JPH02187864A - Picture retrieval information forming device - Google Patents

Abstract

PURPOSE:To automatically form picture retrieval information with high objectivity and to save labor by making set area information and the physical feature of the area information into the picture retrieval information. CONSTITUTION:The features of the pixel data of areas A and B set in an area setting means 2 are extracted by a feature quantity extracting means 3, and the feature data and the area information are made into the retrieval information. In a feature data setting means 4, when each color component quantity is larger than a previously set certain threshold, it is judged that the component exists. Consequently, the picture retrieval information can be automatically formed and input work can be reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to an image retrieval method for an image filing device that uses an optical disk or the like as a storage medium, stores digital image data, and can obtain necessary image data from the storage medium. It is related to.

Conventional technology In recent years, with technological advances such as increased capacity of storage media such as magnetic disks and optical disks, 2 higher resolution of displays, 2 faster image processing, and higher integration of semiconductor circuits, conventional large-capacity data It has now become possible to easily realize an image data filing device, which has been considered to be an ideal device. In addition, the demand for such devices is increasing, and document file devices are already widely used for binary images such as document images. Several companies are also developing and selling multilevel image filing devices. In conventional image file devices, the stored image data and the search data for retrieving the image data are, for example, "input date", "image input user group", etc., or when a person looks at the image data and enters a key. The search keywords are information extracted from the image, such as "name of person in image (Mr. XX)" and "location of image (△△ coast)."

That is, the keyword may not be directly related to the content of the image data, or may reflect the subjectivity of the person who inputs the keyword. Conventional search involves searching using such keywords.
The desired image is retrieved from the file device by keyword search.

In a keyword search, if you have reliably memorized the keyword associated with an image, you can enter that keyword and search. Also, if your memory is vague,
Input several keywords that are thought to be associated with image data, perform logical sums, logical results, etc., narrow down the image data, and finally obtain the desired image. By using keywords effectively like this, you can search smoothly.

I can do it. However, if you forget the exact keyword, you have to rely on your little memory and do some thinking and error. Also, if the image you want is not specific, for example, landscape images such as ``scenery with the sea'' or ``scenery of sunset'', and design images such as ``reddish image'' or ``image with large pattern'', etc. If you want something, there is no problem as long as you enter the keyword corresponding to your request in advance.
If not, it will be very difficult. In this way, in conventional searches using keywords, images are searched for in parts that are not directly related to the content of the images, and the keywords may become meaningless if the search purpose is different. In other words, the image data itself does not effectively utilize the visual information that is most compatible with humans.

Problems to be Solved by the Invention As described above, search information in conventional keyword searches is influenced by the subjectivity of the key information inputter.

Therefore, if the searcher's intent is different from the person entering the key information,
There is a problem that searching becomes difficult. There is still the issue that inputting search keywords is time-consuming.

In view of this, an object of the present invention is to provide an image search information generation device that automatically generates objective image search information.

Means for Solving the Problems The present invention provides means for setting a part or all of an area of image data, and a means for setting a part or all of an area of image data, and a means for setting a part or all of image data.
It has means for extracting physical features of more than one type, and uses the set area information and its physical features as image search information.

Other inventions include <1rI of part or all of image data.
means for setting a region, means for extracting one or more types of physical features from pixel data included in the set region, and means for converting the representation of the physical features, and The conversion result is used as image search information.

Another invention provides means for setting a part or all of the area of image data, and means for extracting one or more types of physical characteristics from pixel data included in the area set by means 1 and 2. means for dividing image data including pixel data within the region into a plurality of blocks; means for extracting one or more types of physical characteristics of the block from the pixel data within the block; and means for obtaining the physical characteristics from the physical characteristics of the block containing the data included in the area.

Effect: By the method described above, objective image retrieval information suitable for the image can be automatically generated from the image data itself.

Example This example is the first example. This embodiment extracts features from image data of a certain set area and automatically generates image search information.

This will be explained below using figures. FIG. 1 shows a processing block diagram, FIG. 6 shows an example of setting image data and areas, and an example of segmentation, and FIG. 7 shows an example of feature data.

In FIG. 1, 1 is image data, and the original data is R
, G, and H. 2 is a region setting means, 3 is a feature extraction means, 4 is a feature data setting means, and 5 is image search data. All or part of the image data 1 is set by the area setting means 2. An example is shown in FIG.

The figure shows a diagram in which a partial area of the entire image is set.

A and B are area names, respectively. The blocks in the figure are not used in this embodiment. The feature amount of the pixel data of the regions A and H set by the region setting means 2 is extracted by the feature amount extraction means 3, and the feature data and the region information are used as search information. The feature extracting means will be explained below. Here, the feature amount is color component information.

Extract in the following order.

■ R and G of each pixel data within the setting area.

Add the data values of each color component of B.

■ The result of the addition is used as the feature amount of the area.

As described above, the color component information within the block is determined, and the feature data setting means 4 in FIG. 1 uses the information to determine the color component information. The setting method is ft, G, B obtained as feature quantities.
If each color component information amount is larger than a preset threshold, it is determined that the block has that color component, and the block is set.

As described above, color component information can be obtained from the set area. If the feature values are described using this result, the information will be as shown in FIG. 7, for example. In Figure 7,
Display colors are roughly classified into eight types depending on the composition of each color component. In other words, if the amount of each color component is greater than a preset threshold, it is determined that that component is present, and the display colors are classified into display colors as shown in Figure 7 depending on the presence or absence of R, G, and H components. .

For example, the color components of a certain setting area are (R=1.G=O.

B=0), the set area information and its color component information are set to be "red" as search information.

According to the present invention, image search information can be automatically generated and input work can be reduced. However, the search information still has objectivity and can be searched without depending on the subjectivity of the person inputting the search information.

The second example will be described below.

This extracts features from image data in a certain set area and automatically generates image search information.

It also improves the accuracy of extracted information.

This will be explained below using figures. FIG. 3 shows a processing block diagram, FIG. 6 shows an example of setting image data and regions, and an example of block division, and FIGS. 7 and 8 show examples of feature data.

In FIG. 3, 1 is image data, and the original data is R
, G, and B. 2 is a region setting means, 8 is a block dividing means, 9 is an intra-block feature extraction means, 6 is an intra-region feature data setting means, and 6 is image search data. The image data 1 is divided by the area setting means 2, and the image data including the set area is divided into 18 blocks by the block dividing means 8.
Divide into blocks of x16 pixels or 8 x 8 pixels. In this embodiment, the image is divided into blocks of 8×8 pixels. From the pixel data in the divided blocks, we will first consider the block feature amounts, here color component information. The following two methods are shown as examples of methods for extracting the feature amount.

(iv) If the data value of each color component of R, C, and B for each pixel data exceeds a preset threshold value, that pixel is assumed to have that color component, and the number of pixels is counted, and the counting result is calculated. Determine the color components of the block.

(B) The data values of each R, G, and B color component of each pixel data are added, and the color component of the block is determined based on the addition result.

Furthermore, frequency component information of the block can be considered as the feature amount of the block. In that case, if the pixel data in the block is R, G, B color data, convert that data into luminance data and chrominance data. If it is separated into data, each luminance data is used as is.

Furthermore, if the image data is a black and white image, the data as it is is used to extract the feature quantity of the frequency component. Below, if it is a color image, brightness data, if it is black and white image data,
Extract the feature amount of the frequency component using the data as it is. In this case, frequency components are extracted using image processing, here discrete cosine transform (hereinafter referred to as DCT), on the image data. An example of DCT is shown in FIG. In the figure,
The DC component of the transform coefficient indicates the average value of the block,
The frequency component of the block is determined based on the other AC components. (1) is when there is little change in the image data, that is, mostly low frequency components, (2) is when the image data is changing rapidly and there are many high frequency components, and (3) is when the image data changes are mostly low frequency components. (4) shows a case where there are many changes in the horizontal direction such as a vertical striped pattern, and (4) shows a case where there are many changes in the vertical direction such as a horizontal striped pattern among the changes in the image.

In the figure, the shaded portions indicate coefficients with large absolute values. By performing DCT on pixel data in this way, frequency components can be easily extracted. An example of a method for extracting the feature amount will be shown below.

(8) After applying DCT to the pixel data, remove the DC component of the transform coefficient, add the absolute value or square value of the AC component, and determine the frequency component based on the addition result. In order to determine (3) and (4) in Figure 4, we focus on the transform coefficient values in the upper and left regions of the transform block, respectively, and calculate the absolute value or square of the transform coefficient value in that region. Add the values and determine the frequency component based on the addition result.

As described above, the color component information and frequency component information within the block are determined, and based on this information, the intra-area analysis 6 in FIG. 1 is performed. An example of that analysis is shown below.

A If the information amount of each R, G, B color component obtained from each block, the overall AC component amount of frequency components, and the AC component amount of each setting area (upper, left) are each larger than the preset threshold value. , assuming that the block has the color component, count the number of pixels, and determine the color component of the area based on the counting result.

CB) Count the amount of R, G, and B color component information obtained from each block, the overall AC component amount of the frequency component, and the AC component amount of each setting area (top, left), and use the counting results to calculate the Determine the color components of the region.

As described above, color component information is obtained from the set area.

Frequency component information can be obtained. Based on this result, when the feature amount is described, the information becomes as shown in FIGS. 7 and 8, for example. In FIG. 7, for example, the color components of a certain setting area are (R=1.G=o.

If E=o), the set area information and its color component information will be "Red 1" as the search information. In addition, in Figure 8, it is classified into six types depending on the frequency component. If the amount of each AC component in the upper part or left part of the block is larger than a preset threshold, it is determined that the component data exists, and the area is classified based on the result.For example, if the AC component data is (all Area=0. Upper area=0.
If the left area=0), the set area information and its frequency component information will be 1, which has the most low frequencies, and will be the search information.

As described above, according to the present invention, image search information can be automatically generated and input work can be reduced. Moreover, the search information is objective, and the search can be performed without depending on the subjectivity of the person inputting the search information.

Next, a third embodiment of the present invention will be described.

This automatically generates objective image search information and makes the user's search more flexible.

This will be explained below using figures. Figures 2 and 4 show processing block diagrams, Figure 6 shows examples of image data and area settings, and block division examples, Figures 7 and 8 show examples of feature data, and Figure 9 shows feature data. An example of each conversion is shown below.

In FIG. 4, 1 is image data, and the original data is R
, G, and B. 2 is a region setting means, 8 is a block dividing means, 9 is an intra-block trace amount extraction means, 4 is an intra-region feature data setting means, 5 is image search data, and 6 is the feature data extracted from the region used to represent the image content. thing(
For example, "mountain".

7 is conversion knowledge data for the data conversion means e (for example, "Blue 1+"Low Frequency'--Sky, Sea)".

The image data 1 is divided by the area setting means 2, and the image data including the set area is divided by the block dividing means 8, and the features of the set area are extracted using the obtained feature amount (color component information 2 frequency acid content information). The process is the same as the second embodiment, so it will be omitted.

Here, parts different from the second embodiment will be explained. This embodiment is characterized by converting the feature data of a certain area obtained in the second embodiment.

A part of the conversion example is shown in FIG. In the figure, the color component determined to be "blue" and the frequency component determined to be "many low frequencies" are converted into search information such as "Sky 1" or "°Sea", for example. Such conversion information is stored in the conversion knowledge data 7, and the data conversion means 6 converts it into an expression that represents the image content shown in the conversion knowledge data 7. By modifying it to suit the purpose, it is also possible to generate search information that is easier to use. Above, we have described the case where the image data containing the image data of the set area is divided into blocks, but in the case where the image data including the image data of the set area is not divided into blocks, the embodiment shown in FIG. For example, by calculating the difference in pixel data between adjacent pixels in a region and creating a histogram of the difference data, the degree of change in the image data can be known. Taking advantage of that, for example, (“little change” + “blue 1 → sky.”

It is also possible to obtain search information by converting it like ``Umi''.

Effects of the Invention According to the present invention, highly objective image search information can be automatically generated, and when added to conventional searches, more flexible searches can be performed. Furthermore, by increasing the knowledge of conversion data, more intelligent searches become possible.

[Brief explanation of the drawing]

FIGS. 1, 2, 3, and 4 are block diagrams of an image search information generation device according to an embodiment of the present invention, and FIG. 6 shows an example of image data, area setting, and block division of the same device. FIG. 6 is a diagram showing an example of orthogonal transformation of the same device, FIG. γ, and FIG. 8 are diagrams showing an example of feature S data of the same device. It is a figure which shows an example. 1... Image data, 3... Feature amount extraction means, 2... Area setting means, 4... Feature data setting means, 6... ...Image search data, e
...Data conversion means, completion...Conversion knowledge data, 8...Block division means, 9...
...Means for extracting a small amount within the block. Name of agent Patent attorney Shigetaka Awano and 1 other person Drawing I Elephant Qi 9 3rd Drawing Image Qi 9 Image search take Image search take No. 2 Drawing image data Figure 4 Jin image Cheetah image search Cheetah image search data condolence Zushozu Funeral Zuzu Zuzu Kazu (1) Bunkanchi 9

Claims (8)

[Claims] (1) It has means for setting a part or all of the area of image data, and means for extracting one or more types of physical characteristics from pixel data included in the set area, and the set area information and its An image search information generation device characterized by using physical characteristics as image search information. (2) means for setting part or all of the area of image data, means for extracting one or more types of physical features from pixel data included in the set area, and means for converting the expression of the physical features An image search information generation device characterized in that the image search information generation device has the above-mentioned set area information and its conversion result as image search information. (3) The means for extracting one or more types of physical features from pixel data included in a set area according to claims 1 and 2, is a means for dividing image data including pixel data in the area into a plurality of blocks. , a means for extracting one or more types of physical characteristics of a block from pixel data in the block, and a means for extracting one or more types of physical characteristics of the set area from the physical characteristics of the block containing data included in the area. An image retrieval information generating device that replaces a means for obtaining information with a means consisting of. (4) means for obtaining the physical characteristics of the set area from the physical characteristics of the blocks included in the area, means for calculating the sum of the quantities of the physical characteristics of the blocks included in the area;
4. The image retrieval information generating apparatus according to claim 1, further comprising means for determining the presence or absence of a physical feature in the set area based on the sum thereof. (5) Means for obtaining the physical characteristics of a set area from the physical characteristics of blocks included in the area, means for determining whether or not a block included in the area has a physical feature based on the physical feature amount of the block. and a means for counting the number of blocks having a physical characteristic, and a means for determining whether or not the set area has a physical characteristic based on the counted value. Image search information generation device. (6) Image search information generation according to claims 1, 2 and 3, wherein the means for extracting the physical characteristics of the block from the pixel data in the block includes means for performing image processing on the pixel data. Device. (7) The image search information generating device according to claim 6, wherein the image processing means is a discrete cosine transform. (8) The image search information generation device according to any one of claims 1, 2 and 3, wherein the physical feature includes at least one of color component information and frequency component information.