JP2002190024A - Image recognition method, image recognition system using the same, command interface and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image recognition method allowing the rapid recognition of the shape of an object using the mapped feature of the luminosity distribution of original image data to the orthogonal base, and to provide an image recognition system using the method, a command interface applying the system, and a recording medium storing a program for realizing them. SOLUTION: This image recognition method of recognizing the object included in the original image data by comparing two-dimensional FFT image data obtained by mapping the original image data to the orthogonal base, with a registration pattern comprising the two-dimensional FFT image data, has a mapping process, a mosaic process and a similarity degree detecting process.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition method for recognizing the state of an object included in original image data, an image recognition system using the same, a command interface using the system, and a program for realizing these. Related to a recorded medium.

[0002]

2. Description of the Related Art Recently, an image recognition technique for recognizing hand gestures and body actions by extracting a part of original image data shot by a camera has been proposed. Such a technique requires a process of extracting a part of the original image data, so that the processing time becomes long. Therefore, it is difficult to use it as an input interface of a video game that requires a response to real time.

On the other hand, as a technique for recognizing human facial expressions, there has been proposed a technique for mosaicizing a face image and inputting brightness distribution information of original image data into a neural network to identify a large number of faces. ing. There is also a technique in which a two-dimensional cosine transform is performed on a difference between original image data with expression and expressionless original image data, and the DCT coefficient in the low frequency region is directly input to a neural network.

[0004]

However, the face image recognition technology described above is a technology for limited use such as personal recognition and facial expression recognition, and cannot be directly applied to other shape recognition. In addition, since the latter recognition technology does not use information in the high-frequency region, not only can the creatures not recognize the rough or slick feeling visually recognized, but also the robustness is reduced and compression operations such as mosaicization are performed. , The processing time becomes longer. This makes it difficult to use it as an input interface for video games that require a response to real time.

The present invention has been made based on such a technical background. That is, an object of the present invention is to provide an image recognition method capable of recognizing a shape of an object at high speed by using a feature obtained by mapping a brightness distribution of original image data to an orthogonal basis, and an image recognition system using the same. An object of the present invention is to provide a command interface using a system and a recording medium storing a program for realizing the command interface.

[0006]

Means for Solving the Problems In order to achieve the above object, a first invention compares two-dimensional FFT image data obtained by mapping original image data on an orthogonal basis with a registered pattern composed of two-dimensional FFT image data. Preferably, the image recognition method recognizes an object included in the original image data, and includes a mapping step, a mosaicing step, and a similarity detection step. Thus, since the cutout processing of the recognition target is not performed, the processing can be sped up. The mapping step is a step of obtaining two-dimensional FFT image data from original image data. Through such a process, it is possible to process information of a Fourier-like rough feeling or a smooth feeling. The mosaicing step is a step of mosaicing the two-dimensional FFT image data obtained in the above-described mapping step. The similarity detection step is a step of calculating the similarity between the mosaic data obtained in the above-described mosaicization step and the registered pattern.

According to a second aspect of the present invention, in the first aspect, it is preferable that the method further includes an image clipping step of setting the number of vertical and horizontal pixels of the original image data used in the mapping step to a power of two. Through these steps, the discrete Fourier transform can be performed at high speed.

According to a third aspect of the present invention, there is provided image recognition for recognizing an object included in the original image data by comparing the two-dimensional FFT image data obtained by mapping the original image data on an orthogonal basis with a registered pattern comprising the two-dimensional FFT image data. The system preferably includes a mapping unit, a mosaic unit, and a similarity detection unit. Thereby, since the cutout processing of the recognition target is not performed, the processing can be speeded up. The mapping means is means for obtaining two-dimensional FFT image data from original image data. By such means, it is possible to process information of a Fourier-like rough feeling or a smooth feeling. The mosaic means is a means for mosaicizing the two-dimensional FFT image data obtained by the above-mentioned mapping means. The similarity detecting means is means for calculating the similarity between the mosaic data obtained by the above-described mosaicizing means and the registered pattern.

In a fourth aspect based on the third aspect, it is preferable that the image processing apparatus further comprises means for setting the number of vertical and horizontal pixels of the original image data used in the mapping means to a power of two. By providing such means, discrete Fourier transform can be performed at high speed.

A fifth aspect of the present invention is a command interface for outputting a command based on a result of recognizing a hand shape or its position included in original image data picked up by an image pickup means. It is desirable to include a similarity detection unit and a command generation unit. As a result, since the cutout processing of the recognition target is not performed,
Enables high-speed processing. The mapping means is means for obtaining two-dimensional FFT image data from original image data. By such means, it is possible to process information of a Fourier-like rough feeling or a smooth feeling. The mosaicing means is a means for mosaicizing the two-dimensional FFT image data obtained by the above-mentioned mapping means. The similarity detecting means is means for calculating the similarity between the mosaic data obtained by the above-described mosaicizing means and the registered pattern. The command generation means is a means for outputting a command based on the result from the similarity detection means.

In a sixth aspect based on the fifth aspect, it is preferable that the image processing apparatus further includes an angle-of-view setting means for adjusting the number of vertical and horizontal pixels of the original image data used in the mapping means to a power of two. By providing such means, discrete Fourier transform can be performed at high speed.

A seventh invention is a computer-readable recording medium in which a program for outputting a command based on a result of recognizing a shape or a position of a hand included in original image data picked up by an image pickup means is recorded. It is preferable that the computer-readable recording medium stores a program for causing a computer to execute the mapping step, the mosaicing step, the similarity detection step, and the command generation step. The mapping step is a step of obtaining two-dimensional FFT image data from original image data. With this step, it is possible to simultaneously process not only the information of the Fourier-like roughness and the feeling of smoothness but also the information of the brightness distribution of the entire screen. The mosaicing step is a two-dimensional FF obtained in the above-described mapping step.
This is the step of mosaicizing the T image data. The similarity detection step is a step of calculating the similarity between the mosaic data obtained in the above-described mosaicization step and the registered pattern. The command generation step is a step of outputting a command based on the result from the above-described similarity detection step.

In an eighth aspect based on the seventh aspect, it is preferable that the image processing apparatus further includes an angle-of-view setting step for aligning the number of vertical and horizontal pixels of the original image data used in the mapping step to a power of two. With such steps, the discrete Fourier transform can be performed at high speed.

In a ninth aspect based on the first aspect or the second aspect, it is preferable that the original image data is obtained through a network interface. Thus, an object can be recognized at high speed using original image data captured at a remote place.

[0015] In a tenth aspect based on the third or fourth aspect, it is preferable that the original image data is obtained through a network interface. This allows
An object can be recognized at high speed using original image data captured in a remote place.

In an eleventh aspect based on the fifth or sixth aspect, it is preferable that the original image data is obtained via a network interface. This allows
An object can be recognized at high speed using original image data captured in a remote place.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS (Embodiment 1) Hereinafter, the present embodiment recognizes the shape of a human hand included in original image data photographed by a video camera and uses the result of detecting the position of the hand as a command. A description will be given assuming that the present invention is applied to a game system that has a command interface (hereinafter, the interface is abbreviated as an I / F) that outputs a command, and that proceeds with a game using a command output from the I / F.

The command I / F in the present embodiment is:
An I / F for inputting a command for controlling the water flow by recognizing the shape and position of the hand existing in the original image data to the game program. Commands output from the I / F include an on / off command, a strength command, a direction command, and a water source command. The ON / OFF command, the strength command, and the direction command are obtained by image recognition processing. The water source command is obtained by the center of gravity detection processing.

The on / off command is a command for recognizing the shape of a hand existing in the original image data and turning on / off the water flow with the shape. For example, when the hand does not exist in the original image data or when the original This is a command to stop the water flow when the hand shape existing in the image data is goo.

The strength command is a command for designating the strength of the water flow. For example, when the shape of the hand existing in the original image data is par, a command for generating a weak water flow, or a command that is continuously captured. When the shape of the hand existing in the original image data changes from goo to par, there is a command in which the force accumulated during goo is released at once by par and a strong water flow is generated.

The direction command is a command for changing the direction of the water flow, for example, changing the direction of the water flow to a direction corresponding to the direction of the par-shaped hand existing in the original image data.

The water source command is a command that identifies the position of the hand and uses that position as the water source. For example, by detecting the center of gravity of the hand existing in the original image data, it is based on the data representing the position of the center of gravity. This command specifies the location of the water source.

There are three types of game contents in this embodiment. The first is a content that rotates a virtual object floating on the water with a water flow, the second is a content that repels many virtual objects descending from the upper part of the screen with a water flow, and the third is a water channel in which various obstacles are set. It is the content that advances the raft by controlling the water flow.

The hardware configuration of the game system according to the present embodiment will be described below with reference to FIG.

FIG. 9 is a block diagram showing a hardware configuration of the game system according to the present embodiment. The game system according to the present embodiment includes a CPU 1 and a memory 2
It comprises an I / F 3 for image input, an I / F 4 for image output, an I / F 5 for input / output, a network I / F 6, a video camera 7, a display 8, and an input / output device 9.

The CPU 1 is a CPU for activating a game program and a command interface program stored in the memory 2, and includes a mapping unit, a mosaicizing unit,
A part of the similarity detection unit, the angle-of-view setting unit, and the command generation unit is configured.

The memory 2 has a frame area, a work area for executing various processes such as a recognition process, and the like.

The frame area is an area for storing original image data fetched through the image input I / F 3 or the network I / F 6, and includes multi-valued R data, multi-valued G data, and multi-valued G data. B data and multi-valued alpha value data are stored in the order of R, G, B,... R, G, B.

The recognition area is an area used for copying image data in order to set an area for performing a recognition process from the frame area. In order to perform a discrete Fourier transform at high speed, the vertical and horizontal directions of the original image data are used. 2 pixels
The storage capacity is set to match the power of. Such an area constitutes a part of the angle-of-view setting means.

The binarized area is the RG stored in the frame area.
This area stores binary data obtained by binarizing the B image data with a threshold value in consideration of not being affected by a change in lighting conditions. This is used in the center of gravity detection processing. Therefore, it constitutes a part of the center of gravity detection means.

The image input I / F 3 is an I / F for taking in original image data from the video camera 7.

The image output I / F 4 is an I / F for outputting the frame data read from the output frame area of the memory 2 to the display 8.

The input / output I / F 5 is an I / F for inputting and outputting data to and from the input device 9.

The network I / F 6 is an I / F for inputting and outputting data to and from a network such as the Internet via a public line or a dedicated line. Using the original image data captured by a mobile phone with a built-in digital camera at a remote location via the network I / F 6, the shape of the object,
The direction and the like can be recognized. This mobile phone with a built-in digital camera corresponds to the imaging means.

The video camera 7 is a home-use DV camera, which captures a human hand above a dark cloth. This corresponds to an imaging unit.

The display 8 is a display device for displaying the contents of the game, and the input device 9 is a keyboard, a mouse, an analog controller, and the like for inputting a personal recognition password and the like.

Next, a schematic process of the game system according to the present embodiment will be described with reference to FIG. FIG. 1 is a flowchart showing the main routine.

When the main routine shown in FIG. 1 is started, the CPU 1 executes a process for initialization (step 1). Specifically, the CPU 1 initializes variables on the program and reserves a frame area and a recognition area as preparation for starting the program.

The CPU 1 fetches the original image data from the video camera 7 via the image input I / F, following the processing of step 1 shown in FIG. 1 (step 2). As a result, the CPU 1 stores the original image data captured in step 2 in the frame area of the memory 2. Thereafter, a process of setting the number of vertical and horizontal pixels of the original image data to a power of 2 is executed. This corresponds to an image clipping step.

After finishing the processing of step 2, the CPU 1 starts an image recognition routine program (step 3). As a result, the main program can acquire the various commands described above.

The CPU 1 executes a processing routine for the game contents (step 4). Thereby, based on the command acquired in step 3, the CPU 1 performs a process for rotating the virtual object floating on the water with the water flow, a process for turning back a large number of virtual objects falling from above the screen with the water flow,
In a waterway in which various obstacles are set, a process for advancing the raft by controlling the water flow is executed. Thereby, the CP
U1 reflects the processing result in step 4 to the memory 2
The frame data is stored in the output frame area.

After the processing in step 4, the CPU 1 outputs the game image stored in the output frame area of the memory 2 to the image output I / F (step 5). Thereby, the image output I / F 4 displays the game image on the display 8.
To be displayed.

The CPU 1 determines whether or not a command to end the game after the processing of step 5 has been input from the input / output I / F 5 (step 6). Note that a command to end the game can be input from the input device 9.

The CPU 1 terminates this routine if it obtains the end command in step 6. If it does not obtain the end command in step 6, the CPU 1 executes the processing from step 2 to step 5 for one inter (about 1/60 sec.). ). The above is the outline of the main routine in the present embodiment.

Next, before describing the image recognition routine in the present embodiment, two-dimensional Fourier transform image data (hereinafter referred to as two-dimensional F
Abbreviated as FT image data. ) Will be described with reference to FIGS.

FIGS. 4 to 8 show the original image data and the two-dimensional FFT.
FIG. 4 is a diagram illustrating a correspondence relationship with image data. 4A shows grayscale original image data obtained by imaging a human hand representing a par, and FIG. 4B shows a discrete two-dimensional Fourier transform of the original image data shown in FIG. Is two-dimensional FFT image data mapped to an orthogonal basis by FIG. 4 (b)
It can be seen that the two-dimensional FFT image data shown in (1) spreads in all directions around the origin of the image center and is point-symmetric about the origin. Therefore, in the present embodiment, the two-dimensional FFT image data of either the left or right half with respect to the origin as a registered pattern is stored in the memory 2.

FIG. 5 (a) shows grayscale original image data of a human hand representing goo, and FIG. 5 (b)
Is two-dimensional FFT image data obtained by mapping the original image data shown in FIG. 5A to an orthogonal basis by a discrete two-dimensional Fourier transform. It can be seen that the two-dimensional FFT image data shown in FIG. 5B is concentrated around the origin of the image center and is point-symmetric about the origin. Therefore, in the present embodiment, the two-dimensional FFT image data of either the left or right half with respect to the origin as a registered pattern is stored in the memory 2.

FIG. 6A shows grayscale original image data obtained by imaging a human hand with the par with the fingers aligned and the par narrowed to the left. FIG. 6B shows the original image data shown in FIG. This is two-dimensional FFT image data obtained by mapping the shown original image data to an orthogonal basis by a discrete two-dimensional Fourier transform. FIG. 6 (b)
It can be seen that the two-dimensional FFT image data shown in (1) is scattered laterally around the origin of the center of the image, and that the scattered shape is rotating counterclockwise. In the present embodiment, the memory 2 stores two-dimensional FFT image data in the left or right half with respect to the origin as a registered pattern.

FIG. 7A shows grayscale original image data of a human hand in which a par with a narrowed par with fingers is arranged without tilting the center of the image, and FIG. 7B is FIG. This is two-dimensional FFT image data obtained by mapping the original image data shown in (a) to an orthogonal basis by a discrete two-dimensional Fourier transform. It can be seen that the two-dimensional FFT image data shown in FIG. 7B is elongated and dispersed on the Y coordinate passing through the origin of the center of the image. In the present embodiment, the memory 2 stores two-dimensional FFT image data in the left or right half with respect to the origin as a registered pattern.

FIG. 8A shows grayscale original image data of a human hand in which a par whose finger is aligned and whose width is narrowed is arranged clockwise from the center of the image.
Is two-dimensional FFT image data obtained by mapping the original image data shown in FIG. 8A to an orthogonal basis by a discrete two-dimensional Fourier transform. It can be seen that the two-dimensional FFT image data shown in FIG. 8B is elongated and dispersed on the Y coordinate passing through the origin of the image center, and is inclined clockwise about the origin. In the present embodiment, the memory 2 stores two-dimensional FFT image data in the left or right half with respect to the origin as a registered pattern.

The image recognition routine according to the present embodiment will be described in detail with reference to FIG. FIG. 2 is a flowchart showing an image recognition routine according to the first embodiment.

When the CPU 1 starts the image recognition routine, the original image data stored in the recognition area of the memory 2 is added to the original image data.
A dimensional Fourier transform is performed (step 31). Such a step corresponds to a mapping step or a mapping step. This is two-dimensional FFT image data in which original image data is mapped to an orthogonal basis by using two-dimensional Fourier transform.
As a result, it is possible to process Fourier-like rough and smooth information like a real creature. Also,
Since the number of pixels of the aspect ratio of the original image data is set to the power of 2 when copying the processing area described above, the discrete Fourier transform can be performed at high speed.

The CPU 1 determines in step 31 that the two-dimensional FFT
The image data is converted into a mosaic (step 32). Such a step corresponds to a mosaicing step or a mosaicing step. As a result, the number of pieces of processing data can be reduced without cutting out the original image data, so that the following processing can be speeded up and an allowable width (robustness) can be provided.

The CPU 1 calculates an error E _(n) between the registration pattern _Pn and the original image data C based on the following equation (1) (step 331).

[0055] Here, n is the identification number of the registered pattern, and is an integer of 1 or more.

When the CPU 1 calculates an error between the original image data C and the registered pattern _Pn based on the equation (1), the CPU 1 obtains n errors E ₍₁₎ , E ₍₂₎ ... E _(n). Can be.

The CPU 1 finds the minimum error among the errors E ₍₁₎ , E ₍₂₎ ... E _(n) calculated in step 331, and recognizes that it is most similar to the registered pattern used to find the error. Get results. This step corresponds to a similarity detection step or a similarity detection step. Such recognition results are various commands determined based on the shape of the hand, its position, and its orientation as described above. The recognition result is returned to the main routine (step 3
4), end this routine. Therefore, this routine corresponds to a command generation step and a command generation step, and constitutes a part of command generation means. The above is the processing procedure of the image recognition routine in the present embodiment, and by adopting such a processing procedure, a high-speed and highly robust recognition processing can be executed.

In this embodiment, a water source command routine for outputting a water source command to the main routine is started in parallel with the image recognition routine. This routine constitutes a command I / F together with the image recognition routine. The water source command routine will be described below.

The water source command routine is a routine for detecting the coordinates of the center of gravity of the hand existing in the original image data in order to detect the position of the hand above the dark cloth. In this center-of-gravity detection processing, the original image data is binarized using a threshold value so as not to be affected by changes in lighting conditions, and then the center-of-gravity position is detected by calculation based on the following equation. The formula for calculating the center of gravity will be described below.

The brightness on the two-dimensional coordinates is represented by a function f (x, y).
Expressed in discrete moment of inertia is represented by _{m ij = Σ y Σ x x} i · y j · f (x, y) Equation (2).

[0061] centroid c is c _{= m} 10 _/ m ₀₀ Equation (3), so when the center of gravity of the coordinates (cx, cy), cx = (Σ y Σ x x · f (x, y) / Σ y Σ _x · f (x, y)) equation _{(4) cy = (Σ y} Σ x y · f (x, y) / Σ y Σ x · f (x, y)) equation (5) (4) (5) is calculated. This routine returns a water source command to the main routine based on the above calculation result.

(Embodiment 2) Hereinafter, the present embodiment
A command I / which recognizes the shape of a human hand included in original image data captured by a video camera and outputs the result of detecting the position of the hand as a command according to the embodiment described above.
F, which is common in that it is a game system in which a game is advanced by a command output from the I / F, so that it has a common configuration. Are omitted, and different configurations will be mainly described with reference to FIGS. 1 and 3 to 9.

The hardware constituting the game system according to the present embodiment includes a CPU 1, a memory 2, and an image input / output I / F as in the first embodiment described with reference to FIG.
Embodiment 1 in that the CPU 1 is a multi-layer neural network comprising an I / F 4, an image output I / F 4, an output I / F 5, a network I / F 6, a video camera 7, a display 8, and an input / output device 9. Is different from

The neural network according to the present embodiment employs a perceptron whose input is an analog value, and is a multilayer neural network including three layers: an input layer, an intermediate layer, and an output layer. BP method (Back Propagation method) as the learning algorithm
Has been adopted.

The input layer is composed of, for example, 64 cells. A half (2 × 8) mosaic of half of the two-dimensional FFT image data (64) is input to each cell (96) of the input layer. Here, the reason why the mosaic is used is to reduce the data to be input to the neural network and have an allowable width (robustness). Since the property of the two-dimensional FFT image data is point-symmetric with respect to the origin, only the two-dimensional FFT image data divided into half from the origin is input to the input layer.

The intermediate layer is empirically set to 30 cells as the number of cells which can correspond to various inputs and can be separated cleanly when the output is about 10 cells. The output layer is the game command I
/ F is set to 10 cells so that 10 types of input commands can be distinguished.

Next, the correspondence between the original image data and the two-dimensional FFT image data will be described with reference to FIGS. FIG.
8 are diagrams showing the correspondence between the original image data and the two-dimensional FFT image data. Since FIG. 4 to FIG. 8 are described in the first embodiment, the two-dimensional FFT image data shown in FIG. 4B is spread in all directions around the origin of the image center. 5 is symmetrical with respect to the center, and FIG.
The two-dimensional FFT image data shown in (b) is concentrated around the origin of the center of the image, and is point-symmetric about the origin.

It can be seen that the two-dimensional FFT image data shown in FIG. 6B is scattered laterally around the origin of the center of the image, and that the scattered shape is rotated counterclockwise. The two-dimensional FFT image data shown in FIG.
It can be seen that they are elongated on the coordinates. FIG. 8B
It can be seen that the two-dimensional FFT image data shown in (1) is elongated and dispersed on the Y coordinate passing through the origin of the center of the image, and is inclined clockwise about the origin.

Next, the general processing of the game system according to the present embodiment will be described. The main routine in the present embodiment is the same as that in the first embodiment, and thus the description is omitted, and the image recognition routine in the present embodiment will be described with reference to FIG.

FIG. 3 is a flowchart showing an image recognition routine according to the second embodiment.

When activating the image recognition routine shown in FIG. 3, the CPU 1 performs a two-dimensional Fourier transform on the original image data stored in the recognition area of the memory 2 (step 3).
1). In this step, two-dimensional FFT image data (gray scale) in which the original image data is mapped to the orthogonal basis can be obtained. As a result, it is possible to simultaneously process not only the information of the Fourier-like roughness and smoothness as in a real creature but also the information of the brightness distribution of the entire screen. Further, as described above, when the processing area is copied, the number of pixels of the aspect ratio of the original image data is set to a power of 2, so that the discrete Fourier transform can be performed at high speed.

The CPU 1 mosaicizes the two-dimensional FFT image data obtained by mapping the original image data on the orthogonal basis in step 31 (step 32). Specifically, two-dimensional FFT
Half of the image data is tessellated to 8 × 8 and input to an input layer consisting of 64 cells. In this way, the sum of the gray scales is obtained for each mosaic unit. Where 2
By mosaicing the dimensional FFT image data, data input to the neural network is reduced, and an allowable width (robustness) is provided. Furthermore, two-dimensional F
Since the properties of the FT image data are point-symmetric with respect to the origin, only the two-dimensional FFT image data divided into half from the origin need be input to the input layer. The number can be reduced and the processing speed can be increased.

The CPU 1 inputs each block value after mosaicing to the neural network, and outputs “probability” for each pattern that has already been registered and learned (step 33). This step corresponds to a similarity detection step or a similarity detection step.

The CPU 1 returns to the main routine the registered pattern with the closest tendency among the output results in step 33 as a recognition result (step 34), and terminates this routine. Such recognition results are various commands determined based on the shape of the hand, its position, and its orientation as described above. Therefore, this routine corresponds to a command generation step and a command generation step, and constitutes a part of command generation means. The above is the processing procedure of the image recognition routine in the present embodiment, and by adopting such a processing procedure, a high-speed and highly robust recognition processing can be executed.

Also in the present embodiment, the water source command routine described in the first embodiment is activated together with the image recognition routine shown in FIG.

[0076]

According to the present invention, there is provided an image recognition method capable of recognizing a shape of an object at a high speed by using a feature obtained by mapping a brightness distribution of original image data to an orthogonal basis, and using the method. An image recognition system, a command interface to which the system is applied, and a recording medium storing a program for realizing the command interface can be provided.

[Brief description of the drawings]

FIG. 1 is a flowchart showing a main routine.

FIG. 2 is a flowchart illustrating an image recognition routine according to the first embodiment.

FIG. 3 is a flowchart illustrating an image recognition routine according to the second embodiment.

FIG. 4 is a diagram showing a correspondence relationship between original image data and two-dimensional FFT image data.

FIG. 5 is a diagram showing a correspondence relationship between original image data and two-dimensional FFT image data.

FIG. 6 is a diagram showing a correspondence between original image data and two-dimensional FFT image data.

FIG. 7 is a diagram showing a correspondence relationship between original image data and two-dimensional FFT image data.

FIG. 8 is a diagram showing a correspondence relationship between original image data and two-dimensional FFT image data.

FIG. 9 is a block diagram illustrating a hardware configuration of a game system according to the present embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 CPU 2 Memory 3 I / F for image input 4 I / F for image output 5 I / F for input / output 6 Network I / F 7 Video camera 8 Display 9 Input device

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G06T 7/60 150 G06T 7/60 150S F term (Reference) 2C001 BA02 BC03 CA00 CA09 CB01 CC02 5B057 AA20 BA02 BA26 BA29 CA01 CB01 CE01 CE09 CG05 CG09 CH03 CH04 DC06 DC09 DC33 5L096 AA02 BA08 BA18 CA02 EA21 EA35 EA43 FA23 FA60 FA69 GA08 HA08 HA11 JA03 LA01 LA15

Claims (11)

[Claims] An image recognition method for recognizing an object included in original image data by comparing two-dimensional FFT image data obtained by mapping the original image data on an orthogonal basis with a registered pattern composed of the two-dimensional FFT image data. A mapping step of obtaining two-dimensional FFT image data from the original image data, a mosaicing step of mosaicizing the two-dimensional FFT image data obtained in the mapping step, A similarity detection step of obtaining a similarity. 2. The image recognition method according to claim 1, further comprising an image cutout step in which the number of vertical and horizontal pixels of the original image data used in the mapping step is a power of two. 3. An image recognition system for recognizing an object included in original image data by comparing two-dimensional FFT image data obtained by mapping the original image data on an orthogonal basis with a registered pattern comprising the two-dimensional FFT image data. A mapping unit for obtaining two-dimensional FFT image data from original image data, a mosaic unit for mosaicing the two-dimensional FFT image data obtained by the mapping unit, and a mosaic data obtained by the mosaic unit and the registered pattern. An image recognition system comprising: a similarity detection unit that obtains a similarity. 4. The image recognition system according to claim 3, further comprising an angle-of-view setting means for aligning the number of vertical and horizontal pixels of the original image data used in said mapping means to a power of two. 5. A command interface for outputting a command based on a result of recognizing a shape or a position of a hand included in original image data captured by an imaging unit,
A mapping unit for obtaining two-dimensional FFT image data from the original image data, a mosaic unit for mosaicizing the two-dimensional FFT image data obtained by the mapping unit, a mosaic data obtained by the mosaic unit, and the registration pattern. A command interface, comprising: a similarity detection unit that calculates the similarity of the command; and a command generation unit that outputs a command based on a result from the similarity detection unit. 6. The command interface according to claim 5, further comprising an angle-of-view setting unit for aligning the number of vertical and horizontal pixels of the original image data used in the mapping unit to a power of two. 7. A computer-readable recording medium in which a program for outputting a command based on a result of recognizing a hand shape or a position thereof included in original image data captured by an imaging unit is recorded, wherein A mapping step of obtaining two-dimensional FFT image data from the image data, a mosaicing step of mosaicing the two-dimensional FFT image data obtained in the mapping step, and a similarity between the mosaic data obtained in the mosaicing step and the registration pattern A computer-readable storage medium storing a program for causing a computer to execute a similarity detection step for determining a degree and a command generation step for outputting a command based on a result from the similarity detection step. 8. The recording medium according to claim 7, further comprising an angle-of-view setting step for aligning the number of vertical and horizontal pixels of the original image data used in the mapping step to a power of two. 9. The image recognition method according to claim 1, wherein the original image data is obtained through a network interface. 10. The apparatus according to claim 3, wherein the original image data is obtained through a network interface.
Or the image recognition system according to claim 4. 11. The apparatus according to claim 5, wherein the original image data is obtained through a network interface.
Or the command interface according to claim 6.