JP2012252691A - Method and device for extracting text stroke image from image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and a device for extracting a text stroke image from an image.SOLUTION: According to one aspect of the invention, a method for extracting a text stroke image from an image includes the steps of: obtaining edge information and gradient information on the image; performing predetermined enhancement processing for the obtained edge information and gradient information to thereby enhance the edge information and gradient information relating to texts in the image; and obtaining a text stroke image corresponding to the enhanced edge information and gradient information.

Description

  The present invention relates to the field of image processing, and more particularly to a method and apparatus for extracting a text stroke image from an image.

  There are a large number of video files in the current information processing field. It is necessary to perform an efficient search on these video files. For video annotation, video search, etc., text information in the video is an accurate and simple clue. Therefore, how to accurately extract and recognize text information included in a video is very important for subsequent video annotation and video search.

Some known text stroke extraction techniques have drawbacks such as slow speed, noisyness, and insensitivity to stroke scale.
A method and apparatus for extracting a text stroke image from an image that can solve the above-described problems is desired.

The following briefly describes the present invention and provides a basic understanding of some aspects of the present invention. This brief description is not exhaustive for the invention. Its purpose is not to determine the essential or critical part of the invention, nor to limit the scope of the invention, but to provide some concepts in a simplified form and to precede the more detailed description below. It's just an explanation.
An object of this invention is to provide the method and apparatus which extract a text stroke image from an image.

  According to one aspect of the present invention, there is provided a method for extracting a text stroke image from an image, acquiring edge information and gradient information of the image, and providing the acquired edge information and gradient information in advance. By performing the enhanced enhancement, a method is provided that includes emphasizing edge information and gradient information relating to text in the image and obtaining a text stroke image corresponding to the enhanced edge information and gradient information. .

  According to another aspect of the present invention, an apparatus for extracting a text stroke image from an image is provided in advance in information acquisition means for acquiring edge information and gradient information of the image, and in the acquired edge information and gradient information. An apparatus including an enhancement unit that enhances edge information and gradient information related to text in an image by performing the enhancement process, and a stroke image acquisition unit that acquires a text stroke image corresponding to the enhanced edge information and gradient information. Provided.

  In the embodiment of the present invention, a computer program for realizing the above method is further provided.

  In the embodiment of the present invention, computer program code for realizing the above method is recorded, and at least a computer program product in the form of a computer-readable medium is further provided.

  The above and other advantages of the present invention will become more apparent from the following detailed description of the preferred embodiment of the present invention with reference to the drawings.

The above objects, features and advantages of the present invention, as well as other objects, features and advantages can be more easily understood with reference to the description of the embodiments of the present invention based on the drawings. The components in the drawings are merely illustrative of the principles of the invention. In the drawings, the same or similar technical features or components are denoted by the same or similar reference numerals.
It is a flowchart which shows the method of extracting the text stroke image from the image based on embodiment of this invention. FIG. 2 shows the step signal, the pulse signal, the signal corresponding to the Sobel operator, the extraction result obtained by extracting the step signal by the Sobel operator, the extraction result obtained by extracting the pulse signal by the Sobel operator, and the absolute value of the extraction result of the step signal. It is a schematic diagram which shows the signal which took the signal which took the absolute value in the extraction signal of the pulse signal, the extraction result of the pulse signal, and the corresponding offset. 4 is a flowchart illustrating a method for extracting a text stroke image from an image using a Sobel operator according to the present invention. It is a figure which shows the primitive image processed. It is a figure which shows the image which performed the convolution process using the Sobel operator. It is a figure which shows the image which performed the convolution process using the Sobel operator. It is a figure which shows the image which performed the convolution process using the Sobel operator. It is a figure which shows the image which performed the convolution process using the Sobel operator. It is a figure which shows the image which performed the convolution process using the Sobel operator. It is a figure which shows the image which performed the convolution process using the Sobel operator. It is a figure which shows the image which performed the convolution process using the Sobel operator. It is a figure which shows the image which performed the convolution process using the Sobel operator. It is a figure which shows four images obtained by offsetting the image processed with the Sobel operator in the opposing direction, and synthesize | combining the offset image. It is a figure which shows four images obtained by offsetting the image processed with the Sobel operator in the opposing direction, and synthesize | combining the offset image. It is a figure which shows four images obtained by offsetting the image processed with the Sobel operator in the opposing direction, and synthesize | combining the offset image. It is a figure which shows four images obtained by offsetting the image processed with the Sobel operator in the opposing direction, and synthesize | combining the offset image. It is a figure which shows four images obtained by offsetting the image processed with the Sobel operator in the opposite direction, and synthesize | combining the offset image. It is a figure which shows four images obtained by offsetting the image processed with the Sobel operator in the opposite direction, and synthesize | combining the offset image. It is a figure which shows four images obtained by offsetting the image processed with the Sobel operator in the opposite direction, and synthesize | combining the offset image. It is a figure which shows four images obtained by offsetting the image processed with the Sobel operator in the opposite direction, and synthesize | combining the offset image. It is a figure which shows the matched thin stroke image. It is a figure which shows the matched thick stroke image. It is a figure which shows the thick stroke image obtained by the filtering process. It is a block diagram which shows the apparatus which extracts the text stroke image from the image based on embodiment of this invention. It is a block diagram which shows the apparatus which extracts a text stroke image from an image with a Sobel operator based on embodiment of this invention. 1 is a configuration diagram illustrating, as an example, a computer apparatus that can implement a method and apparatus for extracting a text stroke image from an image according to the present invention.

  Embodiments of the present invention will be described below with reference to the drawings. Elements and features described in one drawing or one type of embodiment of the invention may be combined with elements and features shown in one or more other drawings or embodiments. Here, for the sake of clarity, in the drawings and description, the display and description of components and processes that are known to those skilled in the art and are not relevant to the present invention are omitted.

A method for extracting a text stroke image from an image according to an embodiment of the present invention will be described below with reference to FIG.
FIG. 1 is a flowchart showing a method for extracting a text stroke image from an image according to an embodiment of the present invention. As shown in FIG. 1, in step S102, edge information and gradient information of the image can be acquired.
Preferably, a step signal or a pulse signal representing edge information and gradient information of an image can be analyzed, and edge information and gradient information can be extracted based on the analysis result.

  In an image, image data of a thin stroke can be indicated by a pulse signal, and image data of a thick stroke and a large scale object similar thereto can be indicated by a step signal. By analyzing the pulse signal, it is possible to extract edge information and gradient information of a thin stroke based on the analysis result. Also, it is possible to analyze the step signal, extract edge information and gradient information of a thick stroke based on the analysis result, and extract edge information and gradient information of a large scale object similar to a thick stroke.

  Hereinafter, with reference to FIG. 2, a process of extracting a step signal by the Sobel operator, a process of extracting a pulse signal by the Sobel operator, and a process of executing enhancement processing by offset processing and synthesis processing will be described.

  2 (i) is a step signal, FIG. 2 (ii) is a pulse signal, FIGS. 2 (iii) and 2 (iv) are signals corresponding to the Sobel operator, and FIG. 2 (v) is a step signal using the Sobel operator. 2 (vi) shows the extraction result of extracting the pulse signal with the Sobel operator, FIG. 2 (vii) shows the signal obtained by taking the absolute value in the extraction result of the step signal, and FIG. 2 (viii) shows the pulse. The signal obtained as an absolute value in the signal extraction result, and FIGS. 2 (ix) and 2 (x) show the corresponding offset signals.

  Here, the coordinates and sizes in FIG. 2 are illustrative rather than limiting, and are only intended to illustrate the principles of related signal processing.

  As shown in FIG. 2, by extracting the step signal (as shown in FIG. 2 (i)) with a Sobel operator (as shown in FIG. 2 (iii)) (as shown in FIG. 2 (v) A single trough signal (as shown) can be obtained. On the other hand, by extracting a pulse signal (as shown in FIG. 2 (ii)) with a Sobel operator (as shown in FIG. 2 (iv)), as shown in FIG. 2 (vi). ) A signal obtained by combining a valley-like signal and a mountain-like signal can be obtained. As described above, by extracting the step signal and the pulse signal with the Sobel operator, two signals having a large difference (as shown in FIG. 2 (v) and FIG. 2 (vi)) are acquired correspondingly. be able to.

  Next, the step signal and the pulse signal can be enhanced by enhancement processing (for example, including offset processing and synthesis processing) described later. As a result, image data of a thin stroke corresponding to the pulse signal and image data of a thick stroke corresponding to the step signal are extracted.

  For example, in the pulse signal extracted by the Sobel operator, the absolute value of the valley signal (as shown in FIG. 2 (vi)) is taken (that is, the magnitude of the valley signal is taken). Such a signal (as shown in FIG. 2 (viii)) can be obtained. After that, the other peak-like signal and the original peak-like signal part (that is, the peak-like signal extracted by the Sobel operator) are offset in the opposite direction, and By superimposing a signal like a peak and a signal like the original peak (as shown in FIG. 2 (x)), the image data of a thin stroke corresponding to the pulse signal can be further enhanced. . Also, taking the absolute value of the valley signal in the step signal extracted by the Sobel operator (as shown in Fig. 2 (v)), another one (as shown in Fig. 2 (vii)) A mountain-like signal can be obtained. Thereafter, the other peak-like signal (as shown in FIG. 2 (ix)) is offset.

  Optionally, the edge information and gradient information of a large scale object similar to a thick stroke can be filtered based on a predetermined filtering condition, and the edge information and gradient information of a thick stroke can be left. Specific filtering processing will be described later.

  Optionally, the source image shown in FIG. 4 may be filtered based on the clarity of the image, or as needed. For example, noise in the image can be suppressed by filtering the original image using a low-pass filter. The low-pass filter may be, for example, a Gaussian filter, but is not limited thereto, and may be any appropriate low-pass filter known to those skilled in the art.

  In step S104, the edge information and the gradient information related to the text can be emphasized in the image by performing the enhancement processing provided in advance for the acquired edge processing and gradient processing.

  The edge information and the gradient information can be enhanced by various methods. Preferably, edge information and gradient information can be emphasized by binarization processing and matching processing. The matching process here may be a process for obtaining a product set, a process for obtaining a maximum value, or a process for obtaining an average value. Preferably, as will be described later, the matching process may be a process for obtaining a product set.

  In step S106, a text stroke image corresponding to the emphasized edge information and gradient information can be obtained.

  Hereinafter, a method for extracting a text stroke image from an image by a Sobel operator according to an embodiment of the present invention will be described with reference to FIGS.

  FIG. 3 is a flowchart showing a method for extracting a text stroke image from an image by a Sobel operator according to an embodiment of the present invention. FIG. 4 is a diagram showing a source image to be processed. FIG. 5A to FIG. 5H are diagrams illustrating images that have been subjected to convolution processing by the Sobel operator. 6A to 6D are diagrams illustrating four images obtained by offsetting images in the facing direction and combining the offset images. 7A to 7D are diagrams illustrating four images obtained by offsetting images in opposite directions and combining the offset images. FIG. 8 is a diagram showing a thin stroke image that is aligned. FIG. 9 is a diagram illustrating a thick stroke image that has been matched. FIG. 10 is a diagram illustrating a thick stroke image subjected to filtering processing.

  As shown in FIG. 3, in step S202, a smoothing process (eg, a low-pass filter) can be performed on an image to be processed (eg, the original image shown in FIG. 4). Step S202 is selectable. In other words, if the image is clear for comparison, or if necessary, smoothing processing (such as a low-pass filter) may not be performed on the image shown in FIG.

  Here, let us say that the original image shown in FIG. 4 is an image to be processed. In FIG. 4, it is desirable that the extracted text stroke includes FUJITSU in the upper right corner of the image and Japanese characters on the lower side of the image. Note that FUJITSU is a character with thin strokes, and Japanese characters are characters with thick strokes. Here, FIG. 4 is merely exemplary. Actually, there may be a case where only thin stroke characters or only thick stroke characters are included in the image.

  In step S204, convolution processing can be performed on an image smoothed by a Sobel operator. When the smoothing process of step S202 is not performed, the convolution process can be performed on the image to be processed (for example, the original image shown in FIG. 4) using the Sobel operator directly.

  Specifically, convolution processing is performed on the image to be processed in a plurality of directions using a Sobel operator. In other words, a convolution operation of a plurality of Sobel operators for acquiring edge information and gradient information in a plurality of directions of the image is used to perform convolution operations on the image data of the image, respectively.

For example, using a Sobel operator, the image to be processed is convolved in four directions. Preferably, these four directions include a horizontal direction, a vertical direction, and two diagonal directions. Images that have undergone the convolution process are shown in FIGS. The reason for selecting these four directions is that each of these four directions is a regular stroke (outside 1)
It is because it corresponds to.

For example, the convolution kernel of the Sobel operator used is as follows:

S _h is the convolution kernel of the Sobel operator in the horizontal direction, S _v is the convolution kernel of the Sobel operator in the vertical direction, S _rd is the convolution kernel of the Sobel operator in the first diagonal direction, S _ld is a convolution kernel of the Sobel operator related to the second diagonal direction. The first diagonal direction is the diagonal direction from the upper right corner to the lower left corner, and the second diagonal direction is the diagonal direction from the lower left corner to the lower right corner.

  Here, since the result of the convolution process includes both positive and negative, the convolution result can be stored using two layers of images. One is used to store a positive pulse response image and the other is used to store a negative pulse response image. In other words, the result of the convolution operation for each direction is divided into positive pulse response image data and negative pulse response image data for each direction.

  Although examples of four directions have been disclosed herein, the present invention is not limited to performing a convolution operation in four directions. When high accuracy is required, or as required, convolution processing can be performed on an image to be processed using a convolution kernel of Sobel operators in more directions, for example, eight directions. A more accurate result can be obtained by more convolution operations. The definition of the Sobel operator and the convolution process with the image data may be by any suitable conventional technique.

  By performing the convolution process on the original image shown in FIG. 4, it is possible to obtain an image obtained by performing the convolution process with the convolution kernel of the Sobel operator shown in FIGS. 5A to 5H.

  Specifically, FIG. 5A shows an image including the positive pulse response image data in the horizontal direction, and FIG. 5B shows an image including the negative pulse response image data in the horizontal direction. FIG. 5C shows an image including positive pulse response image data in the vertical direction, and FIG. 5D shows an image including negative pulse response image data in the vertical direction. FIG. 5E shows an image including positive pulse response image data in the diagonal direction from the upper right corner to the lower left corner, and FIG. 5F shows an image including negative pulse response image data in the diagonal direction from the upper right corner to the lower left corner. FIG. 5G shows an image including positive pulse response image data in the diagonal direction from the upper left corner to the lower right corner, and FIG. 5H shows an image including negative pulse response image data in the diagonal direction from the upper left corner to the lower right corner.

  Next, as shown in FIGS. 6A to 6D and FIGS. 7A to 7D, in step S206, the pre-estimated stroke width is used to offset the positive pulse response image data and the negative pulse response image data. Composite images.

  First positive composite image data for each direction is obtained by performing an operation of offsetting and adding the positive pulse response image data and negative pulse response image data for each direction in the opposite direction. Can do. In addition, by performing an operation of offsetting and adding the positive pulse response image data and the negative pulse response image data for each direction in the opposite directions, the second composite image data for each direction is obtained. Obtainable.

For example, the operation of offsetting in the opposite direction and adding may be performed based on the following equation (1).
I _h (x, y) = (I _h-positive (x, yw / 2) + I _h-negative (x, y + w / 2)) / 2
I _v (x, y) = (I _v-positive (xw / 2, y) + I _v-negative (x + w / 2, y)) / 2
I _rd (x, y) = (I _rd-positive (x + w / 2, yw / 2) + I _rd-negative (xw / 2, y + w / 2)) / 2
I _ld (x, y) = (I _ld-positive (xw / 2, yw / 2) + I _ld-negative (x + w / 2, y + w / 2)) / 2
Further, the operation of offsetting in the opposite direction and adding may be executed based on the following equation (2).

I _h '(x, y) = (I _h-positive (x, y + w / 2) + I _h-negative (x, yw / 2)) / 2,
I _v '(x, y) = (I _v-positive (x + w / 2, y) + I _v-negative (xw / 2, y)) / 2,
I _rd '(x, y) = (I _rd-positive (xw / 2, y + w / 2) + I _rd-negative (x + w / 2, yw / 2)) / 2,
I _ld '(x, y) = (I _ld-positive (x + w / 2, y + w / 2) + I _ld-negative (xw / 2, yw / 2)) / 2
X is the abscissa of the pixel, y is the ordinate of the pixel, I _h (x, y), I _v (x, y), I _rd (x, y), and I _ld (x, y) are the pixel First composite image data for the four directions of the horizontal direction, the vertical direction, and two diagonal directions are respectively shown. In addition, the image containing 1st synthetic | combination image data is shown by FIG.

6A is an image including data I _h (x, y), FIG. 6B is an image including data I _v (x, y), FIG. 6C is an image including data I _rd (x, y), and FIG. 6D. Show images including data I _ld (x, y), respectively.

I _h ′ (x, y), I _v ′ (x, y) and I _rd ′ (x, y), I _ld ′ (x, y) are the horizontal, vertical and two diagonal directions of the pixel. Second composite image data for four directions are respectively shown. Note that images including the second composite image data are shown in FIGS.

7A is an image including data I _h ′ (x, y), FIG. 7B is an image including data I _v ′ (x, y), and FIG. 7C is an image including data I _rd ′ (x, y). FIG. 7D shows images each including data I _ld ′ (x, y).

I _h-positive (x, y), I _v-positive (x, y), I _rd-positive (x, y) and I _ld-positive (x, y) are the horizontal direction, vertical direction and two Positive pulse response image data for four directions in one diagonal direction are shown. The images including the positive pulse response image data are shown in FIGS. 5A, 5C, 5E, and 5G.

I _h-negative (x, y), I _v-negative (x, y), I _rd-negative (x, y), and I _ld-negative (x, y) are the horizontal, vertical, and two Negative pulse response image data for four directions in one diagonal direction are respectively shown. Note that images including negative pulse response image data are shown in FIGS. 5B, 5D, 5F, and 5H.

  W related to the processing is a stroke width estimated in advance. For example, the value of w may be estimated by experience, or the value of w may be estimated by viewing an image. In addition, the above formulas (1) and (2) show only specific examples in which the image data for the four directions of the horizontal direction, the vertical direction, and the two diagonal directions of the pixel is combined. It is not limited to the invention. In practice, any processing is applied as long as it is an appropriate composition processing for emphasizing a stroke.

  Here, performing the offset in the opposite direction and the offset in the opposite direction does not always easily specify in which direction the offset effect can be realized for the non-specific image. It is. Therefore, the effect of enhancing the stroke can be realized without fail by trying the offset in the opposite direction and the offset in the opposite direction. If it is possible to specify in advance in which direction the effect of enhancing the stroke can be realized, it is not necessary to simultaneously execute the offset in the opposite direction and the offset in the opposite direction. That is, any one of the offset in the opposite direction and the offset in the opposite direction may be executed. In the example of the image shown in the present specification, as shown in FIGS. 6A to 6D, the effect of enhancing the stroke can be realized by offsetting in the opposite direction. As shown in FIGS. 7A to 7D, the effect of enhancing the stroke cannot be realized by offsetting in the opposite direction.

  In step S208, binarization processing and matching processing with a high threshold can be performed on the image combined in step S206.

  For example, it is possible to perform binarization processing with a high threshold on each of the first composite image data and the second composite image data for each direction using a high threshold provided in advance. Alignment processing is performed on the first composite image data for each direction that has been binarized by a high threshold value, and matching is performed on the second composite image data for each direction that has been binarized by a high threshold value. Process.

  The matching process here may be any one of a process for obtaining a product set, a process for obtaining a maximum value, and a process for obtaining an average value.

  If the matching process is a process for obtaining a product set, a process for obtaining a product set can be performed according to the following expression.

I _output-thin = I _{h-binarized-high} ∪I _{v-binarized-high} ∪I _{rd-binarized-high} ∪I _{ld-binarized-high}
I ' _output-thin = I' _{h-binarized-high} ∪I ' _{v-binarized-high} ∪I' _{rd-binarized-high} ∪I ' _{ld-binarized-high}
Note that I _output-thin is the first thin stroke image data for which a product set has been obtained, and I _{h-binarized-high} , I _{v-binarized-high} , I _{rd-binarized-high} and I _{ld-binarized-} Each of _high represents binarization processing with a high threshold for the first composite image data I _h (x, y), I _v (x, y), I _rd (x, y), and I _ld (x, y) It is the binarized image data which performed. Note that I ′ _output-thin is the second thin stroke image data for which a product set has been obtained, and I ′ _{h-binarized-high} , I ′ _{v-binarized-high} , I ′ _{rd-binarized-high} and I ′ Each of ' _{ld-binarized-high} includes second composite image data I _h ' (x, y), I _v '(x, y), I _rd ' (x, y) and I _ld '(x, y ) Is binarized image data obtained by performing binarization processing with a high threshold.

In step S210, thin stroke image data can be acquired. That is, I _output-thin and I ′ _output-thin are acquired.

  In step S212, a binarization process and a matching process with a low threshold can be performed on the synthesized image.

  For example, by using a low threshold smaller than the high threshold provided in advance, it is possible to perform binarization processing with a low threshold on each of the first composite image data and the second composite image data for each direction. . The first composite image data for each direction that has been subjected to the binarization process with a low threshold is subjected to the alignment process and the second composite image for each direction that has been subjected to the second binarization process. The data is subjected to a matching process.

  The matching process here may be any one of a process for obtaining a product set, a process for obtaining a maximum value, and a process for obtaining an average value.

If the matching processing is processing for obtaining a product set, for example, processing for obtaining a product set can be performed according to the following expression.
I _output-thick = I _{h-binarized-low} ∪I _{v-binarized-low} ∪I _{rd-binarized-low} ∪I _{ld-binarized-low}
I ' _output-thick = I' _{h-binarized-low} ∪I ' _{v-binarized-low} ∪I' _{rd-binarized-low} ∪I ' _{ld-binarized-low}
Note that I _output-thick is the first thick stroke image data for which a product set is obtained, and I _{h-binarized-low} , I _{v-binarized-low} , I _{rd-binarized-low,} and I _{ld-binarized-} Each of _low is a binarization process with a low threshold for the first composite image data I _h (x, y), I _v (x, y), I _rd (x, y), and I _ld (x, y) It is the binarized image data which performed. Note that I ′ _output-thick is the second thick stroke image data for which a product set has been obtained, and I ′ _{h-binarized-low} , I ′ _{v-binarized-low} , I ′ _{rd-binarized-low} and I ′ ' _{ld-binarized-low} is respectively the second composite image data I _h ' (x, y), I _v '(x, y), I _rd ' (x, y) and I _ld '(x, y ) Is binarized image data that has been binarized with a low threshold.

  As described above, if it is possible to specify in advance in which direction the effect of enhancing the stroke can be realized, the offset in the opposite direction and the offset in the opposite direction are executed simultaneously to obtain the first composite image data. There is no need to acquire the second composite image data. That is, offset is performed in a direction that can realize the effect of enhancing the stroke (for example, the opposite direction or the opposite direction), and corresponding composite image data is acquired. On the other hand, the binarization process and the matching process are performed on the composite image data.

  In step S214, the data processed in step S212 is selectively filtered according to filtering conditions provided in advance.

  Specifically, by using a filtering condition that is provided in advance, a filtering condition that is provided in advance is performed by performing a filtering process on data relating to the communication area in the first thick stroke image data and the second thick stroke image data, respectively. The image data of the first communication area and the image data of the second communication area satisfying the above are acquired.

  The filtering process performed on the first thick stroke image data and the second thick stroke image data by using the filtering condition provided in advance here means that the first thick stroke image data obtained in step S212 and the first thick stroke image data. This is because the thick stroke image data of 2 may include edge data of other objects, such as human edge data in an image, or edge data similar to other thick strokes, in addition to a thick stroke. By performing filtering processing on the first thick stroke image data and the second thick stroke image data using filtering conditions provided in advance, the image data of the communication area corresponding to the thick stroke can be acquired more accurately. Can do.

  The filtering conditions provided in advance are: (1) the gradation dispersion of the pixels in the communication area is smaller than a predetermined dispersion threshold; and (2) the polarity of the pixels from the inner edge to the outer edge of the communication area. Or (3) that the size of the communication area is within a predetermined size threshold value. Alternatively, the filtering condition may be any combination of two or a plurality of conditions among the above (1), (2), and (3).

  Here, the communication area is a closed area composed of edges indicated by the thick stroke image data obtained by the process of step S212.

  More specifically, the filtering condition (1), that is, that the gradation dispersion of the pixels in the communication area is smaller than a predetermined dispersion threshold, specifically, coordinate information related to the communication area is first acquired in the image processed in step S212. be able to. Next, the gradation dispersion of the communication area is specified based on the coordinate information and the original image shown in FIG. 4, and it is determined whether or not the gradation dispersion in the communication area is smaller than a predetermined dispersion threshold. . When the gradation variance in the communication area is smaller than a predetermined dispersion threshold, the communication area is reserved, that is, the communication area is considered to be related to a thick stroke. Otherwise, the communication area is filtered.

  The filtering condition (2), that is, the fact that the pixel polarities from the inner edge to the outer edge of the communication area are the same, specifically, the coordinate information regarding the communication area can be acquired first in the image processed in step S212. . Next, it is specified whether or not the pixel polarities from the inner edge to the outer edge of the communication area match based on the coordinate information and the original image shown in FIG. When the pixel polarities of the inner edge and the outer edge in the communication area are the same, the communication area is reserved, that is, the communication area is considered to be related to a thick stroke. Otherwise, the communication area is filtered.

  The polarity of the pixel here may be understood as the relationship between the gradation of the pixel of one inner edge and the gradation of the pixel of the outer edge adjacent to the inner edge pixel. For example, when the gradation of the pixel in the binarized image is set to “0” or “255”, if both of the gradation values of the two pixels are “0” or “255”, the two It is considered that the polarities of the pixels are the same. Otherwise, the polarities are considered inconsistent. Of course, other appropriate conditions for matching pixel polarities may be set as necessary, and will not be described in detail here.

  Filtering condition (3), that is, that the size of the communication area is within a predetermined size area, specifically, if the size of the communication area is within a predetermined threshold value, It is considered that the communication area is reserved, that is, the communication area is associated with a thick stroke. Otherwise, the communication area is filtered. Here, a thick stroke image is larger than a thin stroke image, but generally there is a comparatively common upper limit. Therefore, the above-described threshold value may be set based on the upper limit. Easy to understand. If the size of the communication area is too large, it may be considered that the communication area corresponds to a non-stroke object rather than a thick stroke. For example, in most cases, a thick stroke size does not occupy half of the overall image size and thus more than half. Of course, if the size of the thick stroke is surely large in a special case, such a mode can be reflected by adjusting the size of the threshold value provided in advance.

  The pre-set dispersion threshold and the pre-set size threshold may be specified based on experience values obtained in advance through testing, or may be specified based on visual observation of image samples. good. Of course, any other suitable method may be used.

  Of the filtering conditions (1), (2), and (3), filtering can be executed by combining two or all of the conditions.

  Next, in step S216, first thick stroke edge image data and second thick stroke edge image data respectively corresponding to the image data of the first communication area and the image data of the second communication area are acquired.

  In step S218, a first thin stroke image (as shown in FIG. 8) and a second thin stroke image (not shown) corresponding to the first thin stroke image data and the second thin stroke image data, respectively. And a first thick stroke image (as shown in FIG. 10) and a second thick stroke image (not shown) corresponding to the first thick stroke edge image data and the second thick stroke edge image data, respectively. )).

It should be understood that a thick stroke image can be obtained even when step S214 is not executed. In other words, a first thick stroke image and a second thick stroke image (as shown in FIG. 9) can be acquired based on the I _output-thick and I ′ _output-thick acquired in step S212. .

  In the present specification, the second thin stroke image and the second thick stroke image can be acquired by the above steps, but the effect of enhancing the stroke cannot be realized by the offset in the opposite direction. The second thin stroke image and the second thick stroke image are not specifically shown.

  Optionally, based on the accuracy of the text stroke, the recall rate of the text stroke, or the compromise between the accuracy of the text stroke and the recall rate of the text stroke, either the first thin stroke image or the second thin stroke image is selected. It can be selected as a final thin stroke image. In addition, based on the accuracy of the text stroke, the recall rate of the text stroke, or the compromise between the accuracy of the text stroke and the recall rate of the text stroke, either the first thick stroke image or the second thick stroke image is finalized. It can also be selected as a thick stroke image. In the example of the present specification, the effect of enhancing the stroke can be realized by offsetting in the opposite direction, so that the accuracy of the text stroke, the recall rate of the text stroke, or the compromise between the accuracy of the text stroke and the recall rate of the text stroke Based on this, the first thin stroke image and the first thick stroke image can be acquired as the final stroke extraction result.

  The accuracy of text strokes referred to herein may be understood as the ratio between the number of correct strokes actually detected and the number of strokes actually detected. The recall rate of text strokes may be understood as the ratio of the number of correct strokes actually detected to the number of correct strokes actually present.

  As described above, the method for extracting the text stroke image from the image has been described using the Sobel operator as an example. However, only a method for extracting a text stroke image from an image in an exemplary manner is described. Usable operators are not limited to this, and may be Robert operator, Prewitt operator, Laplace operator, log operator, Canny operator, or any other suitable operator.

  In addition, although it has been described that the matching process is performed after performing the binarization process in steps S208 and S212, the binarization process may be performed after the matching process is actually performed.

  The high and low thresholds mentioned above may be specified by experience or may be specified by several tests. However, this is merely exemplary and the high and low thresholds can be specified in any suitable manner.

  Hereinafter, with reference to FIG. 11, an apparatus 400 for extracting a text stroke image from an image according to an embodiment of the present invention will be described.

  FIG. 11 is a block diagram showing an apparatus 400 for extracting a text stroke image from an image according to an embodiment of the present invention. The apparatus 400 can perform, for example, the method of extracting a text stroke image from an image described with reference to FIG. 1 as described above.

  Specifically, the apparatus 400 relates to the text in the image by performing information enhancement means 402 that obtains edge information and gradient information of the image, and enhancement processing that is provided in advance in the acquired edge information and gradient information. Emphasis means 404 for enhancing edge information and gradient information, and stroke image acquisition means 406 for acquiring a text stroke image corresponding to the emphasized edge information and gradient information.

  Optionally, the apparatus 400 may comprise sorting means (not shown). The sorting means is arranged to sort the plurality of extracted text stroke images.

  The information acquisition means 402 is arranged to analyze the edge information and gradient information step signal or pulse signal of the representative image and extract the edge information and gradient information based on the analysis result.

  In an image, thin stroke image data can be represented by a pulse signal, and thick stroke image data and a large scale object similar to a thick stroke can be represented by a step signal. By analyzing the pulse signal, it is possible to extract edge information and gradient information of a thin stroke based on the analysis result. Further, it is possible to analyze step signals, extract edge information and gradient information of a thick stroke based on the analysis result, and extract edge information and gradient information of a large scale object similar to a thick stroke.

  Optionally, the apparatus 400 may further comprise noise reduction means (not shown). The noise reduction means is arranged so as to reduce noise by filtering the original image based on the clarity of the image or as necessary. For example, noise in the original image can be reduced by filtering the image using a low-pass filter. The low-pass filter may be a Gaussian filter, but is not limited thereto, and may be any appropriate low-pass filter known to those skilled in the art.

  The enhancement unit 404 is arranged to perform enhancement processing on edge information and gradient information based on various methods. Preferably, edge information and gradient information can be emphasized by binarization processing and matching processing. The enhancement processing here may be any of processing for obtaining a product set, processing for obtaining a maximum value, and processing for obtaining an average value. Preferably, as will be described later, the matching process may be a product set process.

  Hereinafter, with reference to FIG. 12, an apparatus 400 'for extracting a text stroke image from an image using a Sobel operator according to an embodiment of the present invention will be described.

  FIG. 12 is a block diagram showing an apparatus 400 'for extracting a text stroke image from an image using a Sobel operator according to an embodiment of the present invention.

  The apparatus 400 'can execute the method of extracting the text stroke image from the image with the Sobel operator according to the embodiment of the present invention shown in FIGS. 3 to 10 as described above.

  Specifically, like the apparatus 400, the apparatus 400 'includes an information acquisition unit 402', an enhancement unit 404 ', and a stroke image acquisition unit 406'. Optionally, the device 400 'may comprise a sorting means and a noise reduction means (none shown).

  The selecting means uses either the first thin stroke image or the second thin stroke image as a final thin stroke image based on at least one of the accuracy of the text stroke and the recall rate of the text stroke. It sorts and / or arranges so that either the 1st thick stroke image and the 2nd thick stroke image may be sorted as a final thick stroke image. Since the specific sorting process has been described with reference to FIGS. 3 to 10, the description is not repeated here.

  The noise reduction means is arranged to filter the original image (that is, the image to be processed) to reduce noise. Since the specific noise reduction processing has been described with reference to FIGS. 3 to 10, description thereof will not be repeated here.

  The information acquisition unit 402 ′ may include a convolution calculation sub unit 4022 and a division sub unit 4024. The convolution operation sub-unit 4022 performs convolution operations with the image data of the image by using a convolution kernel of a plurality of Sobel operators for acquiring edge information and gradient information in a plurality of directions of the image, respectively. Be placed. The sorting sub means 4024 is arranged to sort the result of the convolution operation for each direction into positive pulse response image data and negative pulse response image data for each direction.

  The enhancement unit 404 ′ may include a first composite image acquisition sub unit 4042 and / or a second composite image acquisition sub unit 4044. The first composite image acquisition sub means 4042 executes the offset and addition in the opposite direction to the positive pulse response image data and the negative pulse response image data for each direction, thereby adding the first composite image data for each direction. Arranged to get. The second composite image acquisition sub means 4044 performs the offset and addition in the opposite direction to the positive pulse response image data and the negative pulse response image data for each direction, thereby obtaining the second composite image data for each direction. Arranged to get.

  The stroke image acquisition unit 406 ′ includes a first binarization processing sub unit 4062, a first thin stroke image acquisition sub unit 4064, a second thin stroke image acquisition sub unit 4066, and a second binary value. The image processing sub-unit 4068, the first thick stroke image acquisition sub-unit 40610, and the second thick stroke image acquisition sub-unit 40612 may be provided.

  The first binarization processing sub-unit 4062 applies the first binarization processing to the first composite image data and the second composite image data for each direction by using a first threshold value provided in advance. Arranged to do.

  The first thin stroke image acquisition sub means 4064 performs alignment processing on the first composite image data for each direction in which the first binarization processing has been performed to acquire the first thin stroke image data. The first thin stroke image data corresponding to the first thin stroke image data is arranged to be acquired.

  The second thin stroke image acquisition sub means 4066 performs alignment processing on the second composite image data for each direction in which the first binarization processing has been performed to acquire the second thin stroke image data. The second thin stroke image data corresponding to the second thin stroke image data is arranged to be acquired.

  The second binarization processing sub-unit 4068 uses the second threshold value which is provided in advance and is smaller than the first threshold value, and outputs the first synthesized image data and the second synthesized image data for each direction. 2 are arranged so as to respectively perform binarization processing.

  The first thick stroke image acquisition sub-unit 40610 performs alignment processing on the first composite image data for each direction in which the second binarization processing has been performed to acquire the first thick stroke image data. The first thick stroke image data corresponding to the first thick stroke image data is arranged to be acquired.

  The second thick stroke image acquisition sub means 40612 performs alignment processing on the second composite image data for each direction in which the second binarization processing has been performed to acquire the second thick stroke image data. The second thick stroke image data corresponding to the second thick stroke image data is arranged to be acquired.

  The stroke image acquisition unit 406 may include only the first binarization processing sub-unit 4062, the first thin stroke image acquisition sub-unit 4064, and the second thin stroke image acquisition sub-unit 4066 as necessary. Alternatively, only the second binarization processing sub-unit 4068, the first thick stroke image acquisition sub-unit 40610, and the second thick stroke image acquisition sub-unit 40612 may be provided. In other words, the stroke image acquisition unit 406 is arranged to acquire only a thin stroke image or a thick stroke image as necessary. For example, a case where only text information of a thick stroke or a thin stroke exists in an image to be processed or only text information of a thick stroke or a thin stroke needs to be acquired corresponds to this case.

  Further, as described above, if it is possible to specify in advance in which direction the effect of enhancing the stroke can be realized, the enhancement unit 404 ′ includes the first synthesized image acquisition sub unit 4042 and the second Among the composite image acquisition sub means 4044, a composite image acquisition sub means corresponding to the processing in the direction may be provided. On the other hand, the stroke image acquisition unit 406 ′ includes a first thin stroke image acquisition sub-unit 4064 and a second thin stroke image acquisition sub-unit 4066, which correspond to the processing in the direction, and the first thick stroke. Of the image acquisition sub means 40610 and the second thick stroke image acquisition sub means 40612, only the one corresponding to the processing in the direction may be provided.

  The apparatus 400 ′ is arranged to execute a method for extracting a text stroke image from an image with the Sobel operator according to the embodiment of the present invention shown in FIGS. 3 to 10 as described above, for example. . For the sake of simplicity, here, the first binarization processing sub-unit 4062, the first thin stroke image acquisition sub-unit 4064, the second thin stroke image acquisition sub-unit 4066, and the second binarization The processing sub-unit 4068, the first thick stroke image acquisition sub-unit 40610, and the second thick stroke image acquisition sub-unit 40612 are not specifically described.

  Optionally, the stroke image acquisition unit 406 ′ may further include a filtering sub unit (not shown). The filtering sub means performs a filtering process on data related to the communication area in the first thick stroke edge image data and the second thick stroke edge data by using a filtering condition that is set in advance, and the filtering condition that is set in advance The image data of the first communication area and the image data of the second communication area satisfying the above are acquired.

  The pre-set filtering condition is the same as the condition that the gradation dispersion of the pixels in the communication area is smaller than the pre-set dispersion threshold and the pixel polarity from the inner edge to the outer edge of the communication area. It includes at least one of a condition and a condition in which the size of the communication area is within a predetermined threshold value. Since the filtering conditions provided in advance have been described with reference to FIGS. 3 to 10, description thereof will not be repeated here.

  At least one of the following technical effects can be realized by the method and apparatus for extracting a text stroke image from an image according to the embodiment of the present invention. That is, there are an effect of increasing the speed and an effect of suppressing noise and insensitivity to the stroke scale due to the bad quality of the video processed simultaneously with the extraction of the stroke. Further, the method and apparatus for extracting a text stroke image from an image according to the embodiment of the present invention eliminates the need for conventional knowledge such as text color and text background contrast.

  Although the basic principle of the present invention has been described above using specific embodiments, all of the methods and apparatuses of the present invention, or any step or component, can be replaced with any arithmetic device (processor, storage medium, etc. Or a combination thereof in a network of computing devices can be realized by those skilled in the art. This can be accomplished with basic programming skills by those skilled in the art when considering the description of the invention.

  Therefore, the object of the present invention can be further realized by executing one program or a set of programs on any arithmetic device. The arithmetic device may be a known general-purpose device. Therefore, the object of the present invention can be realized simply by providing a program product including a program code for realizing the method or apparatus. That is, such a program product and a medium storing such a program product also constitute the present invention. Of course, the storage medium may be any kind of storage medium that is already known or that will be developed in the future.

  When the embodiment of the present invention is realized by software and / or firmware, a program that configures the software from a storage medium or a network to a computer having a dedicated hardware configuration, for example, the general-purpose computer 1200 shown in FIG. Install. The computer can execute functions and the like when various programs are installed.

  In FIG. 13, a central processing unit (CPU) 1201 executes various processes based on a program stored in a read-only memory (ROM) 1202 or a program loaded from a storage unit 1208 into a random access memory (RAM) 1203. To do. The RAM 1203 also stores data necessary for the CPU 1201 to execute various processes as necessary. The CPU 1201, ROM 1202, and RAM 1203 are connected to each other via a bus 1204. An input / output interface 1205 is also connected to the bus 1204.

  An input unit 1206 (including a keyboard, a mouse, and the like), an output unit 1207 (including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker), and a storage unit 1208 (including a hard disk), A communication unit 1209 (including a network interface card such as a LAN card and a modem) is connected to the input / output interface 1205. The communication unit 1209 executes communication processing via a network, for example, the Internet. A driver 1210 is also connected to the input / output interface 1205 as necessary. A removable medium 1211 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is attached to the driver 1210 as necessary. Thereby, the read computer program is installed in the storage unit 1208 as necessary.

  When the series of processing is realized by software, a program constituting the software is installed from a network, for example, the Internet, or a storage medium, for example, a removable medium 1211.

  Those skilled in the art will recognize that such storage media is not limited to the removable media 1211 shown in FIG. 13 in which the program is stored and delivered remotely from the device to provide the program to the user. Should be understood. Examples of the removable medium 1211 include a magnetic disk (including a floppy disk (registered trademark)), an optical disk (including a compact disk read-only memory (CD-ROM) and a digital versatile disk (DVD)), and a magneto-optical disk (mini-disk). Disk (MD) (including registered trademark) and semiconductor memory. Alternatively, the storage medium may be a hard disk included in the ROM 1202 and the storage unit 1208, in which a program is stored, and a hard disk delivered to the user together with a device including them.

  The present invention further discloses a program product for storing a computer readable command code. When the command code is read and executed by a computer, the above-described method according to the embodiment of the present invention can be executed.

  On the other hand, a storage medium that holds a program product that stores the computer-readable command code is also included in the disclosure of the present invention. Storage media include, but are not limited to, floppy disks, optical disks, magneto-optical disks, memory cards, memory sticks, and the like.

  Those skilled in the art should understand that the listing herein is exemplary and not limiting.

  In the present specification, descriptions such as “first”, “second”, and “Nth” are for clearly explaining the present invention by distinguishing the described features in characters. Therefore, it does not have any limiting meaning.

  Each step of the method and each component module and / or means of the device may be implemented by way of example as software, firmware, hardware or a combination thereof and may be part of a corresponding device. Since a specific method or form that can be used when each component module and means in the above apparatus is arranged in the form of software, firmware, hardware, or a combination thereof is well known to those skilled in the art, it will not be described in detail here.

  When implemented by software or firmware, for example, installing a program constituting the software from a storage medium or a network to a computer having a dedicated hardware configuration (for example, the general-purpose computer 1200 shown in FIG. 13) Can do. The computer can execute functions and the like when various programs are installed.

  In the above description of specific embodiments of the invention, features described and / or illustrated for one type of embodiment may be used in one or more other embodiments in the same or similar form. In addition, the features in the other embodiments can be combined or replaced with the features in the other embodiments.

  It should be emphasized that the term “comprising / comprising” as used herein refers to the presence of a feature, element, step or component, but one or more other features, elements, steps or components. The presence or addition of is not excluded.

  Further, the method of the present invention is not limited to being executed in the order of time described in the specification, and may be executed sequentially, in parallel, or individually according to other times. Therefore, the order of execution of the methods described herein does not limit the technical scope of the present invention.

As described above, the present invention has been disclosed by the description of the specific embodiments of the present invention. However, those skilled in the art can make various modifications, improvements, or equivalents to the present invention within the spirit and scope of the present invention appended below. You can design things. These changes, improvements or equivalents should be admitted to be within the protection scope of the present invention.
(Appendix)
(Appendix 1)
A method for extracting a text stroke image from an image,
Obtaining edge information and gradient information of the image;
Emphasizing edge information and gradient information relating to text in the image by performing enhancement processing provided in advance on the acquired edge information and gradient information;
Obtaining a text stroke image corresponding to the enhanced edge information and gradient information;
A method comprising the steps of:
(Appendix 2)
The step of acquiring edge information and gradient information of the image includes
Analyzing a step signal or a pulse signal representing edge information and gradient information of the image, and extracting the edge information and the gradient information based on an analysis result,
The method according to appendix 1.
(Appendix 3)
The step of acquiring edge information and gradient information of the image includes
Using convolution kernels of a plurality of Sobel operators for acquiring edge information and gradient information in a plurality of directions of the image, respectively performing a convolution operation with the image data of the image,
Partitioning the result of the convolution operation for each direction into positive pulse response image data and negative pulse response image data for each direction;
Emphasizing edge information and gradient information about text in the image,
Performing offset and addition in opposite directions on the positive and negative pulse response image data for each direction to obtain first composite image data for each direction, and / or
Performing offset and addition in opposite directions on the positive and negative pulse response image data for each direction to obtain second composite image data for each direction;
Obtaining a text stroke image corresponding to the emphasized edge information and gradient information,
A first binarization process is performed on each of the first composite image data and / or the second composite image data for each direction by using a first threshold value provided in advance. The first thin stroke image corresponding to the first thin stroke image data is obtained by performing the alignment process on the first composite image data for each direction and acquiring the first narrow stroke image data. And / or performing the matching process on the second composite image data for each direction in which the first binarization process has been performed, and acquiring the corresponding second thin stroke image data, Obtaining a second thin stroke image corresponding to the second thin stroke image data, and / or
A second binarization process is performed on each of the first composite image data and / or the second composite image data for each direction using a second threshold value smaller than the first threshold value provided in advance. The first thick stroke image data is obtained by performing matching processing on the first composite image data for each direction that has been subjected to the binarization processing of 2, and acquiring the first thick stroke image data. The first thick stroke image is acquired and / or the second binarization processing is performed, the second combined image data for each direction is subjected to matching processing, and the corresponding second thick stroke image Acquiring a second thick stroke image corresponding to the second thick stroke image data by acquiring data,
The method according to appendix 2.
(Appendix 4)
The matching process includes any one of a process for obtaining a product set, a process for obtaining a maximum value, and a process for obtaining an average value.
The method according to attachment 3.
(Appendix 5)
The steps of performing convolution operations with the image data of the image using a convolution kernel of a plurality of Sobel operators for acquiring edge information and gradient information in a plurality of directions of the image, respectively.
Using the convolution kernel of four Sobel operators to acquire edge information and gradient information in four directions of the horizontal direction, vertical direction and two diagonal directions of the image, respectively, the image data and convolution operation of the image Including doing,
The method according to appendix 3 or 4.
(Appendix 6)
The offset and addition in the opposite direction are as follows:
I _h (x, y) = (I _h-positive (x, yw / 2) + I _h-negative (x, y + w / 2)) / 2,
I _v (x, y) = (I _v-positive (xw / 2, y) + I _v-negative (x + w / 2, y)) / 2,
I _rd (x, y) = (I _rd-positive (x + w / 2, yw / 2) + I _rd-negative (xw / 2, y + w / 2)) / 2, and
I _ld (x, y) = (I _ld-positive (xw / 2, yw / 2) + I _ld-negative (x + w / 2, y + w / 2)) / 2,
Based on
The offset and addition in the opposite direction is given by
I _h '(x, y) = (I _h-positive (x, y + w / 2) + I _h-negative (x, yw / 2)) / 2,
I _v '(x, y) = (I _v-positive (x + w / 2, y) + I _v-negative (xw / 2, y)) / 2,
I _rd '(x, y) = (I _rd-positive (xw / 2, y + w / 2) + I _rd-negative (x + w / 2, yw / 2)) / 2, and
I _ld '(x, y) = (I _ld-positive (x + w / 2, y + w / 2) + I _ld-negative (xw / 2, yw / 2)) / 2,
Based on
x is the abscissa of the pixel, y is the ordinate of the pixel, I _h (x, y), I _v (x, y), I _rd (x, y) and I _ld (x, y) are the pixels First composite image data for four directions of the horizontal direction, the vertical direction, and the two diagonal directions, respectively,
I _h ′ (x, y), I _v ′ (x, y), I _rd ′ (x, y) and I _ld ′ (x, y) are the horizontal direction, vertical direction and two diagonal directions of the pixel The second composite image data for the four directions are shown respectively.
I _h-positive (x, y), I _v-positive (x, y), I _rd-positive (x, y) and I _ld-positive (x, y) are the horizontal direction, vertical direction and Positive pulse response image data for four directions in two diagonal directions are shown respectively.
I _h-negative (x, y), I _v-negative (x, y), I _rd-negative (x, y) and I _ld-negative (x, y) are the horizontal direction, vertical direction and The negative pulse response image data for four methods in two diagonal directions are shown respectively, and w is a stroke width estimated in advance.
The method according to any one of appendices 3 to 5.
(Appendix 7)
The steps of obtaining the first thick stroke image data and the second thick stroke image data corresponding to the first thick stroke edge image data and the second thick stroke edge image data, respectively.
By performing a filtering process on the communication area data in the first thick stroke edge image data and the second thick stroke edge image data by using a predetermined filtering condition, the predetermined filtering condition is set. Acquiring image data of the first communication area and image data of the second communication area to be satisfied,
Obtaining a first thick stroke image and a second thick stroke image corresponding to the image data of the first communication area and the image data of the second communication area, respectively.
The method according to any one of appendices 3 to 6.
(Appendix 8)
The filtering condition provided in advance is at least:
A condition in which gradation dispersion of pixels in the communication area is smaller than a predetermined dispersion threshold;
A condition in which pixel polarities from the inner edge to the outer edge in the communication region match,
A condition in which the size of the communication area is within a predetermined threshold value;
Including one of
The method according to appendix 7.
(Appendix 9)
Based on at least one of the accuracy of the text stroke and the recall rate of the text stroke, one of the first thin stroke image and the second narrow stroke image is selected as a final thin stroke image, And / or further comprising selecting one of the first thick stroke image and the second thick stroke image as a final thick stroke image.
The method according to any one of appendices 3 to 8.
(Appendix 10)
An apparatus for extracting a text stroke image from an image,
Information acquisition means for acquiring edge information and gradient information of the image;
Emphasis means for emphasizing edge information and gradient information related to text in the image by performing enhancement processing provided in advance on the acquired edge information and gradient information;
A stroke image acquisition means for acquiring a text stroke image corresponding to the emphasized edge information and gradient information;
A device comprising:
(Appendix 11)
The information acquisition unit analyzes a step signal or a pulse signal representing edge information and gradient information of the image, and extracts the edge information and the gradient information based on an analysis result.
The apparatus according to appendix 10.
(Appendix 12)
The information acquisition means includes
Using a convolution kernel of a plurality of Sobel operators for acquiring edge information and gradient information in a plurality of directions of an image, a convolution operation sub-unit for performing a convolution operation with the image data of the image, and
A division sub-unit for dividing the result of the convolution operation for each direction into positive pulse response image data and negative pulse response image data for each direction;
The highlighting means is
A first composite image acquisition sub-unit that performs offset and addition in opposite directions to the positive pulse response image data and the negative pulse response image data for each direction to acquire first composite image data for each direction; And / or
A second composite image acquisition sub-unit that performs offset and addition in opposite directions on the positive pulse response image data and the negative pulse response image data for each direction to acquire second composite image data for each direction; Prepared,
The stroke image acquisition means includes:
First binarization processing sub-unit for respectively performing first binarization processing on the first composite image data and / or the second composite image data for each direction using a first threshold value provided in advance; The first thin stroke image data is obtained by performing the alignment processing on the first composite image data for each direction that has been subjected to the first binarization process and obtaining the first thin stroke image data. The first thin stroke image acquisition sub-unit for acquiring the corresponding first thin stroke image and / or the second binarized image data that has been subjected to the first binarization processing is matched with the second combined image data. And obtaining second thin stroke image data corresponding to the second thin stroke image data by obtaining the corresponding second thin stroke image data; and And / or
A second binarization process is performed for each of the first composite image data and / or the second composite image data for each direction using a second threshold value that is smaller than the first threshold value provided in advance. The binarization processing sub-unit of the first and second binarization processing, the first combined image data for each direction is subjected to alignment processing to obtain the first thick stroke image data, The first thick stroke image acquisition sub-unit for acquiring the first thick stroke image corresponding to the first thick stroke image data and / or the second binarization processing is performed for each direction. The second thick image data corresponding to the second thick stroke image data is acquired by performing matching processing on the composite image data 2 and acquiring the corresponding second thick stroke image data. A thick stroke image acquisition sub-unit,
The apparatus according to appendix 11.
(Appendix 13)
The matching process includes any one of a process for obtaining a product set, a process for obtaining a maximum value, and a process for obtaining an average value.
The apparatus according to appendix 12.
(Appendix 14)
The convolution operation sub means uses a convolution kernel of four Sobel operators for obtaining edge information and gradient information in four directions of the horizontal direction, the vertical direction, and two diagonal directions of the image. The apparatus according to appendix 12 or 13, which performs image data and a convolution operation, respectively.
(Appendix 15)
The offset and addition in the opposite direction are as follows:
I _h (x, y) = (I _h-positive (x, yw / 2) + I _h-negative (x, y + w / 2)) / 2,
I _v (x, y) = (I _v-positive (xw / 2, y) + I _v-negative (x + w / 2, y)) / 2,
I _rd (x, y) = (I _rd-positive (x + w / 2, yw / 2) + I _rd-negative (xw / 2, y + w / 2)) / 2, and
I _ld (x, y) = (I _ld-positive (xw / 2, yw / 2) + I _ld-negative (x + w / 2, y + w / 2)) / 2,
Based on
The offset and addition in the opposite direction is given by
I _h '(x, y) = (I _h-positive (x, y + w / 2) + I _h-negative (x, yw / 2)) / 2,
I _v '(x, y) = (I _v-positive (x + w / 2, y) + I _v-negative (xw / 2, y)) / 2,
I _rd '(x, y) = (I _rd-positive (xw / 2, y + w / 2) + I _rd-negative (x + w / 2, yw / 2)) / 2, and
I _ld '(x, y) = (I _ld-positive (x + w / 2, y + w / 2) + I _ld-negative (xw / 2, yw / 2)) / 2,
Based on
x is the abscissa of the pixel, y is the ordinate of the pixel, I _h (x, y), I _v (x, y), I _rd (x, y) and I _ld (x, y) are the pixels 1st composite image data for four directions of the horizontal direction, the vertical direction, and two diagonal directions, respectively,
I _h ′ (x, y), I _v ′ (x, y), I _rd ′ (x, y) and I _ld ′ (x, y) are the horizontal direction, vertical direction and two diagonal directions of the pixel The second composite image data for the four directions are shown respectively.
I _h-positive (x, y), I _v-positive (x, y), I _rd-positive (x, y) and I _ld-positive (x, y) are the horizontal direction, vertical direction and Positive pulse response image data for four directions in two diagonal directions are shown respectively.
I _h-negative (x, y), I _v-negative (x, y), I _rd-negative (x, y) and I _ld-negative (x, y) are the horizontal direction, vertical direction and The negative pulse response image data for four methods in two diagonal directions are shown respectively, and w is a stroke width estimated in advance.
The apparatus according to any one of appendices 12 to 14.
(Appendix 16)
The stroke image acquisition means further performs a filtering process on the communication area data in the first thick stroke edge image data and the second thick stroke edge image data by using filtering conditions provided in advance. A filtering sub-unit for acquiring image data of a first communication area and image data of a second communication area that satisfy the filtering condition provided in advance.
The apparatus according to any one of appendices 12 to 15.
(Appendix 17)
The predetermined filtering condition includes at least a condition in which gradation dispersion of pixels in the communication area is smaller than a predetermined dispersion threshold, and pixel polarity from the inner edge to the outer edge in the communication area. Including one of a matching condition and a condition in which the size of the communication region is within a predetermined size threshold,
The apparatus according to appendix 16.
(Appendix 18)
Based on at least one of the accuracy of the text stroke and the recall rate of the text stroke, one of the first thin stroke image and the second narrow stroke image is selected as a final thin stroke image, And / or further comprising a selecting means for selecting either the first thick stroke image or the second thick stroke image as a final thick stroke image.
The apparatus according to any one of appendices 12 to 17.
(Appendix 19)
A computer program comprising instructions for causing a computer to execute the method for extracting a text stroke image from an image according to any one of appendices 1 to 9.
(Appendix 20)
A computer-readable recording medium on which the computer program according to attachment 19 is recorded.

400: An apparatus for extracting a text stroke image from an image 402: Information acquisition means 404: An emphasis means 406: A stroke image acquisition means 400 ′: An apparatus for extracting a text stroke image from an image 402 ′: An information acquisition means 4022: A convolution operation sub Means 4024: Classification sub means 404 ′: Emphasis means 4042: First composite image acquisition sub means 4044: Second composite image acquisition sub means 406 ′: Stroke image acquisition means 4062: First binarization processing sub means 4064 First thin stroke image acquisition sub-unit 4066 Second thin stroke image acquisition sub-unit 4068 Second binarization processing sub-unit 40610 First thick stroke image acquisition sub-unit 40612 Second thick stroke Image acquisition sub means

Claims (10)

A method for extracting a text stroke image from an image,
Obtaining edge information and gradient information of the image;
Emphasizing edge information and gradient information relating to text in the image by performing enhancement processing provided in advance on the acquired edge information and gradient information;
Obtaining a text stroke image corresponding to the enhanced edge information and gradient information;
A method comprising the steps of: The step of acquiring edge information and gradient information of the image includes
Analyzing a step signal or a pulse signal representing edge information and gradient information of the image, and extracting the edge information and the gradient information based on an analysis result,
The method of claim 1. The step of acquiring edge information and gradient information of the image includes
Using convolution kernels of a plurality of Sobel operators for acquiring edge information and gradient information in a plurality of directions of the image, respectively performing a convolution operation with the image data of the image,
Partitioning the result of the convolution operation for each direction into positive pulse response image data and negative pulse response image data for each direction;
Emphasizing edge information and gradient information about text in the image,
Performing offset and addition in opposite directions on the positive and negative pulse response image data for each direction to obtain first composite image data for each direction, and / or
Performing offset and addition in opposite directions on the positive and negative pulse response image data for each direction to obtain second composite image data for each direction;
Obtaining a text stroke image corresponding to the emphasized edge information and gradient information,
A first binarization process is performed on each of the first composite image data and / or the second composite image data for each direction by using a first threshold value provided in advance. The first thin stroke image corresponding to the first thin stroke image data is obtained by performing the alignment process on the first composite image data for each direction and acquiring the first narrow stroke image data. And / or performing the matching process on the second composite image data for each direction in which the first binarization process has been performed, and acquiring the corresponding second thin stroke image data, Obtaining a second thin stroke image corresponding to the second thin stroke image data, and / or
A second binarization process is performed on each of the first composite image data and / or the second composite image data for each direction using a second threshold value smaller than the first threshold value provided in advance. The first thick stroke image data is obtained by performing matching processing on the first composite image data for each direction that has been subjected to the binarization processing of 2, and acquiring the first thick stroke image data. The first thick stroke image is acquired and / or the second binarization processing is performed, the second combined image data for each direction is subjected to matching processing, and the corresponding second thick stroke image Acquiring a second thick stroke image corresponding to the second thick stroke image data by acquiring data,
The method of claim 2. The matching process includes any one of a process for obtaining a product set, a process for obtaining a maximum value, and a process for obtaining an average value.
The method of claim 3. The steps of performing convolution operations with the image data of the image using a convolution kernel of a plurality of Sobel operators for acquiring edge information and gradient information in a plurality of directions of the image, respectively.
Using the convolution kernel of four Sobel operators to acquire edge information and gradient information in four directions of the horizontal direction, vertical direction and two diagonal directions of the image, respectively, the image data and convolution operation of the image Including doing,
The method according to claim 3 or 4. The offset and addition in the opposite direction are as follows:
I _h (x, y) = (I _h-positive (x, yw / 2) + I _h-negative (x, y + w / 2)) / 2,
I _v (x, y) = (I _v-positive (xw / 2, y) + I _v-negative (x + w / 2, y)) / 2,
I _rd (x, y) = (I _rd-positive (x + w / 2, yw / 2) + I _rd-negative (xw / 2, y + w / 2)) / 2, and
I _ld (x, y) = (I _ld-positive (xw / 2, yw / 2) + I _ld-negative (x + w / 2, y + w / 2)) / 2,
Based on
The offset and addition in the opposite direction is given by
I _h '(x, y) = (I _h-positive (x, y + w / 2) + I _h-negative (x, yw / 2)) / 2,
I _v '(x, y) = (I _v-positive (x + w / 2, y) + I _v-negative (xw / 2, y)) / 2,
I _rd '(x, y) = (I _rd-positive (xw / 2, y + w / 2) + I _rd-negative (x + w / 2, yw / 2)) / 2, and
I _ld '(x, y) = (I _ld-positive (x + w / 2, y + w / 2) + I _ld-negative (xw / 2, yw / 2)) / 2,
Based on
x is the abscissa of the pixel, y is the ordinate of the pixel, I _h (x, y), I _v (x, y), I _rd (x, y) and I _ld (x, y) are the pixels First composite image data for four directions of the horizontal direction, the vertical direction, and the two diagonal directions, respectively,
I _h ′ (x, y), I _v ′ (x, y), I _rd ′ (x, y) and I _ld ′ (x, y) are the horizontal direction, vertical direction and two diagonal directions of the pixel The second composite image data for the four directions are shown respectively.
I _h-positive (x, y), I _v-positive (x, y), I _rd-positive (x, y) and I _ld-positive (x, y) are the horizontal direction, vertical direction and Positive pulse response image data for four directions in two diagonal directions are shown respectively.
I _h-negative (x, y), I _v-negative (x, y), I _rd-negative (x, y) and I _ld-negative (x, y) are the horizontal, vertical and The negative pulse response image data for four methods in two diagonal directions are shown respectively, and w is a stroke width estimated in advance.
6. A method according to any one of claims 3-5. The steps of obtaining the first thick stroke image data and the second thick stroke image data corresponding to the first thick stroke edge image data and the second thick stroke edge image data, respectively.
By performing a filtering process on the communication area data in the first thick stroke edge image data and the second thick stroke edge image data by using a predetermined filtering condition, the predetermined filtering condition is set. Acquiring image data of the first communication area and image data of the second communication area to be satisfied,
Obtaining a first thick stroke image and a second thick stroke image corresponding to the image data of the first communication area and the image data of the second communication area, respectively.
The method according to claim 3. The filtering condition provided in advance is at least:
A condition in which gradation dispersion of pixels in the communication area is smaller than a predetermined dispersion threshold;
A condition in which pixel polarities from the inner edge to the outer edge in the communication region match,
A condition in which the size of the communication area is within a predetermined threshold value;
Including one of
The method of claim 7. Based on at least one of the accuracy of the text stroke and the recall rate of the text stroke, one of the first thin stroke image and the second narrow stroke image is selected as a final thin stroke image, And / or further comprising selecting one of the first thick stroke image and the second thick stroke image as a final thick stroke image.
9. A method according to any one of claims 3 to 8. An apparatus for extracting a text stroke image from an image,
Information acquisition means for acquiring edge information and gradient information of the image;
Emphasis means for emphasizing edge information and gradient information related to text in the image by performing enhancement processing provided in advance on the acquired edge information and gradient information;
A stroke image acquisition means for acquiring a text stroke image corresponding to the emphasized edge information and gradient information;
A device comprising: