinate value in the entire image, compares the two-dimensional coordinate value in the entire image with the two-dimensional coordinate value in the character string region, and extracts the character string from the image Recognition device. The character recognition device according to claim 1,
The character string extracting means includes
A character recognition apparatus that separately holds an image subjected to the expansion processing as mask information, and extracts the character string from a logical product of the image and the mask information. In the character recognition device according to any one of claims 1 to 3,
The character candidate recognition means includes:
It generates a reference area based on the position of the indicator which moves within the image, recognizing a character recognition apparatus the character candidate region from the image by using the reference region described above generated. Computer
Detecting the position of the index in the binarized image , and recognizing a character candidate area which is an area estimated as one character based on the position of the index ;
For each of the character candidate areas, an expansion process is performed to expand an area having a color estimated to be a character, and an area connected between the character candidate areas subjected to the expansion process is recognized as a character string area.
A character recognition method for extracting a character string from the image based on the character string region. The character recognition method according to claim 5,
The computer is
A character that holds a two-dimensional coordinate value in the entire image, compares the two-dimensional coordinate value in the entire image with the two-dimensional coordinate value in the character string region, and extracts the character string from the image Recognition method. The character recognition method according to claim 5,
The computer is
A character recognition method for separately holding an image subjected to the expansion process as mask information and extracting the character string from a logical product of the image and the mask information. In the character recognition method according to any one of claims 5 to 7,
The computer is
Wherein generating the reference region based on the position of the indicator, a character recognition method for recognizing the character candidate region from the image by using the reference area the generated moving within the image. Computer
Means for detecting a position of an index in the binarized image and recognizing a character candidate area which is an area estimated as one character based on the position of the index ;
Means for expanding an area having a color estimated to be a character for each character candidate area, and recognizing an area connected between the character candidate areas subjected to the expansion process as a character string area;
A program for functioning as means for extracting a character string from the image based on the character string region. The program according to claim 9,
The computer,
Means for holding a two-dimensional coordinate value in the entire image;
A program for further functioning as means for extracting the character string from the image by comparing the two-dimensional coordinate value of the entire image with the two-dimensional coordinate value of the character string region. The program according to claim 9,
The computer,
Means for separately holding the image subjected to the expansion processing as mask information;
A program for further functioning as means for extracting the character string from the logical product of the image and the mask information. The program according to any one of claims 9 to 11, wherein the computer is
Wherein based on the position of the index to generate a reference area, a program for further functions as a means for recognizing the character candidate region from the image by using the reference area the generated moving within the image.

Images