JP2002117409A - Image processing method and device thereof - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve a problem that image processing is inefficient because an image is excessively divided and even the background is judged to be an object when performing similar image retrieval by extracting the object by the division of a region by utilizing clustering. SOLUTION: An adjacent relation judging part 13 integrates image data 101 divided by a clustering part 11 and a labeling part 12 based on the adjacent relation. An integrated information generation part 14 generates background mask of an image based on image classification data 102. A background detection part 15 detects a background region from the region integrated based on the background mask and removes it to extract the object.

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 and apparatus, for example, an image processing method and apparatus for performing a similar image search using an outline of an object as a feature amount of an image.

[0002]

2. Description of the Related Art Conventionally, an object is extracted from an image,
2. Description of the Related Art There is known a technique for performing a similar image search by determining an outline similarity between the object and a plurality of model images.

As an object extraction method, for example,
There is a method of classifying an image into a predetermined number of classes (clustering) to divide the region and connecting the obtained label images. For example, 1 for RGB 8 bits
When performing clustering by color on a multivalued image expressing 7.6 million colors, the number of classes (representative colors) is 1
In many cases, about 0 to 50 colors are set.

[0004] As a method for dividing an image into regions by such clustering, many methods are known, such as k clustering and a method using a Markov random field.

[0005]

Generally, one object is formed by combining a plurality of regions. Therefore, in order to extract an object from a region-divided image, as described above, a process of integrating the divided regions is necessary, but this is a complicated process, which has caused a reduction in processing speed. .

[0006] In the area division method based on clustering, an image is over-divided.
For example, even the background is a target for similarity search of an object, which is inefficient.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problem, and has as its object to provide an image processing method and an apparatus for efficiently searching for an outline similarity of an object.

[0008]

As one means for achieving the above object, the image processing method of the present invention comprises the following steps.

More specifically, there is provided an image processing method for extracting an object from an image, comprising: an area dividing step of dividing the image into a plurality of areas based on color information; Integration process to integrate
A background information generating step of generating background information of the image based on predetermined classification information related to the image; a background detecting step of detecting a background area from the integrated area based on the background information; An object extraction step of extracting an object by removing the background area from the area.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment according to the present invention will be described below in detail with reference to the drawings.

First Embodiment System Configuration FIG. 1 shows an outline of an object as a feature amount of an image.
FIG. 1 is a diagram illustrating a configuration of a system that performs similar image search. In FIG. 1, reference numeral 21 denotes a model image database, which stores a large amount of model images. An object extraction unit 28 extracts an object from a model image by performing a region division and integration process, which is a feature of the present embodiment. Reference numeral 22 denotes a contour extraction unit that extracts a contour of an object. Reference numeral 23 denotes a contour database that stores contour information for each object of each model image.

On the other hand, reference numeral 24 denotes an example image for performing a similar image search. Reference numeral 29 denotes an object extraction unit, similar to the object extraction unit 28, which performs region division and integration processing on the example image 24 to extract objects. Reference numeral 25 denotes a contour extraction unit, which extracts a contour of an object, similarly to the contour extraction unit 22.

Reference numeral 26 denotes a similarity detection unit which detects, from the outline information of the model image stored in the outline database 23, one having a high degree of similarity to the outline of the example image 24, and displays the result on the display unit 27. I do. As a display method, for example,
Displaying the name of the detected model image, accessing the model image database 21 to display the detected model image, and the like can be considered.

FIG. 2 is a block diagram showing a detailed configuration of the object extracting unit 28. Note that the object extraction unit 29 also has the same configuration as in FIG. Hereinafter, the object extraction process and the contour extraction process for each object in the present embodiment will be described in detail with reference to FIG.

The object extracting section 28 (29) comprises a clustering section 11, a labeling section 12, an adjacency determining section 13, an integrated information generating section 14, and a background detecting section 15. Image classification data 102 indicating information about the background is input to the integrated information generation unit 14, which is, for example,
It may be input by a user via a user interface (not shown) or may be predetermined data.

The pixels constituting the image data 101 (model image or exemplifying image) are classified by the clustering unit 11 into, for example, a class of the same color. The number of classes to be classified is 10 to 50 colors with respect to the original number of colors (for example, 16.7 million colors with 8 bits of RGB). The clustering in the clustering unit 11 is realized by a known method such as a method using k-class cling or a Markov random field.

Reference numeral 12 denotes a labeling unit, which labels a portion where pixels are connected based on the clustering result in the clustering unit 11. An area labeled here (hereinafter, referred to as a label area) is a minimum component of the object in the present embodiment.

Here, the object is generally formed by combining a plurality of label areas. Therefore, in the present embodiment, the adjacency determination unit 13 determines the adjacency of a plurality of label areas to integrate them and generate a sub-object.

FIG. 3 is a flowchart showing a sub-object generation process in the adjacency determination section 13.
In the present embodiment, when two label areas are in contact at many points and the average colors are close, it is determined that they are in the same group.

First, in step S601, a label area in contact with the target label area is searched to count the number of adjacent points. In step S602, the counted number of adjacent points is compared with a predetermined threshold value T1, and if it is larger than T1, the color difference of the average color in the two areas is calculated in step S603. And step S604
Then, the calculated color difference is compared with a predetermined threshold T2,
If it is smaller than T2, it is determined that the target label area and the searched adjacent label area belong to the same group, and a flag indicating that these two areas have a connection relation is set in step S605.

In step S606, it is determined whether the search for all label areas has been completed, that is, whether the search for all adjacent areas has been completed for the target label area and whether all label areas have been set as the target label area. judge. If it is determined that the search has been completed, a plurality of connected label areas are grouped based on the connection flag in step S607. This group is a sub-object.

Here, as described in the above conventional example,
In the area division method using clustering, generally, an image tends to be over-divided. For example, even with a single background color such as gray, if there is a density gradient (gradation) from the right edge to the left edge or from the upper edge to the lower edge of the image, the right edge and the left edge, the upper edge, the lower edge, etc. ,
In some cases, the area is divided as another area despite the same background. As described above, since the background having the density gradient (gradation) or the background portion of the moving image is over-divided, the over-divided background is also determined as an individual object. Or the similarity calculation process.

Therefore, in the present embodiment, the problem caused by over-segmentation is solved by creating integrated information for area segmentation using the image classification data 102 in the integrated information generation unit 14. As a more specific example, image classification data 1
Reference numeral 02 denotes information on the background. The integrated information generation unit 14 generates mask data (background mask) used when integrating the background based on the image classification data 102 and the image data 101.

FIG. 4 is a diagram for explaining a method of generating a background mask in the integrated information generating unit 14. In this example, an example of a method of detecting a background candidate pixel when assuming data of "still object photographed in a uniform background" as the image classification data 102 is shown. According to the figure, first, a maximum value and a minimum value of a pixel value are detected in an N × N rectangular area centered on a target pixel. When the difference is small, that is, when the difference is equal to or less than a predetermined value,
The target pixel is set as a background region candidate. By performing this processing on all the pixels of the image data 101, a background mask is generated.

The background detection unit 15 extracts a background region based on the integration result (sub-object) of the label region in the adjacency determination unit 13 and the background mask generated in the integration information generation unit 14.

FIG. 5 is a diagram showing a specific example of the background extraction processing in the background detector 15. In the figure, reference numeral 31 denotes an original image, and reference numeral 32 denotes a result (sub-object) obtained by integrating the label areas by the adjacent relationship in the adjacent relationship determination unit 13. In the sub-object 32, the background area is also divided into a plurality of areas. Reference numeral 33 denotes background information (background mask) generated by the integrated information generation unit 14. The original image 31 is classified into the above-described background region candidate pixels (white portions in the drawing) and other pixels (colored portions in the drawing). Have been. The background detection unit 15 determines an area of the sub-object 32 that includes many background area candidate pixels indicated by the background mask 33 as a background area. The image of the determination result is shown at 34. According to the image 34, it can be seen that the background area other than the object has been substantially extracted from the original image 31.

The background detection section 15 outputs the object area by excluding the detected background area from the sub-objects output from the adjacency determination section 13. This is the object extraction result in the object extraction unit 28 (29).

The contour extraction unit 22 (25) outputs a contour data 103 by subjecting the object extracted by the object extraction unit 28 (29) to a known contour extraction process. In the present embodiment, the result of performing the wavelet transform on the contour point coordinates of the object is defined as contour data 103. If the contour data 103 is for a model image, it is stored in the contour database 23, and if it is for the example image 24, it is directly provided to the similarity detection unit 26.

As described above, in the contour extraction processing of the present embodiment, at least one object in the image
In order to solve the problem that the background is over-divided by clustering based on the fact that the outline of the object roughly matches the boundary of the label area, In addition to the integration of label areas based on the relationship, background data that has been over-divided using background mask information is integrated.

[Similarity Detection Processing] Hereinafter, in the system of the present embodiment shown in FIG.
This will be described with reference to the flowchart of FIG. Prior to the start of this process, the contour database 23 has already
It is assumed that contour data obtained by performing a wavelet transform on contours of a plurality of model images is stored.

First, in step S1, an object extracting unit 2 extracts an exemplary image 21 which is, for example, a grayscale image.
At 9, an object is extracted. Next, in step S2, the contour extraction unit 25 extracts the contour of the object extracted in step S1. In step S3, the contour of the object obtained in step S2 is divided into N equal parts to obtain contour points. Next, in step S4, the contour point is subjected to polar coordinate transformation centering on the center of gravity of the object. Then, in step S5, the contour represented by the contour points is converted into a wavelet descriptor. The processing of steps S2 to S5 is performed in the contour extraction unit 25.

In step S6, the similarity of the contour after the wavelet transform is calculated with the contour of the model image stored in the contour database 23 using the component corresponding to the low frequency component. This similarity calculation is a matching process between the contour of the object and the contour of the model image.

Thereafter, in step S7, step S
Only for the model image for which a high degree of coincidence was obtained by the matching of No. 6, the component corresponding to the high-frequency component
Perform higher-precision matching. The matching processing in steps S6 and S7 is performed by the similarity detection unit 26.

Then, in step S8, step S
The user is notified by displaying the matching results obtained in Steps 6 and S7 on the display unit 27 as the similarity.

In the following, among the processes shown in FIG. 6, in particular, the weblet conversion and matching process of the contour points shown in steps S3 to S7 will be described in detail.

Since the object extraction processing shown in step S1 has been described above, the description is omitted here. Also, a known method can be applied to the contour line extraction processing shown in step S2, and a description thereof will be omitted.

[Contour Line Division / Polar Coordinate Transformation Processing] The preprocessing for wavelet transformation in steps S3 and S4 will be described below.

First, the process for obtaining the contour points of the object in step S3 will be described with reference to FIG. FIG. 7 is a diagram for describing processing for obtaining a contour point and its curve from a contour line in the present embodiment. FIG. 7A shows 16 contour points obtained by dividing the square contour line into 16 points with the center of the square as the origin. It is assumed that the number assigned to each contour point is a contour point number, and contour tracing is performed in ascending order from the contour point whose number is 1. FIG. 7B shows the trajectory of the tracking result of each contour point shown in FIG. 7A. In FIG. 7B, the horizontal axis is the contour point number n
And the vertical axis indicates the coordinate values of x and y. According to FIG.
It can be seen that the trajectories x (n) and y (n) of the x and y coordinates are both trapezoidal waves of one cycle.

In step S4, the locus is subjected to the following polar coordinate transformation. First, the center of gravity (x0, y0) of the object is determined from the average value of the x coordinate and the average value of the y coordinate of the image. And this center of gravity (x0,
With y0) as the origin, polar coordinate conversion is performed according to the following equation.

[0040]

(Equation 1)

In this embodiment, the following wavelet transform is performed only on r (n) obtained by the polar coordinate transform shown in the above equation.

Wavelet Transform Processing The wavelet transform processing in step S5 described above will be described below with reference to FIGS.

FIG. 8 is a diagram for explaining the wavelet transform in the present embodiment. First, input data r
(n) is branched into a low-pass filter H0 and a high-pass filter H1, and the filter output is down-sampled.
Then, the output of the low-pass filter H0 is further branched into a low-pass filter H0 and a high-pass filter H1, and the filter output is down-sampled to proceed with the division. As a result, the input data r (n) is converted into a wavelet descriptor, and is substantially separated into a plurality of levels of a low frequency component and a high frequency component.

Here, the filter coefficients shown in the figure are, for example, a simple Haar basis. This example is shown below. That is, the conversion in the low-pass filter H0 is indicated by H0 (z), and the conversion in the high-pass filter H1 is indicated by H1 (z).

[0045]

(Equation 2)

As described above, the H0 wave is obtained by converting r (n) with the low-pass filter H0, and the H1 wave is obtained by converting the r (n) with the high-pass filter H1. Similarly, the H00 wave and the H01 wave are obtained by, for example, further converting the H0 wave by the low-pass filter H0 and the high-pass filter H1. The multi-level wavelet transform results for r (n) are stored as ri (n).

The same wavelet transform is applied to a plurality of (for example, n) model images in advance.
The conversion result is stored in the contour database 23 as rj (n).

Hereinafter, a method of determining similarity using the wavelet transform in the present embodiment will be specifically described.

FIG. 9 is a diagram showing an example in which wavelet descriptors for a square and a circle are represented in an orthogonal coordinate system. In the figure, both the square and the circle are diagrams showing the trajectories of the contour points on the left of the three waveform diagrams. As a result of performing polar coordinate transformation and wavelet transformation on these, the H0 wave shown in the center and The H1 wave shown on the right is obtained.

According to the figure, it can be seen that the H0 waveform by the wavelet transform is similar between the square and the circle. As a result, the square and the circle are determined to have high similarity as a figure having substantially the same distance from the center of gravity to the contour point. Therefore, in the present embodiment, they are determined to be similar as a schematic figure. On the other hand, it is determined at this stage that a rectangle, a largely flat ellipse, or a figure with a depression has low similarity to a square or a circle.
It should be noted that the square and the circle are represented by H
Since the waveforms of 1 are significantly different, they can be distinguished by the H1 waveform.

Here, in the determination of the similarity based on the H0 waveform, it can be seen that the number of reference points indicated by the contour point numbers is halved for both the model image and the object image. Therefore, the similarity calculation in the present embodiment can be expected to be approximately twice as fast as when all the reference points are used.

Further, when the H00 wave is obtained by further wavelet transforming the H0 wave, the number of reference points can be reduced to 1/4 in the H00 waveform, so that a fourfold speedup can be expected. That is, when a rough similarity calculation is sufficient, higher-speed processing can be performed by using a higher-order wavelet descriptor.

The matching process in steps S6 and S7 will be described below with reference to FIG. As described above, steps S6 and S7 differ in whether the comparison between the model image and the object image is performed based on the low-frequency component or the high-frequency component.

This embodiment is characterized in that matching is performed in consideration of the influence of the start point of contour tracing. If the starting point of the contour tracking between the model image and the object image is different, the Sim value, which is an index of the similarity described later, also differs. Therefore, in the present embodiment, the starting point of the contour tracking in the model image is shifted by one point,
A plurality of Sim values for the object image are calculated, and the minimum value of the obtained Sim values is used as the similarity between the model image and the object image.

FIG. 10 is a flowchart showing the similarity calculation processing, that is, matching between the object image and the model image in the present embodiment. First, step S601
, The tracking start point of the waveform of the referenced model image is set to 1
Shift one. Then, in step S602, a Sim value serving as an index of similarity is calculated by the following equation.

[0056]

(Equation 3)

In the above equation, ri (n) and rj (n) indicate values obtained by performing a wavelet transform on the contour points of the object image and the model image, respectively. Note that, as described above, a plurality of levels of conversion results can be obtained by performing a wavelet transform on the contour points, but ri (n) and rj to be compared are obtained.
(n) are assumed to be at the same level as each other.

Next, in step S603, by comparing the Sim value obtained in step S602 with the value in a minimum value register (not shown), it is determined whether or not the Sim value is the minimum in tracking up to the present. And if it is the minimum, the value in the minimum value register is
Replace with m value. Then, in step S605,
If it is determined that the shift of the entire circumference has been completed, the process is terminated. That is, the Sim value finally stored in the minimum value register is an index of the similarity to the model.

By executing the matching process shown in the flowchart of FIG. 10 for each of the contour data of a plurality of model images registered in the contour database 23, Sim which is an index of the contour similarity of the object image with respect to each model image is obtained. A value is calculated for each. Among them, the contour of the model image having the minimum Sim value is
It can be determined that the contour is most similar to the contour of the object image.

In step S6 in FIG. 1, the matching process shown in FIG. 10 is performed on the levels corresponding to the low frequency components of ri (n) and rj (n), so that a predetermined value among a plurality of calculated Sim values is obtained. Extract those that are less than or equal to the value. That is, a candidate of the most similar object image is obtained from the plurality of model images. Here, if it is sufficient to obtain a rough similarity, the process proceeds to step S8 without performing the matching in step S7, and the model image extracted in step S6 is replaced with the similarity (Sim
Value) together with the value).

When it is desired to make a higher-resolution similarity determination, in step S7, the model image extracted in step S6 is subjected to the level corresponding to the high-frequency components of ri (n) and rj (n). Perform higher-definition matching. Then, in step S8, step S
The model image having the minimum Sim value obtained by the matching in step 7 is displayed on the display unit 27 together with the similarity (Sim value).

As described above, according to the similarity detection processing of the present embodiment, the calculation of the contour similarity between the model image and the object image is performed based on the wavelet descriptor of the contour, so that the number of reference points can be reduced. Can be reduced, so that the processing speed is increased.

Since the global similarity or the local similarity can be detected in accordance with the level of the wavelet transform, it is possible to calculate the similarity more suited to the needs of the user.

As described above, according to the present embodiment,
Since the background area can be appropriately integrated by creating the background mask, similar image retrieval using the contour of the object as a feature amount of the image can be performed at high speed.

<Second Embodiment> Hereinafter, a second embodiment according to the present invention will be described.
An embodiment will be described. In the second embodiment, the same components as those in the above-described first embodiment are denoted by the same reference numerals, and description thereof is omitted. In the second embodiment, an example in which the similar image search processing described in the first embodiment is applied to a moving image will be described. That is, the example image 24 shown in FIG. 1 of the first embodiment is a moving image in the second embodiment.

In the second embodiment, it is assumed that the example image (moving image) 24 to be searched is a video image obtained by panning the camera, as preliminary knowledge. In this case, in the video, the movement of the background portion is uniform. Therefore, the second embodiment is characterized in that the integrated information generation unit 14 in the object extraction unit 29 creates integrated information (background mask) for area division using uniform motion vectors.

FIG. 11 is a block diagram showing a detailed configuration of the integrated information generation unit 14 in the object extraction unit 29 according to the second embodiment. Here, information that “the image is panned” is input as the image classification data 102 to the integrated information generation unit 14. Then, the motion vector search unit 210 obtains a motion vector by a known method such as block matching. A clustering unit 211 clusters the obtained motion vectors, and a background extraction unit 212 extracts, for example, a wide area having uniform motion vector sizes as a background based on the clustering result, and creates a background mask 201. I do.

In the second embodiment, the image classification data 102 can further include information on the background. For example, the above-mentioned “video by panning”
Other than that, information such as “uniform background” may be included as in the first embodiment. In this case, the integrated information generation unit 14 may create the background mask 201 based on not only the magnitude of the motion vector but also whether or not the pixel value is uniform as in the first embodiment.

As described above, according to the second embodiment, when an example image is a moving image, a portion having a uniform motion vector is extracted as a background region, thereby providing mask data for background integration. Generate This allows
Similar images can be searched for an object in a moving image, as in the first embodiment.

In the second embodiment, the structure of the object extracting unit 28 for the model image is the same as that of the first embodiment.
This may be the same as the embodiment. That is, even if the example image 24 is a moving image, the model image may be a still image.

[0071]

[Other Embodiments] Even if the present invention is applied to a system including a plurality of devices (for example, a host computer, an interface device, a reader, a printer, etc.), an apparatus (for example, a copying machine) Machine, facsimile machine, etc.).

Further, an object of the present invention is to supply a storage medium (or a recording medium) on which a program code of software for realizing the functions of the above-described embodiments is recorded to a system or an apparatus, and a computer (a computer) of the system or the apparatus. Or a CPU or MPU) reads out and executes the program code stored in the storage medium,
Needless to say, this is achieved. In this case, the program code itself read from the storage medium implements the functions of the above-described embodiment, and the storage medium storing the program code constitutes the present invention.
In addition, by the computer executing the readout program code, not only the functions of the above-described embodiments are realized, but also based on the instructions of the program code,
The operating system (OS) running on the computer performs part or all of the actual processing,
It goes without saying that a case where the functions of the above-described embodiments are realized by the processing is also included.

Further, after the program code read from the storage medium is written into the memory provided in the function expansion card inserted into the computer or the function expansion unit connected to the computer, the program code is read based on the instruction of the program code. , The CPU provided in the function expansion card or the function expansion unit performs part or all of the actual processing,
It goes without saying that a case where the functions of the above-described embodiments are realized by the processing is also included.

[0074]

As described above, according to the present invention, it is possible to efficiently search for the contour similarity of an object.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of a similar image search system according to an embodiment of the present invention.

FIG. 2 is a block diagram illustrating a detailed configuration of an object extraction unit.

FIG. 3 is a flowchart illustrating a sub-object generation process in an adjacency determination unit.

FIG. 4 is a diagram illustrating a method of generating a background mask in an integrated information generation unit.

FIG. 5 is a diagram illustrating a specific example of a background extraction process in a background detection unit.

FIG. 6 is a flowchart illustrating a similarity calculation process.

FIG. 7 is an explanatory diagram of a contour curve.

FIG. 8 is a diagram illustrating wavelet transform.

FIG. 9 is a diagram illustrating an example of a wavelet descriptor.

FIG. 10 is a flowchart illustrating a matching process.

FIG. 11 is a block diagram illustrating a detailed configuration of an integrated information generation unit according to a second embodiment.

[Explanation of symbols]

 21 Model Image Database 22, 25 Contour Extraction Unit 23 Contour Database 24 Example Image 26 Similarity Detection Unit 27 Display Unit 28, 29 Object Extraction Unit

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) H04N 1/387 H04N 1/387 1/46 1/46 Z F-term (Reference) 5B057 CA01 CA08 CA12 CA16 CB01 CB08 CB12 CB16 CC01 CD18 CE06 CE09 CH09 DB02 DB06 DB09 DC14 DC16 DC25 DC33 5C076 AA02 AA11 AA19 AA36 BA06 CA10 5C079 LA01 LA02 LA10 LA39 LA40 NA13 5L096 AA02 EA37 FA06 FA15 FA32 GA34 GA55 HA09 JA11 MA07

Claims (12)

[Claims] 1. An image processing method for extracting an object from an image, comprising: an area dividing step of dividing the image into a plurality of areas based on color information; A background information generating step of generating background information of the image based on predetermined classification information relating to the image; and detecting a background region from the integrated region based on the background information. An image processing method, comprising: a background detection step; and an object extraction step of extracting an object by removing the background area from the integrated area. 2. The area dividing step includes: a clustering step of classifying constituent pixels of the image into a plurality of color classes; and a labeling step of labeling a portion where pixels are connected in the plurality of color classes. 2. The image processing method according to claim 1, comprising: 3. In the integrating step, areas adjacent to each other are integrated when the number of adjacent points is equal to or more than a predetermined number and the difference between the average colors in these areas is equal to or less than a predetermined value. The image processing method according to claim 1, wherein 4. The image processing method according to claim 1, wherein the background information is background mask information indicating whether or not constituent pixels of the image are pixels in a background area. 5. In the background information generating step, when it is indicated that the background is uniform as the classification information,
5. The image processing method according to claim 4, wherein the background mask information is generated by setting a pixel having a variation of a neighboring pixel value equal to or less than a predetermined value as a background pixel. 6. The image processing according to claim 5, wherein, in the background detecting step, when the integrated region includes the background pixels in a predetermined ratio or more, the integrated region is detected as a background region. Method. 7. In the object extracting step,
2. The image processing method according to claim 1, wherein contour information of an area obtained by removing the background area from the integrated area is extracted. 8. The image processing method according to claim 1, further comprising a similarity detecting step of detecting a similarity between the object extracted in the object extracting step and a plurality of model images. 9. The image according to claim 8, wherein in the similarity detecting step, a similarity between the object and a predetermined model image is detected based on a result of a wavelet transform performed on an outline point of the object. Processing method. 10. The image processing method according to claim 1, wherein in the background information generating step, when the image is a moving image, the background information is further generated based on the motion information. 11. An image processing apparatus for extracting an object from an image, comprising: an area dividing unit that divides the image into a plurality of areas based on color information; Means for generating background information of the image based on predetermined classification information relating to the image; and detecting a background area from the integrated area based on the background information. An image processing apparatus comprising: a background detection unit; and an object extraction unit that extracts an object by removing the background region from the integrated region. 12. A recording medium on which is recorded a code of an image processing program for extracting an object from an image, the program code comprising: at least an area dividing step of dividing the image into a plurality of areas based on color information A code of an integration step of integrating the plurality of divided areas based on the adjacent relationship thereof; and a code of a background information generation step of generating background information of the image based on predetermined classification information regarding the image. Based on the background information, a code of a background detection step of detecting a background area from the integrated area, a code of an object extraction step of extracting an object by removing the background area from the integrated area, A recording medium comprising: