JPH05216992A - Picture clipping device - Google Patents

Abstract

(57) [Abstract] [Object] An object of the present invention is to provide a device for cutting out an image composed of a pattern made up of various colors and combinations of patterns without setting troublesome cutting conditions. [Structure] The image data input from the image input means a is
The spatial frequency conversion means b detects changes in the strength and weakness of the spatial frequency for each local region, and the clustering means d detects the similarity of the strength and weakness changes in the spatial frequency and performs clustering. The area-divided image generation means e groups together local areas having similar spatial frequency characteristics based on the clustering result. When a region to be extracted is designated from the grouped regions, the region setting means g obtains region image data to be extracted in local region units. The contour portion setting means h accurately cuts out the contour portion of the region image data extracted using the color quantized image data grouped by the color quantizing means c for each similar color, and obtains data necessary for mask creation.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image clipping device for extracting only an image of a required area from image data.

[0002]

2. Description of the Related Art Conventionally, in an image processing system such as a layout scanner, as a cropping device for cropping a necessary image area from image data, an operator refers to an image displayed on a display, There is a method of designating a cutout area by tracing the contour portion of the cutout area using a digitizer or a pointing device such as a mouse. Also,
In order to automatically perform these operations, there is a device that detects a color difference or a density difference between a region to be cut out and a background region, divides the two based on a preset threshold value, and cuts out the background region. Further, Japanese Patent Application No. 2-256280 filed by the present applicant has a device for indicating a cutout area by performing area division from color quantization and edge extraction of image data.

[0003]

However, in a cutting device using a trace, if a region having a complicated shape is to be cut out, the tracing work becomes complicated, the burden on the operator increases, and the working time is very long. I can't help it. Further, when the point information read by the trace is connected to form line information, it is necessary to perform smoothing processing or vectorization processing, and complicated processing must be incorporated in the device.

On the other hand, even in an apparatus for automatically cutting out an image from a color signal or a density signal, an operator needs to set an optimum threshold value each time while checking the image, and the image to be cut out is also required. The data is diverse,
When trying to cut out a specific area from a photographic image taken in a natural environment such as outdoors, when the same color as the color in the area to be cut out exists in the background area. Can't survive well.

Further, in a device for performing area division based on color quantization of image data and edge information, an area composed of a pattern made up of various colors and combinations of patterns (eg, sweaters, patterned clothes, flower gardens, etc.) When trying to cut out all together, it is not possible to divide the area all together because the edges are extracted too finely.

[0006]

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and it is an image data inputting means a and a spatial frequency for generating spatial frequency characteristic data of a local region from the image data. Transforming means b, color quantizing means for generating color quantized image data from the image data, c, clustering means d for clustering spatial frequency feature data of the local region according to similarity of spatial frequency features, and the clustering result Area division image generation means e for generating area division image data in units of local areas from the image data, display means f for area division image data in units of local area, and area division image data in units of local area The area setting means g for extracting the area instructed from, and the area division image data in the extracted local area unit Constituting a contour part setting means h for retrieving color quantized image data of a color and setting a contour part, and a mask creating means i for creating mask data from the area division image data in which the contour part is set. The above-mentioned problem is solved by a characteristic image clipping device.

[0007]

In the image cropping device according to the present invention, the spatial frequency conversion means b detects the change in the spatial frequency of each local area in the image data input from the image input means a, and the space of each local area is detected. Frequency feature data is generated. The clustering unit d performs the clustering of the spatial frequency feature data of the local area while detecting the similarity of the strength change of the spatial frequency from the spatial frequency feature data of the local area.

On the other hand, the image data input from the image input means a is grouped by similar colors by the color quantization means c, and the color quantized image data is created by reducing the number of colors. Then, the region-divided image generation means e uses the clustering result of the spatial frequency feature data of the local region to collect the local regions having similar spatial frequency features, thereby dividing the region-divided image data into local regions. Is generated and displayed on the display means f.

When an area to be extracted is designated from the displayed area division image data, the area setting means g obtains area image data to be extracted in local area units. In order to accurately cut out the contour portion of the image data instructed in units of the local area in pixel units, the contour portion setting means h uses the image data color-quantized for the local area including the contour portion, By specifying the color included in the area to be extracted, the contour is accurately cut out, and the area-divided image data necessary for mask creation is determined. Mask data is created by the mask creating means i based on this data. Then, a desired clipped image can be obtained by extracting the area of the image data corresponding to the mask data.

[0010]

An embodiment of the present invention will be described in detail with reference to the drawings. FIG. 2 is a block diagram showing a configuration example of the image clipping device according to the present invention. The cutting device of this embodiment is
Image input means 1 for receiving image data from an external system
A storage means 2 including a semiconductor memory and a magnetic disk, a display means 3 including a color display, and an input means 4 including a pointing device such as a mouse.
It is roughly configured by an arithmetic processing means 5 including a computer and the like, and an output means 6 for outputting the arithmetic result to an external system and the like.

The image input means 1 is for taking in at least image data as R, G, B or Y, M, C signals, and is a scanner for taking in a color original as image data, or image data from an external system. Is a data communication device or the like for taking in.

The storage means 2 stores the image data from the image input means 1, and also stores spatial frequency feature data of a local area, color quantized image data, area division image data in units of local areas, which will be described later. Various processed image data such as region image data, mask data, and a cutout image for each local region selected by instruction are stored.

The operation of the display means 3 is controlled by the arithmetic means 5, and various image data from the storage means 2 are displayed.

Since the input means 4 designates the area to be extracted from the area-divided image data displayed on the display means 3,
And in order to accurately set the contour of the area you want to extract,
This is for designating the color included in the local area including the contour portion.

The arithmetic processing means 5 samples the image data for each local region to generate spatial frequency characteristic data of the local region. Then, the similarity of changes in the spatial frequency is detected from the spatial frequency feature data of the local region, and clustering is performed (a group of similar patterns is called a cluster: Cluster, based on the evaluation criterion of the pattern similarity). Clustering is performed to search for clusters from the distribution), local area spatial frequency feature data is grouped, and local areas having similar spatial frequency characteristics are grouped using this clustering result, so that image data is divided into local area units. To divide. The divided areas are labeled to generate area-divided image data in units of local areas and displayed on the display unit 3.
Then, the input unit 4 designates a region-divided region in the image to be cut out from the displayed region-divided image data in units of the local region, whereby the image region of the local region unit including the region to be extracted is specified. Pick out.

On the other hand, the arithmetic processing means 5 performs color quantization for each pixel based on the input image data. And
This color quantized image is applied to the contour portion of the extracted image area in local area units, and the contour is accurately specified (in pixel units) by specifying the color included in the area to be extracted.
Cut out and select the area required for mask creation. The output means 6 is for outputting the mask data or the cut-out image data to another device.

Next, the operation of this embodiment will be described with reference to the flow charts shown in FIGS. (A) Generation of Local Area Spatial Frequency Feature Data First, image data is input by the image input means 1 and stored in the storage means 2 (step SP1). The stored image data is displayed by the display means 3 (step S
P2), it can be confirmed on the display. The image data on the display means 3 is confirmed (step SP3),
If there are no defects, the image data is a 16 x 16 pixel area x
The spatial frequency feature data of the local region is generated by performing Fourier transform on every four pixels in the direction and the y direction by Fourier transform to express the spatial frequency feature of each local region of 4 × 4 pixels (step SP4), and stored. It is stored in the means 2. (The local area has 4 × 4 pixels as one unit, but since it is preferable to capture the characteristics including the surrounding pixels as the spatial frequency data, the spatial frequency data is obtained from the 16 × 16 pixel data and The local region spatial frequency feature data is assumed to be the feature data of the local region of 4 × 4 pixels. The spatial frequency data in each local region obtained by the Fourier transform is the frequency division (enclosed by a solid line) as shown in FIG. Area), the average of the spatial frequency intensities in each section is calculated, and the value is obtained as a vector whose element is the value for each frequency section for each local area.

That is, the spatial frequency data obtained by the Fourier transform shows what kind of frequency component the image data in the local area has, and this spatial frequency data is divided as shown in FIG. The strength of a specific frequency can be represented by taking out data for each of the created regions and calculating the average. The vector having elements of each average obtained in this way for each area can represent the characteristics (of the frequency distribution) of the data of a certain local area.

As the data to be subjected to the Fourier transform, the gradation data is used as it is in the case of a monochrome image, and the luminance signal is used in the case of a color image. Therefore, when a color image is input as a luminance / color difference signal, the luminance signal can be used, but when input as three primary color data of R, G, B (or C, M, Y), these data can be used in advance. It is necessary to obtain the luminance signal based on

(B) Color Quantization of Image Data Image data is color quantized by a clustering method to generate color quantized image data (step SP5),
It is stored in the storage means 2. Color quantization by this clustering method will be described in detail with reference to FIG. First, R, G, B (or Y, M, C) digital data for each pixel is mapped on the R, G, B color space (step SP51). Next, the variance is calculated for each of the R, G, and B axis data (step SP52), and a threshold is determined for the data of the axis with the largest variance by the discriminant analysis method (step SP53). Divide into two groups. (Step SP54). The same operation is performed for the two divided groups (the variance for each axis data is calculated, the threshold is calculated for the axis with the largest variance, and the two groups are divided), and the group is divided into about 60 to 100. Is performed (step SP55).

Next, for digital data of pixels belonging to the same group, an average value is obtained for each of R, G and B data, and this is set as color quantized image data of pixels belonging to the group (step SP56).

(C) Creating Region-Divided Image Data Using Local Regions as Units Spatial frequency feature data of the local regions generated in step SP4 is clustered using a neural net called Kohonen's feature map to obtain similar features. The grouping of the local spatial frequency characteristic data having is performed (step SP6).

Feature map is Kohone
It is a self-organizing neural network proposed by N. It is composed of two layers as shown in FIG. 7, and has a feature that an inputted vector group can be classified according to its similarity. Region division is performed by assigning the same label to local regions in image data having similar spatial frequency characteristics using the results of grouping by feature map, and region-divided image data in units of local regions is generated. (Step SP7). Based on this data
Data that belong to the same group and are adjacent to each other (4 × 4 pixel block as one unit) are collected and the area dividing line is displayed in the contour portion, so that the operator can distinguish the image area of the same group from other areas. To do so. The area-divided image data is displayed on the display means 3 (step SP8) and stored in the storage means 2.

FIG. 5 shows a case where a human image is cut out, but it is possible to obtain image data divided into regions as shown in FIG. 5B from the image data shown in FIG. 5A.

(D) Designation of extraction area in units of local area With reference to the area-divided image data in units of local area, the image portion to be cut out is specified by the input means 4 (step SP9). This is done by pointing the cursor to a point within the area dividing line for the area to be cut out.

Since the area to be extracted is not necessarily divided into one area as a single area, a plurality of areas can be selected, and the process waits until the area designation is completed (step SP10). When the instruction is completed, the area image data of the selected local area unit is generated and displayed on the display means 3 (step SP1).
1) is stored in the storage means 2.

(E) Extraction of contour portion and creation of mask If necessary for the local area including the contour portion of the area image data while referring to the area image data selected in the unit of the displayed local area The contour part is accurately separated by designating the color in the color quantized image data included in the part to be extracted (step SP12), and the contour part of the area image is processed until the instruction to set the necessary contour part is completed. Is set to obtain the area to be extracted (step SP13).

In this embodiment, the local area is a unit of 4 × 4 pixels, and it is difficult to extract a smooth contour line because the contour portion is stepwise in this unit. Therefore, in order to set the contour line on a pixel-by-pixel basis, the contour portion is extracted from the data of the local region including the contour line by using the color-quantized data.

First, the color quantized image corresponding to the area designated in step SP9 is displayed on the display device 3. Since the color quantized image is divided into about 60 to 100 groups, different colors are displayed for each group so that they can be distinguished.

Next, the color component of the contour portion is included in the image inside (or outside) the contour line of each area including the contour portion (corresponding to the areas a to i in FIG. 5C). The pixel is specified by pointing to the cursor with a cursor or the like. Next, a pixel having the same color quantized data as the specified pixel is searched from the 4 × 4 pixel region forming the contour portion. If you cannot extract the pixels of all contours in one operation,
The sample data may be taken again with the cursor and the same process may be repeated.

Since the image often has different color components inside and outside the contour line, by performing the above processing, the inside (or outside) contour line is selected from the local area including the contour line.
Only the pixel data of can be extracted. A mask for cutting out the designated area is generated by mapping the set area of the contour portion (step SP).
14).

The image data divided into regions as shown in FIG. 5B are cut out as shown in FIG.
A mask as shown in FIG. 5D can be generated by specifying a plurality of i.

(F) Creation of cut-out image data The generated area extraction mask is displayed on the display means 3 (step SP15), and this data is stored in the storage means 2. The shape of the area to be extracted and the shape of the displayed area extraction mask are compared and confirmed (step SP1).
6) By multiplying the image data and the area extraction mask data (AND operation), cutout image data as shown in FIG. 4E is generated (step SP1).
7) The cut-out image data is stored in the storage means 2 and displayed on the display means 3 (step SP1).
8). The cut-out image data displayed on the display means 3 is confirmed (step SP19), and if OK, the output means 6
To output data (step SP20).

If the cut-out image cannot be extracted successfully, a loop for returning to step SP8 is prepared in order to extract the area-divided image data again.

As described above, according to the present invention, the characteristic that the spatial frequency is similar in the local region is used by using the property that the spatial frequencies are similar in the regions where the regularity of the shade is similar. Since the image is automatically divided into regions based on the nature, there is no need for manual tracing work as in the past, and since the regions are not divided only by the difference in color, the background area and part of the If the same color is present in the background, or if you are trying to cut out a region (for example, a sweater, patterned clothes, flower garden, etc.) that is made up of a combination of various colors and patterns, the clipping that involves human hands The work can be easily performed in a short time.

In this embodiment, the case where an image is cut out by a layout scanner or the like has been described. However, a peel-off film cutout used for cutting out a specific portion of a disassembled film in the plate making process of ordinary photolithography. Machine data can be easily obtained by extracting mask contour data.

The present invention is not limited to the above embodiment. As the image input means, the image data may be input using a receiving device that receives the data from an external system, or the image data may be directly addressed from the photo original by a scanner. Further, as the image output means, the cropped image data may be output using a communication device that transmits data to an external system, may be output to a layout scanner, or may be output to a hard copy machine or the like. The spatial frequency conversion means for obtaining the spatial frequency characteristic for each local region is not limited to the above embodiment, and a Gabor conversion method or the like may be used. The local spatial frequency data clustering method is not limited to the above-mentioned embodiment, and another clustering method or vector quantization method having equivalent ability, for example, M
A method such as AXNET (a type of neural network model) may be used.

Further, the input image data is R,
The data described in the G, B color space or the Y, M, C color space may be data described in another color space, for example, a uniform color space such as a Lab space.

[0039]

As described above, according to the image cropping device of the present invention, it is possible to easily crop an image without setting troublesome cropping conditions and to significantly reduce the working time. it can. In particular, it is possible to cut out from image data by a photograph taken in a natural environment, and it is possible to cut out an image area composed of a pattern made up of various colors and patterns.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an image clipping device according to the present invention.

FIG. 2 is a block diagram showing a schematic configuration of an image clipping device according to the present invention.

FIG. 3 is a flowchart for explaining the operation of the present invention.

FIG. 4 is a flow chart for explaining the operation of the present invention.

FIG. 5 is an explanatory diagram for explaining the operation of the present invention.

FIG. 6 is an explanatory diagram showing an example of spatial frequency division of a local region for explaining the operation of the present invention.

FIG. 7 is an explanatory diagram of Kohonen's feature map.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Image input means 2 ... Storage means 3 ... Display means 4 ... Input means 5 ... Arithmetic processing means 6 ... Output means

Claims (1)

[Claims] 1. An image data input means, a spatial frequency conversion means for generating spatial frequency feature data of a local region from the image data, and a color quantization means for generating color quantized image data from the image data. Clustering means for clustering the spatial frequency feature data of the local area according to the similarity of the spatial frequency characteristics, and area division image generation means for generating area division image data in units of local area from the image data using the clustering result, A display unit for displaying the region-divided image data in units of the local region, a region setting unit for extracting a region instructed from the region-divided image data in the unit of local region, and a region-division image data in the extracted unit of local region Contour part setting means for retrieving color quantized image data of a color designated by the following and setting the contour part; Image cutout apparatus characterized by comprising a mask generating means for generating mask data from the region-divided image data.