JP2003223642A - Face image recognition system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a face image recognition system capable of matching a face image with enormous images at high speed by applying matched filtering to the preprocessed face image and combining it with a high-speed correlation processor. <P>SOLUTION: This face image recognition system filters the face image formed by preprocessing an inputted image based on edge selection and binarization using the matched filter as a filter for spatial filtering and a recognition result is provided using a detect device receiving the outputted signal using a deviated image by time integration. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a face image recognition system.

[0002]

2. Description of the Related Art Individual recognition using a face image has been attracting much attention as a security technique which can be replaced with an ID card or a personal identification number because the individual can be identified without contact. For this reason, much research has been done centering on digital processing, but an optimal recognition method has not yet been established because a facial image is a three-dimensional object that changes due to facial expression changes and surrounding conditions such as lighting. .

On the other hand, at present, an electrically addressed liquid crystal spatial light modulator that can display at a high speed for 400 μs is on the market, and it is expected that a device with a higher speed response will be developed in the future. However, devices such as ordinary CCD cameras have not kept up with the speed, and a practical high-speed correlation system has not yet been provided.

[0004]

By the way, the matched filter described above can be designed in various ways, and the robustness such as rotation and size can be enhanced. As a result of applying this technique to face recognition that has performed edge extraction, binarization, and half processing, a high recognition rate and exclusion rate can be obtained. Therefore, a practical high-speed recognition system can be realized by performing a high-speed calculation using a high-speed Fourier transform (FFT chip, optical correlation calculation) or the like.

On the other hand, rewriting the database (for example, 4
We propose a technique that can perform the correlation calculation of two-dimensional images as well as the speed of 00 μs). High-speed display SLM (Spatial
This is an optical operation system in which the Light Modulator) and matched filtering are combined and the time integration detect method is further combined. If there is a detect device capable of accurately detecting high-speed display, another detector may be used.

With this method, rewriting of the database and the same speed calculation can be realized. Also, in order to improve robustness, a technology that can perform filtering at the same time as several filters is required, but if arraying a single bare chip as an array light source and performing replication at the speed of light, calculation with multiple filters at the speed of light Is possible. This is a particularly effective system when a very large amount of databases exist.

The inventor of the present application has already found that a face image recognition system using a binary optical image element for parallel correlation calculation (Japanese Patent Application No.
00-307055), and a pattern search method (Japanese Patent Application No. 2000-307058) for performing a search by performing optical coupling conversion correlation calculation in space parallel using a Fourier transform element.

However, in that case, there is a problem that the face recognition processing speed is suppressed by the CCD camera and other optoelectronic devices, and the processing speed cannot be increased even if the number of data to be identified at one time is increased and the processing speed is increased. It was

In view of the above situation, it is an object of the present invention to provide a face image recognition system capable of high speed face image processing.

[0010]

In order to achieve the above-mentioned object, the present invention [1] in a face image recognition system, a face on which an input image is preprocessed based on edge extraction / binarization It is provided with means for filtering an image using a matched filter as a filter for spatial frequency filtering, and a detect device for receiving an output signal by time integration using a shifted image.

[2] In the face image recognition system described in [1] above, matched filtering at the spatial frequency is performed by a high-speed correlation system.

[3] In the face image recognition system described in [2], the fast correlation system is a fast correlation optical system.

[4] In the face image recognition system described in [2], the high speed correlation system is a system using a high speed FFT processor.

[5] In the face image recognition system according to the above [1], the matched filtering at the spatial frequency is performed as a computer graphic or kinoform on a computer by performing an image obtained by subjecting the input image to a specific preprocessing. The feature is that a matched filter is calculated and displayed on a rewritable liquid crystal spatial light modulator, and a huge database is rewritten by a high-speed liquid crystal spatial light modulator for correlation.

[6] In the face image recognition system according to the above [2], the high-speed correlation optical system is a duplicate optical system using a plurality of light sources, and an optical correlator for performing matching with a plurality of matched filters in parallel. Is characterized in that.

With the above arrangement, according to the present invention, the correlation calculation of the two-dimensional image can be performed at the same speed as the rewriting of the database. Specifically, a high-speed display SLM (spatial light modulator) is combined with matched filtering that can be calculated at the speed of light, and a time integration detection method is further combined to build an optical calculation system. The detect device may be any detect device as long as it has a time integration function capable of accurately detecting high-speed display.

Further, in order to enhance robustness, a technique capable of simultaneously filtering with various filters is also adopted. If a single bare chip is arranged to form an array light source and replication is performed at the speed of light, a plurality of filters at the speed of light can be obtained. Calculation is possible. This is a particularly effective system when very large amount of database processing is required.

[0018]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below.

First, the operation of the face image recognition system using matched filtering will be described.

FIG. 1 is an operation flowchart of a face image recognition system using matched filtering according to an embodiment of the present invention, and FIG. 2 is a principle diagram of the optical matched filtering.

(1) In the case of real-time calculation, first, the captured image is subjected to preprocessing such as edge extraction and binarization (step S1).

(2) Next, for the detected image A, a filter which is basically a complex conjugate of Fourier transform is created. Multiplication is performed on the frequency space using this filter, and inverse Fourier transform is performed (step S2).

(3) After that, the correlation signal is acquired and used as the recognition result (step S3). The Fourier transform here may use a high-speed FFT processor or an optical system.

FIG. 2 shows the principle of optical matched filtering.

As shown in this figure, input f on the P1 surface
Place (x, y) and illuminate this with coherent parallel light B
Reveal Obtain the Fourier transform of this signal with lens 1 and
Place the H filter here and again use the lens 2 to change the Fourier transform.
Then, the result is observed on the output surface P3. At this time, the output surface P
The signal obtained in 3 is expressed as in equation (1) g ′.
Then, the convolution of g and h is obtained. However,
h (x, y) is H (v_x, V _y) Is the inverse Fourier transform of
Is called the point response function of the filter. Patta
As shown in FIG. 1, the filter for the scene detection is a Fourier filter.
D) It is a complex conjugate of the transformation. The correlation signal detection position is filled
It can be changed depending on the design of the computer and the position of the detected image.

[0026]

[Equation 1]

(Preprocessing) The preprocessing method of the calculated face image will be described. The same process is performed on all face images (input image and reference image) used for the calculation.

FIG. 3 shows the pre-processing process of the face image.

First, in order to obtain only face information from a human image taken in from a digital camera to a computer, a square is cut from the upper part of the eyebrows to the lower part of the chin [Fig. 3].
(See (a)]. Next, it is normalized to a size of 128 pixels × 128 pixels which is an image size required for recognition [see FIG. 3 (b)]. After that, pre-processing of edge extraction and binarization is performed, and half processing is also performed if necessary [Fig.
(See (c)].

(Example of Post-Processing) FIG. 4 is a diagram showing a method of identifying a registered person, and FIG. 5 is a diagram showing a method of identifying an unregistered person.

These figures show an example of a method of identifying a registered person and an unregistered person when each is an input image, which will be described below.

Let i be the code number of the input image, j be the code number of the reference image in the database, and N be the number of reference images.
(In this case, N = 10). The horizontal axis of the graph is the code numbers 1 to N of the reference image, and the vertical axis of the upper graph is the input image c.
The measured value P _ij of the correlation signal strength between the reference image code # j and the reference image code # j.

When the registrant (code # 1) is input, the autocorrelation value P _i1 takes a higher value than the others.
On the other hand, when an unregistered person is input, the person himself / herself does not exist, so that any P _ij takes a similar low value. Here, it is the lower graph that normalizes P _ij with the maximum value P _imax of P _{i1 to} P _iN.
It's called mapping. Correlation
In mapping, the maximum value (=
It can be seen that the average values other than 1) are high, and the values are low for unregistered persons. This is the input image code #
The comparison value C _i for _i is expressed as in Expression (2).

[0034]

[Equation 2]

If C _i is higher than the initially set threshold value, the input image code # i is identified as an unregistered person, and if it is low, the input image code # i is identified as a registrant, and if it is recognized as a registrant. , P _imax , it is possible to identify the individual as the reference image.

(High Speed Correlation Calculation by Optical System) FIG. 6 is a block diagram of a high speed optical correlation face image recognition system according to the present invention, and FIG. 7 is an operation flowchart of the high speed optical correlation face image recognition system.

In FIG. 6, 11 is a VCSEL array (array light source in which single bare chips are arranged), 12 is B
ZPA1 (binary array optical element), 13 is BZPA
2 and 14 are lenses L1 and 15 are FLC (ferroelectric liquid crystal:
Ferroelectric Liquid Crystal
al) -SLM, 16 is a beam splitter, 17 is a lens L2, 18 is BZPA4, 19 is an LC-SLM MF
Array (matched filter), 21 is BZPA5, 22
Is a detect device having a time integration function, and 23 is P
C (personal computer), 24 is a digital camera connected to a PC, and 25 is a database of the PC.

A high-speed correlation calculation can be performed by the optical system as shown in FIG.

An outline of the operation of this optical system will be described.

The input image is acquired from the digital camera 24 connected to the PC (personal computer) 23.

The matched filter is calculated by the PC 23. The calculated matched filter is displayed on the LC-SLM 19.

The database 25 is displayed on the FLC-SLM 15.

The VCSEL array light source 11 is set to BZPA.
1 (12) collimates, BZPA2 (13) and lens L1 (14) are pinched | interposed, and it irradiates to FLC-SLM15.
The reading is performed by the reflection type, and the duplication is performed by the lens L2 (17). Matched filter displayed on the LC-SLM 19 by performing Fourier transform on the BZPA4 (18) (input image filter which is face information from a human image captured by the computer 23 from the digital camera 24)
Pass through. After passing through BZPA5 (21) again and performing an inverse Fourier transform, a correlation signal is acquired.

This correlation process is repeated for the number of databases 25. Here, the database 25 is an image that has been preprocessed in advance and is stored in the RAM.

Finally, the signal is detected and post-processed to obtain a result output. At this time, the detect method may use time integration.

(Sequential correlation flow) Then, VCSEL
The light beam emitted from the array (array light source) 11 is BZP.
It is collimated by A1 (12) and BZPA2 (13) and becomes a luminous flux of about 2 to 3 mm, and the lens L1 (14)
To form an image on the FLC-SLM 15. This light is FL
The light is reflected from the C-SLM 15, passes through the beam splitter 16, and is collimated by the lens L2 (17) with the image information (refer to the duplication method by the coherent light source), and the same number of images as the VCSEL array 11 is duplicated.

This duplicated image is Fourier transformed by BZPA4 (18) and passed through the matched filter displayed on the LC-SLM 19. The frequency response of the basic matched filter is the Fourier transform and complex conjugate of the detected pattern (see Correlation operation by matched filtering). An image captured by the digital camera 24 is subjected to preprocessing such as edge extraction and binarization, and a PC (computer) 23 produces a phase matched filter.

Here, in order to enhance robustness, rotation,
Filters having invariants such as size are also prepared, and the duplicate image is passed through each filter. Again BZPA5 (2
By performing the Fourier transform in 1), the correlation signal is obtained by the detect device 22 having a time integration function. FLC
-It is possible to perform the calculation at the speed of light from the time the image is displayed on the SLM 15 until the correlation signal is obtained, and the system is capable of ultrahigh-speed correlation calculation.

As shown in FIG. 7, the general process is to obtain an input image (step S11), calculate and display a matched filter (step S12), and perform an optical correlation calculation (step S13). The result is output (step S14).

The optical correlation calculation (step S13) is
More specifically, the database is displayed on the LC-SLM 15 through several loops of the database, passed through an input image filter, and a signal is acquired by a detect device 22 having a time integration function.

(Example of VCSEL array light source for duplicate) FIG. 8 shows an example of a VCSEL array light source. This is an array light source in which VCSEL single bare chips are arranged on a substrate in a size of about 1 to 2 mm, and is not two-dimensionally manufactured on a silicon substrate. The structure is composed of a Peltier device 31, a VCSEL array 32, a mask 33, and a lens array (BZPA) 34, and it is possible to generate beams of the number of lens arrays. The distance R between the VCSEL array 32 and the lens array (BZPA) 34 is about 5 to 13 mm.

(Detection Method Using Time Integration) The time integration detection method is a method in which a receiving side (detection device) accumulates a plurality of output signals for a certain period of time and outputs them as one image.

In particular, in the correlation calculation, a plurality of image data input at high speed is converted into a correlation distribution, and the distributions are accumulated for a certain period of time and output as one image.

FIG. 9 is a schematic diagram showing a detect method using temporal integration according to the present invention. In addition, here
As a photodetector, a commercially available CCD (Charge Coupled Device) having a charge function (integration function) is used.
ce) can be used.

For example, as shown in this figure, the input is T
When 0, T1, T2, ... Tn, I0, I1, I2, ...
Correlation processing is performed on the In image data, and f (I0), f
Output as (I1), f (I2), ... F (In).
The output data corresponding to the input number of sheets subjected to the correlation processing are superimposed on each other during the detection device response speed (T0 to Tn) on the output surface, and output as one image.

The input images are arranged so that the outputs of the local correlation distributions do not overlap. In this way, the correlation distribution of each input is temporally integrated during a certain time (response speed time of the detect). This means that high-speed image processing is possible even when the input signal interval is faster than the output detect response speed.

For example, if the input signal is 500 μs (2000
Hz), and when the detect device is 5 ms (200 Hz), the time taken for each detection element is 5 ms.
There are 10 inputs in between. At this time, a signal is output at different positions depending on the input. By using this method and shifting the position ten times, ten output signals can be obtained.

(Edge Extraction Processing) The edge extraction processing corresponds to checking the luminance change rate of the two-dimensional image, and is obtained by a differential operation. However, because it is processed on a computer, it is considered discretely and the differential is treated as a difference. Original image f (i,
j) the eight neighboring pixels of the target pixel (i, j) and the 3 × 3 matrix operator A B C D E F G H I represented by the formula, the new gray scale number g (i, j) is g (I, j) = A * f (i-1, j-1) + B * f (i, j-1) + C * f (i + 1, j-1) + D * f (i-1, j) + F * It can be expressed as f (i + 1, j) + G · f (i−1, j + 1) + H · f (i, j + 1) + I · f (i + 1, j + 1), which serves as a filter. If attention is paid to the brightness gradient, the first derivative is sufficient. Among them, in the Sobel filter, the weight of four pixels close to the image of interest is large, so that the edge is further emphasized. Therefore, the Sobel filter is used as the edge extraction in the preprocessing of this face image recognition system.
Table 1 shows the matrix operators and filter expressions of Sobel filters in the horizontal and vertical directions.

[0059]

[Table 1]

By combining g _H (i, j) and g _V (i, j), the two-dimensional Sobel filter is defined by the following equation.

G (i, j) = | g _H (i, j) | + | g
_V (i, j) | (Binarization processing) Binarization means that each pixel of a grayscale image is converted into 0, 1 (black and white) based on the magnitude relationship with an appropriate reference value.
The gradation is defined and f (i, j) is defined as g (i, j), which can be defined as in Expression (3).

[0062]

[Equation 3]

The binarization has an effect of removing excess noise and clarifying the shape, and thus is an essential element for preprocessing. The methods can be roughly classified into two types, a fixed threshold method and a region designation method. The former is a method of determining a threshold value with a certain value of gray scale gradation, and the latter is a method of first determining a binary ratio of the entire image and then determining the threshold value from a histogram of density values. We decided to use the area designation method to keep the light intensity of each image constant.

(Optical duplicate) An image can be optically reproduced by using an array lens or a diffused light source. This can substantially increase the number of pixels. In the optical correlation system, by duplicating the input image, it is possible to pass through a plurality of matched filters and enhance robustness.

(Coherent Duplicate Optical System) FIG. 10 is a diagram showing a coherent duplicate optical system using a transmissive input element according to the present invention.

As shown in this figure, the VCSEL array 4
1 is collimated by a BZPA (lens array) 42 to create a multi-beam. This is again passed through the BZPA (lens array) 43, the lens 44, and the duplicated image 45, and duplicated images 48 of the same number as the lenses 46 and BZPA (lens array) 47 are created again.

The VCSEL is a vertical cavity surface emitting laser (Vertical Cavity Surf).
ace Emitting diode Laser)
Is the abbreviation for, and the name is derived from the fact that the direction in which light resonates is perpendicular to the substrate surface. A conventional edge-emitting laser diode is composed of a horizontal resonator having an end face, which is cut out by cleaving a GaAs substrate or the like, as a reflection mirror, and emits from the cleaved end face.

However, in such a structure, the light emission point 2
While it is said that dimensional arraying is difficult, VCSEL
Due to the structure, there is a merit that elements can be freely produced within the substrate surface, and an array having a complicated design pattern can be easily produced. Therefore, it is attracting attention as a light source for many applications such as optical interconnections requiring a two-dimensional array light source, light sources for parallel optical logic devices, and so on.

Table 2 shows the specifications of the FLC-SLM.

[0070]

[Table 2]

The driving method of the FLC-SLM used is different from that of the conventional image display panel, and LCOS (Liquid) which can directly send image information from the back surface to each pixel.
(Crystal on Silicon: shows a liquid crystal display built on a silicon substrate) An active matrix driving system is used, and high-speed display at 400 μs is possible.

The present invention is not limited to the above embodiments, and various modifications can be made based on the spirit of the present invention, and these modifications are not excluded from the scope of the present invention.

[0073]

As described in detail above, according to the present invention, the following effects can be achieved.

(A) A face image recognition system capable of high-speed face image processing can be constructed.

(B) By applying the matched filtering to the preprocessed face image and combining it with the high-speed correlation processor, it is possible to perform high-speed matching with a huge amount of images.

(C) Rewriting the database (for example, 4
Correlation calculation of a two-dimensional image can be performed similarly to the speed of 00 μs). By combining the high-speed display SLM and the matched filtering, it is possible to construct an optical operation system that further combines the time integration detect method.

With this method, rewriting of the database and the same speed calculation can be realized. Also, in order to improve robustness, a technology that can filter several filters at the same time is required, but if arraying a single bare chip as an array light source and performing replication at the speed of light, calculation with multiple filters at the speed of light is possible. Is possible.

[Brief description of drawings]

FIG. 1 is an operation flowchart of a recognition system using matched filtering according to the present invention.

FIG. 2 is a principle diagram of optical matched filtering of the present invention.

FIG. 3 is a diagram showing a preprocessing step of a face image.

FIG. 4 is a diagram showing a method of identifying a registrant.

FIG. 5 is a diagram showing a method for identifying an unregistered person.

FIG. 6 is a configuration diagram of a high-speed optical correlation face image recognition system according to the present invention.

FIG. 7 is an operation flowchart of the high-speed optical correlation face image recognition system according to the present invention.

FIG. 8 is a diagram showing an example of a VCSEL array light source according to the present invention.

FIG. 9 is a schematic diagram showing a detect method using temporal integration according to the present invention.

FIG. 10 is a diagram showing a coherent duplicate optical system including a transmissive input element according to the present invention.

[Explanation of symbols]

1,2,44,46 lens 11,32,41 VCSEL array 12 BZPA1 13 BZPA2 14 Lens L1 15 FLC-SLM 16 beam splitter 17 Lens L2 18 BZPA4 19 LC-SLM MF array 21 BZPA5 22 Detect device with time integration function 23 PC (personal computer) 24 digital camera 25 database 31 Peltier element 33 mask 34, 42, 43, 47 BZPA (lens array) 45 Reproduced image 48 duplicate image

Continued front page    F-term (reference) 5B043 BA04 EA10 EA18 GA04 GA15                 5B057 CA12 CA16 CE06 CH09 DA11                       DB02 DC30 DC36                 5L096 BA18 EA43 FA06 FA23 FA34                       GA49 HA07 JA03 LA01

Claims (6)

[Claims] 1. A means for performing filtering using a matched filter as a spatial frequency filtering filter for a face image which has been preprocessed based on edge extraction / binarization in the input image, and a shifted image. Face image recognition system including a detect device that receives an output signal by time integration using the. 2. The face image recognition system according to claim 1, wherein the matched filtering at the spatial frequency is performed by:
A facial image recognition system characterized by a high-speed correlation system. 3. The face image recognition system according to claim 2, wherein the fast correlation system is a fast correlation optical system. 4. The face image recognition system according to claim 2, wherein the high-speed correlation system is a system using a high-speed FFT processor. 5. The face image recognition system according to claim 1, wherein the matched filtering at the spatial frequency is performed by:
An image obtained by subjecting the input image to specific preprocessing is calculated as a computer graphic or kinoform on a computer, a matched filter is calculated and displayed on a rewritable liquid crystal spatial light modulator. A face image recognition system characterized by rewriting with a modulator to perform correlation. 6. The face image recognition system according to claim 2, wherein the high-speed correlation optical system is an optical correlator that performs matching with a plurality of matched filters in parallel by a duplicate optical system using a plurality of light sources. A facial image recognition system characterized by the above.