JP2000194851A - Character area extraction method and apparatus, and recording medium recording this method - Google Patents

Abstract

(57) [Problem] To provide a character area extraction method and apparatus for extracting a character area from an image including a telop character or the like, smoothing a contour shape and suppressing extraction failure. SOLUTION: Adjacent pixels having similar luminance values in a color image are defined as connected regions by a luminance similar region forming unit 2. Here, after binarization for each line in the horizontal and vertical directions, the edge pixels are removed, the character part is separated and extracted from the background, and the binarization results for each direction are superimposed, compared, integrated and integrated to form a connected area. To form Since the character portion has the same position and shape for the binarization results in the horizontal and vertical directions, extraction failure of the character region is suppressed. Neighboring pixels having similar saturation values in the connected region pixels are formed as connected regions by the saturation value similar region forming unit 3. The hue value similarity area forming unit 4 forms a plurality of pixels having similar hue values in the connected area pixels having similar luminance values and saturation values, and adjacent to each other in the image space. This connected area is defined as a character area image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to, for example, a frame image in which characters such as telop characters in a television broadcast image are displayed in a plurality of frames constituting a color moving image such as a television broadcast image. The present invention relates to a character region extraction technique for extracting a character portion as a pixel connection region, and more specifically, when extracting a character region as a binary image from a video including characters, the binary value of the image is determined by a maximum value method / minimum value method. TECHNICAL FIELD The present invention relates to a character area extraction technique which has an advantage that the number of pixels constituting a character is small when the conversion technique is used.

[0002]

2. Description of the Related Art A character region extraction technique for extracting a character portion as a pixel connection region from a frame image in which characters in a plurality of frames constituting a color moving image are displayed or a still color image in which characters are displayed. about,
Many researches have been performed conventionally. As shown in FIG. 13, when the input character display color image is an image in an NTSC signal format image, the color of each pixel is blurred along each horizontal line, and a narrow vertical stroke component is also present in the character portion. May deteriorate in color. There is a need for a method for stably extracting a character area even when color degradation occurs in the same character.

[0003] To solve this problem, a character area extraction method focusing on the characteristics of the luminance distribution on a line has been proposed (reference document [1]: Kuwano, Arai, Kurakake, Sugimura: "Binary in line units"). Extraction Method of Character Areas in Video by Image Processing, Shingaku Akidai, D-12-42, pp. 264, (1996-0
9)).

This method focuses on the feature that the luminance distribution on a horizontal line has a high contrast between the telop character and the background even if color bleeding occurs. Perform the conversion process. According to the binarization method, a range in which the luminance value is locally higher in the line by a preset value and a range in which the luminance value is lower by a preset value as compared with the peripheral portion is extracted as a character area. The binarization method in this line is performed for each horizontal line and each vertical line, and the binarization results in each direction are compared. Is the final character area extraction result.

[0005]

However, in the above-described conventional method, as shown in FIG. 14, the character range to be extracted in each line is determined in advance from the pixel having the extreme value in the luminance distribution on the line. In the horizontal stroke component where color deterioration occurs only at the end, if the luminance difference between the deteriorated part and the non-deteriorated part is larger than a predetermined value, the color deteriorated part is set to the character range. Cannot be included and extracted as a range. In addition, since there are various degrees of color deterioration, depending on the threshold value of the predetermined luminance difference, it may be possible to extract the image including the color deterioration part. Therefore, FIG.
As shown in (2), near the end of the horizontal stroke component, there is a problem that the contour of the region becomes jagged or the stroke is interrupted in the final character region extraction result.

In the binarization processing of the horizontal line, the horizontal boundary between the character and the background can be detected, but the luminance distribution in the vertical direction is not considered at all. Therefore, as shown in FIG. Since the boundary between the character and the background is not detected, the region pixel of the background may be extracted while being connected in the vertical direction of the character. Similarly, in the binarization processing of the vertical line, the detection of the boundary between the character and the background in the vertical direction is guaranteed, but the detection of the boundary between the character and the background in the horizontal direction is not performed. In some cases, extraction is performed while the area pixels of the background part are connected in the horizontal direction of the character part. Therefore, even if the binarization result for each horizontal line is compared with the binarization result for each vertical line, the area of the character portion is not necessarily
The positions, sizes, and shapes are not necessarily the same, there is also a problem that they cancel each other out and fail to extract the character area.

Accordingly, the present invention provides a character region extracting method and apparatus which smoothes the shape of a character outline portion and suppresses a failure in extracting a character region even when brightness and color deteriorate in the same character. That is the task.

[0008]

In order to solve the above-mentioned problems, a character area extracting method according to the present invention comprises a first step of inputting and storing a color image in which characters are displayed;
A plurality of pixels having similar luminance values in the color image input in the first step and adjacent in the image space
And a first connected area image having similar luminance values obtained in the second step and having similar saturation values in each area, and In a third step of forming a plurality of pixels adjacent to each other as a second connected area, and a second connected area image similar in both luminance value and saturation value obtained in the third step, A fourth step of forming a plurality of pixels having similar hue values in the region and adjacent in the pixel space as a third connected region;
And a fifth step in which only the third connected area pixel obtained in the step is left in the image and stored as a character area image.

Alternatively, in the above character area extracting method, the first step of inputting and storing a color image in which characters are displayed is performed by using the same telop character or the like in a color moving image such as a television broadcast image. A stage 1-1 in which a plurality of temporally continuous color images in which characters are displayed are input and stored, or a single still color image in which characters are displayed is input and stored And when a plurality of color images in which the same character is displayed are input, an average color having an average value of pixel values of corresponding pixels at corresponding positions among the plurality of color images as a value of each pixel An image is created and output as an input color image.
It is characterized by realizing using the two steps.

Alternatively, in the above-described character area extracting method, the second step of forming a plurality of pixels having similar luminance values in a color image and adjacent to each other in an image space as a first connected area is defined as Performing a binary conversion of luminance within each horizontal line in the image to obtain a binary image as a result, and performing binary conversion of luminance within each of the vertical lines in the color image. 2-2 as a binary image,
2 obtained by the (2-1) th step and the (2-2) th step
The connected components in two binary images are compared, and the area, shape,
The connected components whose differences in position are within a preset allowable range are determined as character regions, and are left in the binary image, and the connected components whose differences in area, shape, and position exceed the preset allowable range are not included. The second and third steps of determining the background noise portion and removing the binary noise image from the binary image are realized using the method.

Alternatively, in the above-described character region extracting method, a two-first step of performing binarization of luminance in each horizontal line in a color image and obtaining a result as a binary image is performed. Step 2-1-1: examining a luminance value distribution in a certain horizontal line, and extracting a range in which the luminance value is locally higher in the horizontal line by a predetermined value as compared with a peripheral portion as a connected pixel area. And the absolute value of the luminance gradient in the horizontal direction is the largest among a plurality of pixels selected by a predetermined method near the left and right end pixels of the connected pixel area obtained in the step 2-1-1. Pixel on each of the left and right
Step 2-1-1 of extracting a range having both ends of the pixel as a high-luminance character area, and examining a luminance value distribution in a certain horizontal line in the color image, and locally examining luminance in the horizontal line. A step 2-1-3 of extracting a range in which the value is lower by a preset value as compared with the peripheral part as a connected pixel area,
The absolute value of the luminance gradient in the horizontal direction is maximized from among a plurality of pixels selected by a predetermined method existing near both right and left pixels of the connected pixel area obtained in the step 2-1-3. A pixel is obtained on each of the left and right sides, and a range having both ends of the pixel is extracted as a low-luminance character area, a stage 2-1-4, and each horizontal line extracted in the stage 2-1-2 is extracted. The high brightness character areas are collected to create a high brightness character area image as the entire image, and the low brightness character areas on each horizontal line extracted in the stage 2-1-4 are collected to obtain the entire image. 2-1 for creating a low-brightness character area image as an image
-5, a 2-1-6th step of detecting an edge pixel in the color image, a high-luminance character area image obtained in the 2-1-5th step, and a low-luminance From the pixels constituting the character connected component in the character area image,
2-1-7 step of removing a pixel corresponding to a vertical edge portion in the color image obtained in step -6;
It is characterized by realizing by using.

Alternatively, in the above-described character region extraction method, the step 2-2 in which the luminance is binarized in each vertical line in the color image and the result is obtained as a binary image, A step of examining a luminance value distribution in a certain vertical line, and extracting a range in which the luminance value is locally higher than a peripheral part by a predetermined value in the vertical line as a connected pixel area, as a 2-2-1 step; And the absolute value of the luminance gradient in the vertical direction is the largest among a plurality of pixels selected by a predetermined method near the left and right end pixels of the connected pixel region obtained in the step 2-2-1. Pixel on each of the left and right
Step 2-2-2 of extracting a range having both ends of the pixel as a high-brightness character area, and examining a luminance value distribution in a certain vertical line in the color image, and locally examining luminance in the vertical line. A step 2-2-3 of extracting a range in which the value is lower by a preset value as compared with the peripheral part as a connected pixel area,
The absolute value of the luminance gradient in the vertical direction is maximized from among a plurality of pixels selected by a predetermined method existing in the vicinity of both left and right pixels of the connected pixel region obtained in the step 2-2-3. A pixel is obtained on each of the right and left sides, and a range having both ends of the pixel is extracted as a low-luminance character area, and a high level on each vertical line extracted in the second-2-2 level is extracted. The luminance character areas are collected to create a high luminance character area image as the entire image, and the low luminance character areas on each vertical line extracted in the stage 2-2-4 are collected to form the entire image. 2-2 for creating a low-brightness character area image
Stage 5, a stage 2-2-6 for detecting a pixel at an edge portion in the color image, a high-luminance character area image and a low-luminance character obtained in the stage 2-2-6. From the plurality of pixels constituting the character connected component in the area image,
And a step 2-2-7 of removing a pixel corresponding to a vertical edge portion in the color image obtained in step 6 is realized.

Alternatively, in the above-described character region extraction method, the two-dimensional data obtained in the steps 2-1 and 2-2 can be obtained.
The connected components in two binary images are compared, and the area, shape,
The connected components whose differences in position are within a preset allowable range are determined as character regions, and are left in the binary image, and the connected components whose differences in area, shape, and position exceed the preset allowable range are not included. A second to third step of determining the background noise portion and removing the binary image from the binary image is performed by superposing the obtained two binary images at the same position, and the corresponding two pixels are both connected components. If so, the corresponding pixel is stored as a provisional character pixel, and if one of the two corresponding pixels does not belong to the connected component, the corresponding pixel is stored as a non-provisional character area, and is composed of the provisional character pixel. A 2-3-1 stage in which an image including a connected component to be defined is defined as an AND image, and the obtained two binary images are aligned and overlapped, and two corresponding pixels are If at least one of them belongs to the connected component, If the corresponding pixel is stored as a provisional character area, and if the two corresponding pixels do not belong to a connected component, the corresponding pixel is stored as a non-provisional character area, and an image including a connected component composed of provisional character pixels is stored. Step 2-3-2-2, which is defined and created as an OR image;
And the OR image obtained in the 2-3-2nd step is aligned and superimposed, and among the connected components in the AND image, the connected component in the OR image and If the number of overlapping pixels is equal to or greater than a preset threshold value and the area of the connected component in the OR image having the overlap is within a preset range, the connected component in the AND image is determined as a character connected component. And leave the other connected components in the AND image as background noise areas.
Step 2-3-3 for removing from the D image, and step 2-3
In the AND image in which only the character connected component obtained in the step -3 remains, of the connected pixels in each line obtained by the binarization for each horizontal line obtained in the step 2-1 A 3-2-4: storing a connected pixel in which the number of connected pixels is smaller than a preset value as a character area, and storing a pixel in the AND image existing at the same position as the pixel as a character area pixel.
Is realized by using the steps (1) and (2).

Further, a program for causing a computer to execute the steps in the above-described character region extracting method is recorded on a computer-readable recording medium.

Also, in order to solve the above-mentioned problem, a character region extracting apparatus according to the present invention comprises: a color image input storage means for inputting and storing a color image in which characters are displayed; A luminance value similar region forming unit that has similar luminance values in the color image input by the unit and forms a plurality of adjacent pixels in the image space as a connected region; and a luminance value similar region forming unit. For a connected region image having similar luminance values, a saturation value similar region forming means for forming a plurality of pixels having similar saturation values in each region and forming adjacent pixels in the image space as a connected region, A brightness value obtained by the saturation value similar area forming means,
For connected area images with similar saturation values, within each area,
Hue values are similar, and a plurality of hue value similar area forming means for forming a plurality of adjacent pixels in a pixel space as a connected area, and only the connected area pixels obtained by the hue value similar area forming means are left in the image, A character area image storing means for storing as a character area image, an execution order of the color image input storage means, the luminance value similar area forming means, the saturation value similar area forming means and the hue value similar area forming means are controlled. Processing control means.

Alternatively, in the above-described character region extracting apparatus, the color image input storage means for inputting and storing a color image in which characters are displayed is provided with the same telop character in a color moving image such as a television broadcast image. Or a plurality of temporally continuous color images in which characters such as characters are displayed and stored, or a color moving image / still image in which one still color image in which characters are displayed is input and stored. When a plurality of color images in which the same character is displayed are input, the average value of the pixel values of the corresponding pixels at positions between the plurality of color images is set as the value of each pixel. Multi-frame averaging means for creating an average color image having and outputting as an input color image;
The color moving image / still image input storage means and the processing control means for controlling the execution order of the plural frame averaging means are provided.

Alternatively, in the above-described character region extracting apparatus, the luminance value similar region forming means for forming a plurality of pixels having similar luminance values in the color image and forming a plurality of adjacent pixels in the image space as a connected region may be provided. A horizontal line unit binarizing means for performing binarization of luminance in each horizontal line in, and obtaining a result as a binary image, and performing binarization of luminance in each vertical line in the color image, and Is compared as a binary image, and the connected components in two binary images obtained by the horizontal line unit binarization means and the vertical line unit binarization means are compared with each other. The connected components whose differences in area, shape, and position are within a preset allowable range are determined to be character regions, and are left in the binary image, and the differences in area, shape, and position exceed the preset allowable range. Connected components are background noise Binarizing result integrating means for determining from the binary image and removing from the binary image, processing for controlling the execution order of the horizontal line unit binarizing means, the vertical line unit binarizing means, and the binarizing result integrating means And control means.

Alternatively, in the above-described character region extracting apparatus, the horizontal line unit binarizing means for binarizing the luminance within each horizontal line in the color image and obtaining a result as a binary image is provided. A maximum brightness region detection means for examining a brightness value distribution in a certain horizontal line, extracting a range in which the brightness value is locally higher than a peripheral portion by a predetermined value as a connected pixel region in the horizontal line, The pixel having the maximum absolute value of the horizontal luminance gradient is selected from among a plurality of pixels selected by a predetermined method existing in the vicinity of both left and right pixels of the connected pixel region obtained by the maximum luminance region detecting means. A high-brightness character region extracting means for extracting a region having both ends of the pixel as a high-brightness character region, and examining a luminance value distribution in a certain horizontal line in the color image, A luminance minimum area detection unit that extracts a range in which the luminance value is lower by a preset value as compared with the peripheral part as a connected pixel area, and near the left and right end pixels of the connected pixel area obtained by the luminance minimum area detection unit. From among a plurality of pixels selected by an existing predetermined method, a pixel in which the absolute value of the luminance gradient in the horizontal direction is the maximum is determined for each of the left and right,
A low-luminance character area extracting means for extracting a range having both ends of the pixel as a low-luminance character area, and a high-luminance character area on each horizontal line extracted by the high-luminance character area extracting means are collected as an entire image. A high-brightness character area image, and low-brightness character areas on each horizontal line extracted by the low-brightness character area extraction means. Line-based binarization result integrating means, the luminance maximum area detecting means, the high luminance character area extracting means, the luminance minimum area detecting means, the low luminance character area extracting means, and the horizontal line-based binarizing result integrating means , An edge detecting means for detecting a pixel at an edge portion in the color image, a high-luminance character area image obtained by the horizontal-line-based binarization result integrating means, Brightness statement Edge removing means for removing a pixel corresponding to a vertical edge in a color image obtained by the edge detecting means from a plurality of pixels constituting a character connected component in the region image. And

Alternatively, in the above-described character region extracting apparatus, the vertical line unit binarizing means for binarizing the luminance within each vertical line in the color image and obtaining the result as a binary image is provided. A luminance maximum area detection means for examining a luminance value distribution in a certain vertical line, and extracting a range in which the luminance value is locally higher than a peripheral value by a preset value in the vertical line as a connected pixel area, The pixel having the maximum absolute value of the vertical luminance gradient is selected from among a plurality of pixels selected by a predetermined method existing in the vicinity of both left and right pixels of the connected pixel region obtained by the maximum luminance region detecting means. A high-brightness character area extraction unit that obtains each and extracts a range having both ends of the pixel as a high-brightness character area, and examines a luminance value distribution in a certain vertical line in the color image, A luminance minimum area detection unit that extracts a range in which the luminance value is lower by a preset value as compared with the peripheral part as a connected pixel area, and near the left and right end pixels of the connected pixel area obtained by the luminance minimum area detection unit. A pixel in which the absolute value of the luminance gradient in the vertical direction is the largest from among a plurality of pixels selected by an existing predetermined method is obtained for each of the left and right,
A low-luminance character area extracting unit that extracts a range having both ends of the pixel as a low-luminance character region, and a high-luminance character region on each vertical line extracted by the high-luminance character region extracting unit are collected to form an entire image. To create a high-brightness character region image of each of the vertical lines extracted by the low-brightness character region extraction means, and to create a low-brightness character region image as an entire image. Vertical line-by-line binarization result integration means, the luminance maximum area detection means, the high luminance character area extraction means, the luminance minimum area detection means, the low luminance character area extraction means, and the vertical line-by-line binarization result integration Processing control means for controlling the execution order of the means, edge detection means for detecting a pixel of an edge portion in the color image, and a high-luminance character area image obtained by the vertical line-based binarization result integration means, and Low brightness Edge part removing means for removing a pixel corresponding to a vertical edge part in a color image obtained by the edge extracting means from a plurality of pixels constituting a character connected component in the character area image. Features.

Alternatively, in the above character region extracting apparatus, the connected components in the two binary images obtained by the horizontal line unit binarization unit and the vertical line unit binarization unit are compared with each other, and the area and the shape are compared. The connected components whose differences in position are within a preset allowable range are determined to be character regions, and are left in the binary image, and the connected components whose area, shape, and position differ from each other in a preset allowable range are determined. Is a background noise part, and the binarization result integrating means for removing the binary image from the binary image is arranged such that the two obtained binary images are aligned and overlapped, and both corresponding pixels are connected components. , The corresponding pixel is stored as a provisional character pixel, and if any one of the two corresponding pixels does not belong to the connected component, the corresponding pixel is stored as a non-provisional character area, and is composed of the provisional character pixel. Connected component Image generating means for defining an image as an AND image, and superimposing the obtained two binary images at the same position, and any one of two corresponding pixels belongs to a connected component. In this case, the corresponding pixel is stored as a provisional character area, and when the two corresponding pixels do not belong to a connected component, the corresponding pixel is stored as a non-provisional character area, and the connected component formed of the provisional character pixel is stored. An OR image creating means for creating an image including the image as an OR image, and an AND image obtained by the AND image creating means and an OR image obtained by the OR image creating means, which are aligned and superimposed, and In the connected components in the image, the number of pixels overlapping with the connected components in the OR image is equal to or greater than a predetermined threshold, and the area of the connected components in the OR image having overlaps is set in a predetermined range. If, the AN
The connected component in the D image is determined as a character connected component.
In the ND image, the other connected components are removed from the AND image as a background noise area by a character connected component determining means, and in the AND image in which only the character connected components obtained by the character connected component determining means remain. Of the connected pixels in each line obtained by binarization for each horizontal line obtained by the horizontal line unit binarization means, a connected pixel whose connected pixel number is smaller than a preset value is defined as a character area, Character pixel registration means for storing a pixel in the AND image which is present at the same position as the pixel as a character area pixel.

As described above, the conventional method has a problem that the extraction of the color deteriorated portion at the end of the horizontal stroke is unstable. Therefore, in the present invention, a range in which the value is locally significant from the luminance distribution on the line is obtained using a predetermined luminance value threshold, and then the luminance gradient value at both ends of the obtained range is further determined. By defining the largest pixel as the end of the true character, the character color degraded part can be stably included in the character area and extracted, thereby smoothing the character outline shape as well as the original image, and breaking the stroke. Can be eliminated (see FIG. 10).

Furthermore, in the conventional method, the detection of the boundary between the character and the background can be guaranteed only in the reference direction in the line unit for each direction, and thus there is a problem that the extraction of the character region eventually fails. In the present invention, after performing line-by-line binarization in each direction, edge pixels in the original image are removed from connected components. Therefore, after the binarization in each direction, all boundaries between characters and the background are detected. Then, the character portion can be extracted as a connected component completely separated from the background. As a result, the binarized results for each line in each direction are finally superimposed, compared, and even in the integrated final result, the character portion has a common position, size, and shape for each result. It is possible to suppress the extraction failure of the character area (FIGS. 11 and 12).
reference).

[0023]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing a configuration of an apparatus according to an embodiment of the present invention and a processing flow.

In FIG. 1, reference numeral 1 denotes a color image input storage unit which inputs color image data displaying characters to be processed and stores it in a memory.

Reference numeral 2 denotes a luminance similar area forming unit which forms a plurality of pixels having similar luminance values in a color image input from the color image input storage unit 1 and adjacent to each other in an image space.

Reference numeral 3 denotes a saturation value similar region forming unit which has similar saturation values in each of the connected region pixels having similar luminance values obtained by the luminance value similar region forming unit 2 and has a similarity in the image space. To form a plurality of adjacent pixels as a connection region.

Reference numeral 4 denotes a hue value similar area forming unit, which has similar hue values in each of the connected area pixels having similar luminance and chroma values obtained by the saturation value similar area forming unit 3; A plurality of pixels adjacent in the image space are formed as a connected region.

Reference numeral 5 denotes a character region image storage unit, which stores the connected region image obtained by the hue value similar region forming unit 4 as a character region extraction result image.

Reference numeral 6 denotes a processing control unit which controls the execution order of each of the above-mentioned units 1 to 5.

FIG. 2 shows the color image input storage unit 1 shown in FIG.
FIG. 3 is a block diagram showing an example of a processing flow together with an example of the configuration of FIG.

In FIG. 2, reference numeral 7 denotes a color image input storage unit, which is a plurality of temporally continuous color images in which characters such as the same telop characters are displayed in a color moving image such as a television broadcast image. Is input and stored, or one still color image including characters is input and stored.

Numeral 8 denotes a multiple frame averaging unit, which is an average color image having, as a value of each pixel, an average value of pixel values of corresponding pixels at positions between a plurality of color images displaying the same character. Create

In general, in a color moving image such as a television broadcast image, a telop character is displayed at the same position on the screen over a plurality of frames. On the other hand, a general object other than the character display and a background portion often have movement.

For this reason, the pixel value of the corresponding pixel between a plurality of temporally consecutive color images in which the same telop character in a color moving image is displayed does not change over a plurality of frames in the character portion. Since the object moves in the background, the value changes over a plurality of frames.

Therefore, when an averaged color image having the average value of the corresponding pixels in the plurality of frames as the value of each pixel is created, the value in the character portion does not change over the plurality of frames, so that the value is almost unchanged after the averaging. Although the values do not change, various values are averaged in the background portion, so that the object outline and the like in the image become unclear and blurred.

This has the effect of suppressing the extraction of the area of the general object from the background in the subsequent binarization processing of the image.

Reference numeral 9 denotes a processing control unit, which controls the execution order of the units 7 to 8.

FIG. 3 is a diagram showing a luminance value similar area forming unit 2 in FIG.
FIG. 3 is a block diagram showing an example of a processing flow together with an example of the configuration of FIG.

In FIG. 3, reference numeral 10 denotes a horizontal line unit binarizing unit, which binarizes the luminance within each horizontal line in the image.

Reference numeral 11 denotes a vertical line unit binarizing unit, which binarizes the luminance within each vertical line in the image.

Reference numeral 12 denotes a binarization result integration unit which compares two binary images obtained by the horizontal line unit binarization unit 10 and the vertical line unit binarization unit 11, and both results are the same. The connected component existing at the position and having the same area value is determined to be a character region, and the other connected components are determined to be background noise portions and are removed from the image.

Reference numeral 13 denotes a processing control unit,
Control the execution order of each part.

FIG. 4 is a block diagram showing an example of the processing flow together with an example of the configuration of the horizontal line unit binarizing section 10 in FIG.

In FIG. 4, reference numeral 14 denotes a horizontal line maximum luminance area detecting unit which examines the luminance distribution in a certain horizontal line in the image, and locally increases the luminance value by a preset value in the horizontal line. A connected pixel area is detected.

Reference numeral 15 denotes a high-brightness character region extracting unit, which is located on a certain horizontal line in the image and includes a plurality of pixels near the left and right ends of the connected pixel region obtained by the on-horizontal-line maximum brightness region detecting unit 14. , A pixel having the maximum absolute value of the luminance gradient in the horizontal direction is obtained for each of the left and right sides, and a range having both ends of the pixel is extracted as a high luminance character area.

Reference numeral 16 denotes a horizontal line minimum luminance area detecting unit for examining the luminance distribution in a certain horizontal line in the image and detecting a connected pixel area in which the luminance value is locally lower by a preset value in the horizontal line. To detect.

Reference numeral 17 denotes a low-luminance character area extraction unit which is located on a certain horizontal line in the image and includes a plurality of pixels near the left and right ends of the connected pixel area obtained by the horizontal line minimum luminance area detection unit 16. , A pixel having the maximum absolute value of the luminance gradient in the horizontal direction is obtained on each of the left and right sides, and a range having both ends of the pixel is extracted as a low luminance character area.

Numeral 18 denotes a binarization result integrating unit for each horizontal line, which integrates the character regions on each horizontal line obtained by the high-luminance character region extraction unit 15 and the low-luminance character region extraction unit 17 to form an entire image. To create a character area image.

Reference numeral 19 denotes a processing control unit, which controls the execution order in each of the units 14 to 18.

Reference numeral 20 denotes an edge detection unit which detects a pixel at an edge in an image. For edge detection, for example, Robi
Use existing methods, such as using the nson operator.

Reference numeral 21 denotes an edge removing unit, which is composed of a plurality of pixels constituting a character connected component in the high-luminance character region image and the low-luminance character region extraction image obtained by the horizontal line binarization result integration unit 18. The pixel corresponding to the vertical edge portion in the average value image obtained by the edge detection unit 18 is removed from the image data.

FIG. 5 is a block diagram showing an example of the processing flow along with an example of the configuration of the vertical line unit binarizing section 11 in FIG.

In FIG. 5, reference numeral 22 denotes a vertical line maximum luminance area detecting unit for examining the luminance distribution in a certain vertical line in the image, and locally increasing the luminance value within the vertical line by a preset value. A connected pixel area is detected.

Reference numeral 23 denotes a high-brightness character region extracting unit which is located on a certain vertical line in the image and includes a plurality of pixels in the vicinity of both left and right ends of the connected pixel region obtained by the on-vertical-line maximum brightness region detecting unit 22. , A pixel in which the absolute value of the luminance gradient in the vertical direction is the maximum is determined for each of the left and right, and a range having both ends of the pixel is extracted as a high-luminance character area.

Reference numeral 24 denotes a vertical line minimum luminance area detecting unit which examines the luminance distribution in a certain vertical line in the image, and detects a connected pixel area in which the luminance value is locally lower by a preset value in the vertical line. To detect.

Numeral 25 denotes a low-luminance character area extracting unit, which is located on a certain vertical line in the image and includes a plurality of pixels near the left and right ends of the connected pixel area obtained by the vertical line minimum luminance area detecting unit 24. , A pixel in which the absolute value of the luminance gradient in the vertical direction is the maximum is determined for each of the left and right, and a range having both ends of the pixel is extracted as a low-luminance character area.

Reference numeral 26 denotes a vertical line-based binarization result integrating unit which integrates the character regions on each vertical line obtained by the high-luminance character region extraction unit 23 and the low-luminance character region extraction unit 25 to form an entire image. To create a character area image.

Reference numeral 27 denotes a processing control unit, which controls the execution order in each of the units 22 to 26.

Reference numeral 28 denotes an edge detection unit which detects a pixel at an edge in the image. For edge detection, for example, Robi
Use existing methods, such as using the nson operator.

Reference numeral 29 denotes an edge portion removing section, and a plurality of pixels constituting a character connected component in the high brightness character area image and the low brightness character area extracted image obtained by the vertical line binarization result integration section 26. From the average value image obtained by the edge detection unit 28, the pixel corresponding to the vertical edge portion is removed.

FIG. 6 is a block diagram showing an example of the flow of processing together with an example of the configuration of the binarization result integration section 12 in FIG.

In FIG. 6, reference numeral 30 denotes an AND image forming unit which superimposes two input binary images at the same position and, when both corresponding pixels belong to a connected component, the corresponding pixel. Is stored as a provisional character pixel, and if one of the two corresponding pixels does not belong to the connected component, the corresponding pixel is stored as a non-provisional character area, and the image including the connected component composed of the provisional character pixel is stored. (Here, this is defined as an AND image.)

Numeral 31 denotes an OR image creating section, which superimposes the input two binary images at the same position and superimposes them on the corresponding binary image.
If either one of the two pixels belongs to the connected component, the corresponding pixel is stored as a provisional character area, and the corresponding 2
If both of the pixels do not belong to a connected component, the corresponding pixel is stored as a non-provisional character area, and an image including a connected component composed of provisional character pixels (here, this is defined as an OR image) is created. I do.

Reference numeral 32 denotes a character connected component judging unit,
The AND image obtained by the image generating unit 30 and the OR image obtained by the OR image generating unit 31 are aligned and superimposed, and a pixel overlapping the connected component in the OR image among the connected components in the AND image. Is greater than or equal to a preset threshold, and if the area of the connected component in the overlapping OR image is within a preset range, the connected component in the AND image is determined as a character connected component.

Reference numeral 33 denotes a character pixel registration unit. In an AND image in which only the character connected component obtained by the character connected component determination unit 32 remains, each of the lines obtained by the horizontal line unit binarization unit 10 has Among the connected pixels, a connected pixel having a connected pixel number smaller than a preset value is set as a character area, and a pixel in the AND image located at the same position as the pixel is stored as a character area pixel.

FIG. 7 is a flowchart for explaining an embodiment of the processing of the maximum luminance area on horizontal line detecting section 14 and the high luminance character area extracting section 15 in FIG.

In FIG. 7, step (30) is a process of reading the luminance image into the memory.

Step (31) is a step of judging whether there is an unprocessed horizontal line in the luminance-similar character area extraction process. If there is an unprocessed horizontal line, the process proceeds to step (32). If there is no unprocessed horizontal line, the process ends.

In a step (32), a luminance distribution is checked for each horizontal line in the image obtained in the step (31).
This is a process of obtaining all the pixels having the maximum value.

Step (33) is a step of judging whether or not there is an unprocessed maximum pixel in the high-luminance character area extraction processing among the maximum pixels obtained in step (32). If the maximum value pixel exists, the process proceeds to step (34), and if there is no unprocessed maximum value pixel, the process proceeds to step (31).

In step (34), starting from the maximum value pixel obtained in step (33) as a starting point, the horizontal line is scanned one pixel at a time to the left and right, and the left and right sides of the maximum value are scanned. A process of detecting the first pixel whose luminance value is lower than the maximum value by a preset threshold value, and extracting the two detected pixels and the pixel existing between the two pixels on the horizontal line as a connected pixel area It is.

Step (35) is a process of obtaining the absolute value of the luminance gradient value for a plurality of pixels within a certain range on the left and right ends of the connected pixel area obtained in step (34).

The horizontal coordinate values of the left and right end points of the connected pixel area on a certain horizontal line obtained in step (34) are x and y, respectively (0 <x <y, where the coordinate value is the leftmost end of the line). Is set to 0, and thereafter, it is increased by 1 each time the pixel is shifted to the right by one pixel.) The fixed range in x and y is
X-r, x-r + 1, ..., x, ..., x + r-
1, x + r, and y-r, y-r + 1, ..., y, ...,
y + r-1, y + r. Note that r is a value set in advance.

The luminance value of the pixel within the fixed range at both ends is L
(X−r), L (x−r + 1),..., L (x),.
(X + r-1), L (x + r), and R (x-r),
R (x−r + 1),..., R (x),.
1), R (x + r), the absolute value of the luminance gradient value at each coordinate within a certain range at both ends is abs {L (x−r) −L (x−r−1)} for the left end. ,
..., abs {L (x) -L (x-1)}, ..., abs
{L (x + r) -L (x + r-1)}, with respect to the right end, abs {R (x-r) -R (x-r + 1)},.
abs {R (x) -R (x + 1)}, ..., abs @ R
(X + r) -R (x + r + 1)}. However, for each pixel at the left end, the absolute value of the luminance difference between only one pixel to the left and the pixel at the right end is calculated, and for each pixel at the right end, the absolute value of the luminance difference between only one pixel to the right is calculated. Let the value be the absolute value of the luminance gradient value at each pixel.

In step (36), the maximum value of the absolute values of the luminance gradient values obtained in step (35) is determined for each of the left and right sides, and if both are greater than the preset values, the flow proceeds to step (37). If either one is smaller than the preset value, the process proceeds to step (33).

The maximum value of the absolute value of the luminance gradient value of a pixel in a certain range at the left end of the pixel connection area obtained in step (35) is defined as max [abs ｛L (xr) -L (xr-
1) {, ..., abs {L (x) -L (x-1)}, ...,
abs {L (x + r) -L (x + r-1)}],
The maximum value of the absolute value of the luminance gradient value of a pixel in a certain range at the right end is defined as max [abs ｛R (x−r) −R (x−r +
1) {, ..., abs {R (x) -R (x + 1)}, ...,
abs {R (x + r) -R (x + r + 1)}].

Generally, there is a large difference between the luminance of a character on the character display screen and the luminance of the background in contact with the character.
A sharp luminance gradient exists at the boundary between the two. On the other hand, a sharp luminance gradient does not always exist at the boundary of a general object other than the characters displayed on the screen. For this reason, if the absolute values of the luminance gradient values at both the left and right ends of the connected pixel region obtained in step (35) are both larger than a preset value, the connected pixel region is set as a character region, and either one of them is set in advance. When the value is smaller than the value, by removing the background noise area, the extraction of the background noise area can be suppressed more effectively than in the related art, and the character area can be extracted with high accuracy.

The step (37) includes the pixels at both ends where the absolute value of the luminance gradient value obtained in the step (36) is larger than a preset value, and sets the pixels existing between the pixels as the high luminance character area. This is the process of determining and returning to step (33).

FIG. 8 is a flowchart for explaining an embodiment of the processing of the minimum luminance area detecting section 16 and the low luminance character area extracting section 17 on the horizontal line in FIG.

In FIG. 8, step (40) is a process of reading the luminance image into the memory.

Step (41) is a step of judging whether or not there is an unprocessed vertical line in the luminance-similar character area extraction process. If there is an unprocessed vertical line, the process proceeds to step (42). If there is no unprocessed vertical line, the process ends.

In a step (42), a luminance distribution is checked for each vertical line in the image obtained in the step (41).
This is a process of obtaining all the pixels having the minimum value.

Step (43) is a step of judging whether there is an unprocessed minimum-value pixel in the low-luminance character region extraction processing among the minimum-value pixels obtained in step (42). If the minimum value pixel exists, the process proceeds to step (44). If there is no unprocessed minimum value pixel, the process proceeds to step (41).

In step (44), the minimum value pixel obtained in step (43) is used as a starting point, and the vertical line is scanned one pixel at a time in the upward and downward directions. Detects the first pixel whose luminance value is higher than the minimum value by a preset threshold value, and extracts the detected two pixels and the pixel existing between the two pixels on the vertical line as a connected pixel area. It is a process.

Step (45) is a process of obtaining the absolute value of the luminance gradient value for a plurality of pixels within a certain range at the upper and lower ends of the connected pixel area obtained in step (44).

In step (46), the maximum value of the absolute value of the luminance gradient value obtained in step (45) is obtained in each of the upper and lower directions. If both are larger than the preset values, the process proceeds to step (47). If either one is smaller than the preset value, the process proceeds to step (43).

Step (47) includes the pixels at both ends where the absolute value of the luminance gradient value obtained in step (46) is larger than a preset value, and sets the pixels existing between the pixels as the low luminance character area. This is the process of determining and returning to step (43).

FIG. 9 shows the character connection component judging section 32 in FIG.
5 is a flowchart for describing an embodiment of the processing of FIG.

In FIG. 9, step (50) is AND
This is the process of reading the image and the OR image into the memory.

Step (51) is a process of adjusting the positions of the AND image and the OR image.

Step (52) is a process of initializing a variable n for a connected component number in the AND image to 1.

In step (53), the variable n is compared with the total number N of connected components in the AND image. If the value of n is equal to or less than N, the process proceeds to step (54). This is the process of ending the processing.

In step (54), the number Q of connected component pixels in the OR image overlapping the nth connected component in the AND image is determined.
_or (n) and the number Q _itand (n) of pixels having the _overlap in the n-th connected component in the AND image.

Step (55) consists of Q _and (n) and AN
A value (P _and (n) × th) obtained by multiplying the area value P _and (n) of the n-th connected component in the D image by a preset value th1
1), and a value (P _or (n) × th2) obtained by multiplying Q _or (n) and the area value P _or (n) of the connected component including the overlapping pixel in the OR image by a preset value th2. ), And
Q _and (n) are not less than P _and (n) × th1, and Q
_{If or} (n) is equal to _or larger than P _or (n) × th2, the process proceeds to step (56); otherwise, the process proceeds to step (57).

Step (56) is a process of determining the n-th connected component in the AND image as a character area. In general, characters in a video are regions having substantially the same area value in both the AND image and the OR image, while general objects other than characters are not necessarily obtained as regions having substantially the same area in both the AND image and the OR image. Step (56) in Step (55)
By satisfying the condition for (1), the area of the general object other than the character can be effectively removed, and the character area can be extracted with high accuracy.

Step (57) is a process of incrementing the variable n by 1 and returning to step (53).

FIG. 10 is a diagram for explaining a character area range on a line determined by the present invention. In the present invention, a range that is locally significant from the luminance distribution on the line is obtained using a predetermined luminance value threshold, and then the maximum luminance gradient value at both ends of the obtained range is obtained. By defining the pixel as the end of the true character, the character color degraded part can be stably included in the character area and extracted. As a result, it becomes possible to make the character outline shape as smooth as the original image, and to eliminate the discontinuity of the stroke generated by the conventional method.

FIG. 11 is a diagram showing an example of the binarization result according to the present invention. FIG. 12 is a diagram for explaining the details. In the conventional method, as shown in FIG. 16, the line unit for each direction cannot guarantee the detection of the boundary between the character and the background only in the reference direction. . In the present invention, after performing line-by-line binarization in each direction, edge pixels in the original image are removed from connected components. Therefore, after the binarization in each direction, all boundaries between characters and the background are detected. Then, the character portion can be extracted as a connected component completely separated from the background. As a result, the binarized results for each line in each direction are finally superimposed, compared, and even in the integrated final result, the character portion has a common position, size, and shape for each result. It is possible to suppress a failure in extracting a character area.

It is to be noted that some or all of the means shown in FIGS. 1 to 6 can be made to function using a computer, or the steps (steps) of the processing shown in FIGS. 1 to 9 can be performed by a computer. Needless to say, a program for causing a computer to function as the means or a program for executing the stage of the processing by the computer can be stored in a computer-readable recording medium such as an FD.
(A floppy disk), an MO, a ROM, a memory card, a CD, a DVD, a removable disk, and the like.

[0101]

As described above, according to the present invention,
When binarizing an image on a line-by-line basis, find the pixel with the largest luminance gradient at both ends of the part where the value of the luminance distribution on the line is locally significant, and use this as the end of the character range Thus, the effect of stably extracting the character area including the deteriorated portion at the end of the horizontal stroke can be obtained. Further, after binarization in line units for each direction, the edge portion is removed, so that the character and the background can be reliably separated, and the effect of suppressing the failure of character region extraction can be obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration and a processing flow of a character region extracting apparatus according to an embodiment of the present invention.

FIG. 2 is a block diagram showing an example of a processing flow together with an example of a configuration of a color image input storage unit in FIG. 1 in the embodiment.

FIG. 3 is a block diagram showing an example of a processing flow together with an example of a configuration of a luminance value similar area forming unit in FIG. 2 in the embodiment.

FIG. 4 is a block diagram showing an example of a processing flow along with an example of a configuration of a horizontal line unit binarization section in the luminance value similar area forming section in FIG. 3;

FIG. 5 is a block diagram showing an example of a processing flow together with an example of a configuration of a vertical line unit binarizing section in the luminance value similar area forming section in FIG. 3;

FIG. 6 is a block diagram showing an example of a processing flow together with an example of a configuration of a binarization result integrating unit of the luminance value similar area forming unit in FIG. 3;

FIG. 7 is a flowchart illustrating an example of processing contents of a luminance maximum value area detection unit on a horizontal line and a high luminance character area extraction unit in the horizontal line unit binarization unit in FIG. 4;

8 is a flowchart showing an example of processing contents of a luminance minimum value area detection unit on a horizontal line and a low luminance character area extraction unit in the horizontal line unit binarization unit in FIG. 4;

FIG. 9 is a flowchart showing an example of processing contents of a character connection component determination unit in the binarization result integration unit in FIG. 6;

FIG. 10 is a diagram showing an example of a character area range on a line determined by the present invention.

FIG. 11 is a diagram showing an example of a binarization result according to the present invention.

FIG. 12 is a diagram illustrating details of an example of a binarization result of FIG. 11;

FIG. 13 is a diagram illustrating an example of color degradation of characters in an NTSC format video.

FIGS. 14A and 14B are examples 1 of a binarization result by a conventional method.

FIGS. 15A and 15B are second examples of the binarization result by the conventional method.

FIG. 16 is a third example of the binarization result obtained by the conventional method.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Color image input storage part 2 ... Luminance similar area formation part 3 ... Saturation value similar area formation part 4 ... Hue value similar area formation part 5 ... Character area image storage part 6 ... Process control part 7 ... Color image input storage part 8: Multiple frame averaging unit 9: Processing control unit 10: Horizontal line unit binarization unit 11: Vertical line unit binarization unit 12: Binarization result integration unit 13: Processing control unit 14: Maximum luminance on horizontal line Area detecting section 15 High-luminance character area extracting section 16 Minimum luminance on-horizontal-line area detecting section 17 Low-luminance character area extracting section 18 Binary-result-by-horizontal-result integration section 19 Processing control section 20 Edge detecting section Reference Signs List 21: Edge removal unit 22: Maximum luminance area detection unit on vertical line 23: High luminance character area extraction unit 24: Minimum luminance area detection unit on vertical line 25: Low luminance character area extraction unit 26: Binarization for each vertical line result Integration unit 27 ... Process control unit 28 ... Edge detection unit 29 ... Edge removal unit

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Shoji Kurakake 3-19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo Japan Telegraph and Telephone Corporation (72) Inventor Toshiaki Sugimura 3- 192-1 Nishi-Shinjuku, Shinjuku-ku, Tokyo No. F-term in Nippon Telegraph and Telephone Corporation (reference) GA23 GA41 GA51 9A001 BB06 HH22 HH23 HH28 HH30 HH31

Claims (13)

[Claims] 1. A first step of inputting and storing a color image in which characters are displayed, wherein the color images input in the first step have similar luminance values and are adjacent in an image space. A plurality of pixels
And a first connected area image having similar luminance values obtained in the second step, the saturation values are similar in each area, and A third step of forming a plurality of pixels adjacent to each other as a second connected area, and a second connected area image similar in both luminance value and saturation value obtained in the third step. A fourth step of forming a plurality of pixels having similar hue values in the area and adjacent in the pixel space as a third connected area; and only the third connected area pixels obtained by the fourth step A fifth step of leaving in the image and storing as a character area image. 2. The character area extracting method according to claim 1, wherein the first step of inputting and storing a color image in which characters are displayed is performed in the same color moving image such as a television broadcast image. A first one in which a plurality of temporally continuous color images in which characters such as telop characters are displayed are input and stored, or a single still color image in which characters are displayed is input and stored. Step 1, and when a plurality of color images in which the same character is displayed are input, the average value of the pixel values of the pixels corresponding to the positions among the plurality of color images is used as the value of each pixel. A step 1-2 of creating and outputting an average color image as an input color image. 3. The method according to claim 1, further comprising the step of forming a plurality of pixels having similar luminance values in the color image and adjacent to each other in the image space as a first connected region. Performing a binary conversion of luminance within each horizontal line in the color image and obtaining a result as a binary image, and a binarization of luminance within each vertical line of the color image. Step 2-2 of obtaining the result as a binary image, Step 2-1 and Step 2 of the step 2-2.
The connected components in two binary images are compared, and the area, shape,
The connected components whose differences in position are within a preset allowable range are determined as character regions, and are left in the binary image, and the connected components whose differences in area, shape, and position exceed the preset allowable range are not included. Characteristic region extraction method characterized in that the second and third steps of determining as a background noise portion and removing from the binary image are performed. 4. The character area extracting method according to claim 3, wherein the luminance is binarized within each horizontal line in the color image,
The 2-1 step of obtaining the result as a binary image is set in advance by examining the luminance value distribution in a certain horizontal line in the color image and comparing the luminance value locally with the peripheral portion in the horizontal line. A 2-1-1 step of extracting a range higher by the calculated value as a connected pixel area, and a predetermined area near the left and right end pixels of the connected pixel area obtained by the 2-1-1 step. A pixel in which the absolute value of the luminance gradient in the horizontal direction is maximum is obtained on each of the left and right from a plurality of pixels selected by the method, and a range having both ends of the pixel is extracted as a high luminance character area. And examining the luminance value distribution in a certain horizontal line in the color image, and extracting a range in which the luminance value is locally lower in the horizontal line by a preset value as compared with the peripheral portion as a connected pixel area. Step 2-1-3 and Step 2-1-3 From among a plurality of pixels selected by a predetermined method existing in the vicinity of both left and right pixels of the connected pixel area obtained by the floor, a pixel in which the absolute value of the luminance gradient in the horizontal direction is the largest is determined for each of the left and right pixels. A step 2-1-4 of extracting a range having pixels at both ends as a low-luminance character area; and collecting a high-luminance character area on each horizontal line extracted in the step 2-1-2, A high-brightness character area image as an entire image is created, and a low-brightness character area image as an entire image is collected by collecting low-brightness character areas on each horizontal line extracted in the 2-1-4 stage. 2-1-5 to create
And detecting a pixel at an edge portion in the color image.
Step 6, and the high-luminance character area image obtained in Step 2-1-5,
And removing a pixel corresponding to a vertical edge portion in a color image obtained in the step 2-1-6 from a plurality of pixels constituting a character connected component in the low-luminance character area image. A method for extracting a character region, the method comprising: 5. The method according to claim 3, wherein the luminance is binarized within each vertical line in the color image.
The 2-2 stage of obtaining the result as a binary image is set in advance by examining the luminance value distribution in a certain vertical line in the color image, and locally comparing the luminance value with the peripheral portion in the vertical line. A 2-2-1 step of extracting a range higher by the calculated value as a connected pixel area, and a predetermined area near the left and right end pixels of the connected pixel area obtained by the 2-2-1 step. A pixel in which the absolute value of the luminance gradient in the vertical direction is maximum is obtained on each of the left and right from a plurality of pixels selected by the method, and a range having both ends of the pixel is extracted as a high-luminance character area. And the luminance value distribution in a certain vertical line in the color image is examined, and a range in which the luminance value is locally lower in the vertical line by a preset value compared with the peripheral portion is extracted as a connected pixel region. Step 2-2-3 and Step 2-2-3 From among a plurality of pixels selected by a predetermined method existing in the vicinity of both left and right pixels of the connected pixel area obtained by the floor, a pixel having the maximum absolute value of the vertical luminance gradient is obtained for each of the left and right pixels. Step 2-2-4 of extracting a range having pixels as both ends as a low-luminance character area, and collecting high-luminance character areas on each vertical line extracted in the step 2-2-2, A high-brightness character region image as a whole is created, and low-brightness character regions on each vertical line extracted in the stage 2-2-4 are collected to form a low-brightness character region image as the whole image. Step 2-2-5 to create, and Step 2-2 to detect a pixel at an edge portion in the color image
Step 6, and the high-luminance character area image obtained in Step 2-2-6.
And removing a pixel corresponding to a vertical edge portion in the color image obtained in the step 2-2-6 from a plurality of pixels constituting a character connected component in the low luminance character area image. 2-7. A method for extracting a character region, the method comprising: 6. The character area extraction method according to claim 3, wherein the connected components in the two binary images obtained in the step 2-1 and the step 2-2 are compared with each other to determine an area, The connected components whose differences in shape and position are within a preset allowable range are determined to be character regions, and are left in the binary image, and the connected components whose differences in area, shape and position exceed the preset allowable range. Are judged as a background noise part, and a second to third step of removing from the binary image is performed by superposing the obtained two binary images at the same position.
If the two corresponding pixels both belong to the connected component,
If the corresponding pixel is stored as a provisional character pixel and one of the two corresponding pixels does not belong to the connected component,
A second-3-1 step of storing the corresponding pixel as a non-provisional character area and defining an image including a connected component composed of the provisional character pixel as an AND image and creating the AND image; Superimpose the binary images of
If either one of the two corresponding pixels belongs to the connected component, the corresponding pixel is stored as a provisional character area, and if the two corresponding pixels do not belong to the connected component,
A second-third stage of storing the corresponding pixel as a non-temporary character area and defining an image including a connected component composed of the temporary character pixel as an OR image; AND image obtained by the step
The OR images obtained in the step of -3-2 are aligned and superimposed, and among the connected components in the AND image, the OR
If the number of pixels overlapping with the connected component in the image is equal to or greater than a preset threshold, and the area of the connected component in the OR image having the overlap is within a preset range, the connected component in the AND image Is determined as a character connected component, and the AND
The remaining connected components are removed from the AND image as a background noise region in the second and third steps, and only the character connected components obtained in the second and third steps remain. Of the connected pixels in each line obtained by the binarization for each horizontal line obtained in the 2-1 step, a connected pixel having a connected pixel number smaller than a preset value is set in the AND image. A step of storing a pixel in the AND image existing at the same position as the pixel as the character area as a character area pixel. 7. A color image input storage means for inputting and storing a color image in which characters are displayed, wherein the color images input by the color image input storage means have similar luminance values, and Means for forming a plurality of pixels adjacent to each other as a connected area, and a connected area image having similar brightness values obtained by the similar-valued area forming means. A saturation value similar area forming means for forming a plurality of adjacent pixels having a similar value in an image space as a connected area; and a luminance value and a saturation value obtained by the saturation value similar area forming means are similar. Hue value similar area forming means for forming a plurality of pixels having similar hue values in each area and adjacent in the pixel space as a connected area with respect to the connected area image, and obtaining the hue value similar area forming means. Linked Character area image storage means for storing only the area pixels in the image and storing it as a character area image; the color image input storage means; the luminance value similar area forming means; the saturation value similar area forming means; and the hue value similarity. A character region extracting apparatus, comprising: processing control means for controlling the execution order of the region forming means. 8. The character area extracting device according to claim 7, wherein the color image input storage means for inputting and storing a color image in which characters are displayed is the same as a color moving image such as a television broadcast image. A color moving image in which a plurality of temporally continuous color images in which characters such as telop characters are displayed are input and stored, or a single still color image in which characters are displayed is input and stored. Image / still image input storage means, when a plurality of color images displaying the same character are input, the average value of the pixel values of the pixels corresponding to the positions of the plurality of color images is calculated for each pixel Averaging means for creating an average color image having the values of the above and outputting as an input color image; and realization of the color moving image / still image input storage means and the plural frame averaging means. And processing control means for controlling the order, the character region extracting apparatus characterized by comprising. 9. The character region extracting device according to claim 7, wherein the luminance value similar region forming means for forming a plurality of pixels having similar luminance values in the color image and adjacent to each other in the image space as a connected region includes: Horizontal line unit binarization means for performing binarization of luminance in each horizontal line in the color image and obtaining a result as a binary image; and binarization of luminance in each vertical line in the color image. Vertical line unit binarization means for obtaining the result as a binary image, connected components in the two binary images obtained by the horizontal line unit binarization means and the vertical line unit binarization means Are compared, the connected components whose differences in area, shape, and position are within a preset allowable range are determined to be character regions, and are left in the binary image, and the differences in area, shape, and position are set in a predetermined allowable range. Connected components that exceed A binarization result integrating unit that determines that the image is a part and removes from the binary image; and controls an execution order of the horizontal line unit binarization unit, the vertical line unit binarization unit, and the binarization result integration unit. A character region extracting apparatus, comprising: a processing control unit. 10. The character area extracting device according to claim 9, wherein the luminance is binarized within each horizontal line in the color image.
The horizontal line unit binarization unit that obtains the result as a binary image examines a luminance value distribution in a certain horizontal line in the color image, and locally compares the luminance value in the horizontal line with a peripheral portion in advance. A maximum brightness region detecting means for extracting a range higher by the set value as a connected pixel region, and a plurality of pixels selected by a predetermined method existing near both left and right pixels of the connected pixel region obtained by the maximum brightness region detecting device. A high-brightness character area extracting means for obtaining a pixel having the maximum absolute value of the horizontal luminance gradient from the left and right pixels, and extracting a range having both ends of the pixel as a high-brightness character area; A luminance minimum area detection means for examining a luminance value distribution in a certain horizontal line, and extracting a range in which the luminance value is locally lower than a peripheral value by a preset value in the horizontal line as a connected pixel area; The pixel having the maximum absolute value of the horizontal luminance gradient is selected from among a plurality of pixels selected by a predetermined method existing in the vicinity of both left and right pixels of the connected pixel region obtained by the minimum luminance region detecting means. A low-brightness character region extracting means for extracting a range having both ends of the pixel as a low-brightness character region, and collecting a high-brightness character region on each horizontal line extracted by the high-brightness character region extracting means. A high-brightness character region image as an entire image is created, and low-brightness character region images as a whole image are collected by collecting low-brightness character regions on each horizontal line extracted by the low-brightness character region extraction means. A binarization result integrating means for each horizontal line, the luminance maximum area detecting means, the high luminance character area extracting means,
Processing control means for controlling the execution order of the minimum luminance area detecting means, the low luminance character area extracting means and the horizontal line-based binarization result integrating means; and edge detection for detecting a pixel at an edge portion in the color image. A high-brightness character area image obtained by the horizontal line-based binarization result integration means, and a color obtained by the edge detection means from a plurality of pixels constituting a character connected component in the low-brightness character area image. A character region extracting device, comprising: edge portion removing means for removing a pixel corresponding to a vertical edge portion in an image. 11. The character area extraction device according to claim 9, wherein the luminance is binarized in each vertical line in the color image.
The vertical line unit binarizing means for obtaining a result as a binary image examines a luminance value distribution in a certain vertical line in the color image, and locally compares the luminance value in the vertical line with a peripheral portion in advance. A maximum brightness region detecting means for extracting a range higher by a set value as a connected pixel region, and a plurality of pixels selected by a predetermined method existing near both left and right pixels of the connected pixel region obtained by the maximum brightness region detecting device. A high-brightness character area extracting means for obtaining, on each of the left and right, a pixel having a maximum absolute value of a vertical luminance gradient from among the pixels, and extracting a range having both ends of the pixel as a high-brightness character area; A luminance minimum area detection unit that examines a luminance value distribution in a certain vertical line in the vertical line, and extracts a range in which the luminance value is locally lower in the vertical line by a predetermined value as compared with a peripheral part as a connected pixel area; The pixel having the maximum absolute value of the vertical luminance gradient is selected from among a plurality of pixels selected by a predetermined method existing in the vicinity of both left and right pixels of the connected pixel region obtained by the minimum luminance region detecting means. A low-brightness character region extracting means for extracting a range having both ends of the pixel as a low-brightness character region, and collecting a high-brightness character region on each vertical line extracted by the high-brightness character region extracting means. A high-brightness character region image as an entire image, and a low-brightness character region image as an entire image by collecting low-brightness character regions on each vertical line extracted by the low-brightness character region extraction means. The vertical line-based binarization result integrating means for creating, the maximum luminance area detecting means, the high luminance character area extracting means,
Processing control means for controlling the execution order of the luminance minimum area detecting means, the low luminance character area extracting means, and the vertical line-based binarization result integrating means; and edge detection for detecting an edge pixel in the color image. Means, and a color obtained by the edge extracting means from a plurality of pixels constituting a character connected component in the high brightness character area image obtained by the vertical line-based binarization result integrating means and the low brightness character area image. A character region extracting apparatus, comprising: edge portion removing means for removing a pixel corresponding to a vertical edge portion in an image. 12. The character region extracting apparatus according to claim 9, wherein the connected components in the two binary images obtained by the horizontal line unit binarization unit and the vertical line unit binarization unit are compared with each other, The connected components whose differences in area, shape, and position are within a preset allowable range are determined to be character regions, and are left in the binary image, and the differences in area, shape, and position exceed the preset allowable range. The binarization result integrating means for judging the connected components as the background noise part and removing the binarized image from the binary image, superimposing the obtained two binary images at the same position,
If the two corresponding pixels both belong to the connected component,
If the corresponding pixel is stored as a provisional character pixel and one of the two corresponding pixels does not belong to the connected component,
AND image creating means for storing the corresponding pixel as a non-temporary character area, defining an image including a connected component composed of the temporary character pixel as an AND image, and creating the AND image, Align and overlap,
If either one of the two corresponding pixels belongs to the connected component, the corresponding pixel is stored as a provisional character area, and if the two corresponding pixels do not belong to the connected component,
OR image creating means for storing the corresponding pixel as a non-temporary character area and defining an image including connected components composed of temporary character pixels as an OR image, and an AND image obtained by the AND image creating means And the O
The OR images obtained by the R image creating means are aligned and superimposed on each other, and among the connected components in the AND image, the OR
If the number of pixels overlapping with the connected component in the image is equal to or greater than a preset threshold, and the area of the connected component in the OR image having the overlap is within a preset range, the connected component in the AND image Is determined as a character connected component, and the AND
Character connected component determining means for leaving the remaining connected components in the image as a background noise area from the AND image; and in the AND image in which only the character connected components obtained by the character connected component determining means remain, Of the connected pixels in each line obtained by binarization for each horizontal line obtained by the horizontal line unit binarization means, a connected pixel having a connected pixel number smaller than a preset value is set as a character area, A character pixel registration unit that stores a pixel in the AND image existing at the same position as the pixel as a character region pixel. 13. A program for causing a computer to execute the steps of the character area extracting method according to claim 1 on a recording medium readable by the computer. A recording medium on which is recorded a character area extraction method.