JPH0520455A - Picture processing method - Google Patents

Abstract

PURPOSE:To convert the color of an area showing specified information included in a picture. CONSTITUTION:The picture is made into mosaic (step 1), a picture element is binarized (step 2), and the area is enlarged so as to generate a candidate area (step 4). The form of the candidate area is restricted (step 5), and it is judged whether the candidate area shows specified area or not (step 10). The color is converted (step 12) and a picture processing is executed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method for converting a color within a region showing specific information contained in an input image.

[0002]

2. Description of the Related Art Conventionally, a candidate area for specific information from an image,
For example, as a method of selecting candidate regions of human faces,
First, depending on whether the input image information is a flesh color or not, it is binarized as 1 if it is flesh color and 0 if it is not flesh color, and the value of the adjacent pixel of the binarized image information is 1. Then, there is a method of growing a region and forming a cluster of pixels having a value of 1. Further, a technique of mosaicing a face image and inputting it to a neural network to identify faces of a large number of people is disclosed in Proceedings of the Autumn Meeting of the Institute of Electronics, Information and Communication Engineers, 1990, P6-409. further,
Proceedings of the 1990 Autumn Meeting of the Institute of Electronics, Information and Communication Engineers P
6-415 discloses a technique of searching for a person by extracting features from an input face image.

[0003]

SUMMARY OF THE INVENTION In the above conventional technique,
It is known in advance that a face image will be input as a candidate area for specific information, and it is necessary for a human to determine in advance whether or not the image includes a face. Furthermore, it is used when selecting a candidate area for a face. In the area growing method, the shape of the candidate area grows to an arbitrary shape and the shape is not limited, so such a method automatically recognizes a person's face, which is specific information, for an arbitrary image. It was impossible to do.

[0004]

[Means for Solving the Problems] A step of mosaicing input image information, and a binary value by discriminating the image information into a specific color and a color other than the specific color based on the mosaiced image information. Conversion, growing a candidate region based on binarized data, limiting the candidate region to a specific shape, and determining whether the candidate region shows specific information. The image processing is performed by the step and the step of performing the image processing for converting the color in the area determined to be the specific information. In addition, the specific shape is a rectangle, and it is determined by using a neural network whether or not the candidate area indicates specific information, so that the specific information looks preferable in the area determined to be the specific information. Convert to color.

[0005]

According to the present invention, the color can be converted within the area indicating the specific information included in the input image information.

[0006]

In the embodiment of the present invention, only the face of a person is extracted from the input image information as an area showing specific information, and color conversion is performed so that the flesh color of the face can be seen favorably.
Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows a flowchart in the embodiment of the present invention. First, R from the video camera, scanner, etc.
In the case of the present embodiment, the image information input as 8-bit signals for G and B is assumed to be 1280 dots × 1024 lines corresponding to a general high-resolution display, and is about 4 megabytes. This image information is 10 in step 1
A process of mosaicting into 0 dots × 100 lines is performed.
As a result, the amount of data (pixels) becomes about 1/131, and the processing speed is increased.

Next, in step 2, a skin color is selected from the R, G and B values of the image information mosaiced in step 1,
The pixel is binarized. To binarize a pixel,

[0009]

[Equation 1]

It is determined according to the above, and those that correspond to this are set to 1, and those that do not correspond to 0.

Next, in step 3, if there is a pixel which is judged to be a skin color in step 2 and has a binarized value of 1, the process proceeds to step 4, and otherwise proceeds to step 13.

Next, in step 4, the pixels determined to be flesh-colored are expanded based on the simple area expansion method to be candidate areas.

Here, the simple area expansion method in step 4 will be described with reference to FIG. 2. In step 41, a pixel whose binarized value is 1 which does not belong to any area in the image is searched.

Next, at step 42, it is judged whether or not there is a pixel whose newly binarized value is 1. 2 new
If there is a pixel whose binarized value is 1, the process proceeds to step 43, and if not, the process proceeds to step 44, and the process of expanding the region is completed.

Next, in step 43, the binarized value 1 of the pixel is compared with the binarized value of a pixel which does not yet belong to any region in the vicinity thereof, and the value is 1 If there is a pixel, the process proceeds to step 45. Otherwise step 41
Return to.

In step 45, the pixels found in step 43 are integrated into one area and a label is added,
Go to step 46.

Next, in step 46, attention is paid to the pixels newly integrated in step 45. And step 4
Returning to step 3, the operation in step 43 is repeated. If there is no new pixel to be integrated, the process returns to step 41 to search for a new pixel. In this way, the process is repeated until there is no pixel whose binarized value is 1 that does not belong to any region, and the region determined to be skin color is expanded to be a candidate region.

As described above, the candidate area obtained in step 4 is input to the neural network in order to determine whether or not it is an area showing a face. It has an arbitrary shape and cannot be input to the neural network. Therefore, step 5
Then, the shape of the candidate area is limited to a rectangle. A specific example will be described with reference to FIG. In the figure, the area indicated by a circle is a candidate area obtained by the area expansion method. For this candidate region, the rectangle shown by the largest diagonal line that overlaps at a predetermined rate (here, 75%) or more was selected. That is, when the number of pixels in the rectangle is n and the number of pixels in the candidate region included in the rectangle is P, the candidate region for which it is determined whether or not the region is a face is

[0019]

[Equation 2]

Is determined by

Next, in order to determine whether or not the candidate area obtained in step 5 is an area showing a face, the data of the candidate area is input to the neural network.
The following conditions were set when inputting the data of the candidate area.
That is, in step 6, if the number of pixels on one side of the candidate area is less than 10 pixels, it is determined that the area does not represent a face, and the process proceeds to step 7, and if it is 10 pixels or more, the process proceeds to step 8.

At step 7, attention is paid to the new candidate area, and if there is a new candidate area, the process returns to step 5, otherwise to step 9.

In step 9, it is determined that the area is a face area, and if there is an area in which a flag is set, step 12
If not, go to step 13.

In step 8, if the number of pixels on one side of the rectangle of the candidate area is 10 pixels or more, the candidate area is converted into 10 pixels by linear interpolation after converting from RGB 8-bit data to an 8-bit brightness value. Normalized to 10 pixels and input to the neural network in step 9.

The neural network to which the data of the candidate area is input in step 10 will be described. The neural network is a three-layer backpropagator that receives an 8-bit brightness value (black and white data) of 10 × 10 = 100 pixels and outputs a coefficient f (0 ≦ f ≦ 1) indicating whether or not it is a face area. -A neural network is used. The learning of the three-layer structure back propagation neural network is performed by inputting several kinds of standard face data as teacher signals in advance, and the three-layer structure back propagation is made to minimize the square error with the teacher signals.
This is a modification of the internal coefficient of the neural network. Therefore, the learned neural network can be used as a black box with 100 inputs and 1 output.

In step 10, the data of 10 × 10 candidate regions are input to the three-layer structure backpropagation neural network described above and the output f is obtained. Is the candidate region a face region? The final decision as to whether or not to make a decision is made by comparison with the set threshold value. That is, the threshold is Th
And when

[0027]

[Equation 3]

It is determined whether or not the candidate area is an area showing a face, and if it is judged that the candidate area is an area showing a face, step 1
If the area is not judged to be a face area, the process returns to step 7 to focus on a new candidate area.

Next, at step 11, a flag is set in the area determined to be the area showing the face. And step 7
Proceed to.

Next, in step 7, if there is no new candidate region to be noticed, the process proceeds to step 9, and if there is a flagged region in step 9, the process proceeds to step 12.
If not, proceed to step 13.

In step 12, image processing is performed so that the flesh color can be seen favorably in the flagged area. Then, the process proceeds to step 13 to end the image processing.

After that, the image is output from a video printer, a copying machine, a television receiver, or the like.

By performing the image processing as described above, an area (specifically, a human face) indicating specific information is selected from the image, and the color of the area indicating the specific information is converted (for example, specific area). An image can be output from a video printer, a copying machine, a television receiver, or the like by performing image processing so that information can be viewed in a desirable color.

[0034]

According to the present invention, it is possible to perform image processing for selecting a region indicating specific information from an image from input image information and converting a color of the region indicating the specific information. Further, the color conversion is converted into a color in which specific information is preferably viewed so that the image after image processing can be better viewed.

[Brief description of drawings]

FIG. 1 is a flow chart for explaining an embodiment of the present invention.

FIG. 2 is a flowchart for explaining a region expansion method in the embodiment of the present invention.

FIG. 3 is a diagram for explaining the limitation of the shape of a candidate area according to the embodiment of the present invention.

[Explanation of symbols]

1 Step of mosaicing an image 2 Step of binarizing pixels 4 Step of growing a candidate region 5 Step of limiting a candidate region to a specific shape 10 Step of determining whether a candidate region shows specific information Steps of image processing to convert 12 colors

Claims (4)

[Claims] 1. A step of mosaicing input color image information, and determining the color image information into a specific color and a color other than the specific color based on the mosaiced color image information. Binarizing step, and the above 2
Growing a candidate region based on the binarized data; limiting the candidate region to a specific shape; determining whether the candidate region indicates specific information; The image processing method comprising the step of performing image processing for converting the color of the area determined to be the information. 2. The image processing method according to claim 1, wherein the specific shape is a rectangle. 3. The image processing method according to claim 1, wherein whether or not the candidate area indicates specific information is determined by using a neural network. 4. The image processing method according to claim 1, wherein image processing is performed to convert the color into a color in which the specific information is preferably viewed in the area determined to be the specific information.