WO2002069266A1 - Face extracting method, device therefor, storage medium, and program - Google Patents

Face extracting method, device therefor, storage medium, and program Download PDF

Info

Publication number
WO2002069266A1
WO2002069266A1 PCT/JP2001/001541 JP0101541W WO02069266A1 WO 2002069266 A1 WO2002069266 A1 WO 2002069266A1 JP 0101541 W JP0101541 W JP 0101541W WO 02069266 A1 WO02069266 A1 WO 02069266A1
Authority
WO
WIPO (PCT)
Prior art keywords
face
eye
position
image
extraction
Prior art date
Application number
PCT/JP2001/001541
Other languages
French (fr)
Japanese (ja)
Inventor
Nobuyuki Matsui
Takeshi Torada
Original Assignee
Step One Co., Ltd.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Step One Co., Ltd. filed Critical Step One Co., Ltd.
Priority to PCT/JP2001/001541 priority Critical patent/WO2002069266A1/en
Publication of WO2002069266A1 publication Critical patent/WO2002069266A1/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/00221Acquiring or recognising human faces, facial parts, facial sketches, facial expressions
    • G06K9/00228Detection; Localisation; Normalisation
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/70Determining position or orientation of objects or cameras
    • G06T7/73Determining position or orientation of objects or cameras using feature-based methods

Abstract

A digital color image is inputted through an image input unit such as a digital camera. The contour area of a face is extracted using a face template from the inputted digital color image. The accurate positions of the eyes are determined from the face contour area, and the accurate face area is extracted according to the distance between the eyes. Thus the face area can be extracted with high accuracy without being influenced by the individual difference, the light source, the color adjustment of the digital camera in a significantly short necessary time.

Description

Specification face extraction method, apparatus, storage medium, and program art

This invention relates to a method for to extract 'faces from among a color image, the apparatus, storage medium, and a flop port grams. BACKGROUND

Taken to the computer via an image input device such as a digital camera color image. Extracting the human face from are required in various fields.

For example, is required as pretreatment for authenticating an individual is in the security field, and in monitoring the field is required in order to improve the safety when promoting automation, the face of the driver in the automobile field (pupil among them) detects the movement line automatically braking. is required in order to I, in the field such as a TV phone is required in order to carry out an effective sound collection to detect the position of the speaker, amusement areas required for other recognizes the face of the interlocutor like Hiyumanoi Doyape Ttorobotto in.

To satisfy such requirements, conventionally,

(1) Digital cameras such as a color image captured in the computer (for example .. In, RGB image) to extract the color of a pixel close to the skin color or skin color from among concatenates pixels extracted Ri to Lavelin grayed, how to the connection region and face region, and

(2) (for example, RGB image) a digital camera such as a color incorporated into the computer image to generate a luminance value image by performing luminance value conversion process on the luminance values: standard face template for an image method of extracting a face area subjected to Matsuchingu treatment with bets have been proposed.

When extracting a face region by adopting the method (1) are fair-skinned, Individuals vary such color black, the influence of the light source (light intensity changes in), receiving the influence of color adjustment such as a digital camera extraction accuracy of the face region Te is disadvantageously lowered. In addition, incorporated

5 was a color image, whether we it is common that contains in addition to miscellaneous background face, will be summer long time required to extract a face area, high-speed processing is required It can not be applied only to the absence of the field.

When the '' adopts the method (2) for extracting a face region, as compared with the case where the or or using color ft paper, individual differences, the light source influences, such as color adjustment 0, such as a digital camera although less susceptible to the influence, the luminance value greatly affected by the amount of light

• rice, extraction accuracy of the face region is disadvantageously lowered. Further, since it is common that contains in addition to miscellaneous background face to the luminance value image image, it will be summer long time required to extract a face region, the field of high-speed processing is not required It can not only be applied to. Furthermore, field 5 extracts the face region using the luminance value image - in case, in order to normalize the size, it is necessary to perform matching using a plurality of templates, to perform an enormous amount of repetitive processing since the time required for to extract the face area will be summer more long, areas of applicability will be more limited.

. The present invention has been made in view of the problems described above, individual differences, the influence of the light source, to extract 'a face region with high accuracy without being affected by the color adjustment, such as 0 digital force camera can, aims Moreover face extraction method capable of greatly reducing the time required, that device storage medium, and a program.

'Invention Me disclosure

5 '- wherein face extraction method of paragraph 1, creates a converted image luminance value by performing luminance value conversion on the color image including a face region, the luminance value converted image, one using adjacent pixel information derivative create an image is a method of extracting a face region by performing a matching processing using the first-order derivative face template.

The method of face extraction Claim 2, in order to create a first derivative image, a method to ignore the following points higher luminance value of the primary differential face template.

The method of face extraction claim 3 is a method of extracting a face area have use the first derivative 顏 template lighter expressed dimmed rate. '

The method of face extraction claim 4, the 顏領 area extracted by the method of any of claims 1 to 3, extracted L, extracted eye candidate position a point corresponding to the minimum luminance value as eye candidate positions based on, it performs template matching to extract the eye position, a method of extracting a face by performing normalization of a face based on the extracted to the distance between both eyes. Face extraction method according to claim 5, divides the 顏領 area extracted by the method of any of claims 1 to 3 on the left and right, Meko complement positions a point corresponding to the lowest luminance value in each partitioned area it is a method for extracting as'.

Face extraction method according to claim 6, divides the face region extracted by the method of any of claims 1 to 3 on the left and right, to analyze the minimum luminance position in each partitioned area, the template from the analysis information perform eye candidate location expression using the average luminance value by extracting eye candidate position by matching a way to extract the position of the eye by performing template matching of the eye.

Face extraction method according to claim 7 is a method for extracting low brightness set as eye position. The face extraction apparatus of claim 8, and the luminance value converted image generating means for generating transformed image luminance value by performing luminance value conversion on the color image including a face region, the luminance value converted image, adjacent rain Motokan information a first derivative image generating means for generating a first derivative image with those comprising a means out face region extraction for extracting a face region by performing a matching processing using the first-order derivative face template.

'顏抽 detection device according to claim 9, as the next differential image forming means, in order to create a first derivative image, and adopted for the ignore following points higher luminance value of the primary differential face template.

Face extraction apparatus according to claim 1 0, as the facial area extracting unit, and adopts the extracts a face region using first derivative face template lighter expressed dimmed rate. Face extraction apparatus of claim 1 1, from the extracted issued the face region by what Kano apparatus according to claim 1 0 of claims 8, Meko auxiliary position for extracting a point corresponding to the minimum luminance value as eye candidate positions extraction means, based on the extracted eye candidate positions, extracts the eye position extracting means for extracting the eye position by performing the template matching, the means pursuant face the normalization of the face image based on the extracted to the distance between the eyes it is intended to include a face extraction unit.

Section 1 2 of the face extraction device according as the eye candidate position extracting means divides the face region extracted by either device 請 Motomeko 1 0 of claims 8 to the left and right, each segment region. Contact a point corresponding to the minimum luminance value have is to employ what is extracted as eye candidate position.

Minimum face extraction apparatus according to claim 1 3, wherein the eye candidate position extraction means and have your face region extracted by any method 請 Motomeko claims 1-3 divided into right and left, each segmented region adopted to include analysis means for analyzing luminance position, the eye candidate position location expression means for performing eye candidate location expression using the average luminance value by extracting Meushitora candidate position by the template matching from the analysis information, the as the eye position extracting means, and adopts what performing template one bets matching eye extracts the position of the eye.

Face extraction apparatus according to claim 1 4, as the eye position extracting means, and adopts a extracts a low brightness set as the eye position.

Storage medium of claim 1 5, in which a computer program for executing any of the processing procedure of claims 1 to 7 is stored.

'Program according to claim 1 6 is used to perform any of the processing procedures according to claim 7 to a computer from claim 1. If a face extraction method of claim 1, to create a converted image luminance value by performing luminance value conversion on the color 'image containing 顏領 region, the luminance value converted image, one using adjacent pixel information derivative create an image, first derivative from the matching processing using the face template is to extract the means pursuant face area, individual differences, the influence of the light source, significantly eliminate highly accurately the effects such as color adjustment, such as a digital camera If it is possible to extract a face region in both can significantly reduce the time required to extract a face region. If the face extraction method according to claim 2, primary order to create a differential image, since it is to ignore the following points higher brightness values ​​of the primary finely divided face template, a face by eliminating the influence of miscellaneous background image area When the accuracy of extracting can be further enhanced with the 'monitor, the time required to extract a face region can be greatly shortened.

If a face extraction method of claim 3, the primary differential face templates one was lightly expressed dimmed rate, since it is to extract the face region using preparative pressurized example the effect of claim 1 or claim 2, comparative it can be compared certain direction relative to the control.

If a face extraction method according to claim 4, any one of the face region extracted by the method of claims 1 to 3, a point corresponding to the minimum luminance value is extracted as the eye candidate position, issued extracted eyes based on the candidate position, to further accurately extracted face region of because it extracts a paragraph shall face the normalization of the face image based on the distance between both eyes extracted to extract the eye position by performing template matching it is, it is possible to greatly reduce the time required to extract a face region.

If a face extraction method of claim 5, divides the face region extracted by the method of any of claims 1 to 3 on the left and right eye candidate position a point corresponding to the lowest luminance value in each partitioned area the because it extracted as, in addition to the operation of claim 4, the time required to 'because to extract a face region can be more greatly reduced.

. If the face extraction method according to claim 6, by dividing the one of the face region extracted by the method of claims 1 to 3 on the left and right, to analyze the minimum luminance position in each segmented region, from the analysis information It extracts eye candidate position by template matching performed eye candidate location expression using the average luminance value, by performing template matching of the eye since it is to extract the position of the eye, to the operation of claim 4 or claim 5 in addition, it is possible to further increase the extraction accuracy of the face region.

If a face extraction method of claim 7, since Ru Nodea for extracting low brightness set as eye position, in addition claims 4 to any one of the operation of claim 6, and the extraction accuracy of the eyeball position high Melco it can be, it is possible to increase the extraction accuracy of the turn face area. .

If 释抽 detection device of claim 8, the luminance value converted image creation means creates the converted image luminance value by performing luminance value Hen换 the color image including a face region, the first derivative image generating means, from the luminance value conversion image, it is possible to create a first derivative image using adjacent pixel information, extracts tae, the means pursuant face area pine quenching process using the following differential face template to the face region extracting means .

Therefore, individual differences, the influence of light, the influence such as color adjustment, such as digital cameras is greatly eliminated it is possible to extract a face region with high accuracy, significantly the time required to extract a face region it can be shortened.

If the face extraction apparatus according to claim 9, wherein the first derivative image generating means, in order to create a first derivative image, since it is to adopt shall ignore the following points higher luminance value of the primary differential face template, miscellaneous by eliminating the influence of the background image makes it possible to further enhance the extraction accuracy of the face regions, it is possible to greatly reduce the time required to extract a face region.

If the face extraction apparatus according to claim 1 0, wherein the face region extracting means, the first derivative face template lighter expressed dimmed rate Yore, because there since adopting extracts a face region Te, claim 8 or in addition to the operation of claim 9, it is possible to compare the predetermined direction with respect to the comparison.

If the face extraction apparatus of claim 1 1, the eye candidate position extraction means, from I 'derived out the face region by any of the apparatus of claim 8 or we claim 1 0, to correspond to the lowest luminance value extracts that point as eye candidate positions, by the eye position extraction means, based on the extracted eye candidate position by performing template matching to extract the eye position, the distance between both eyes by connexion, and extracted in the face extraction means it can be extracted face by performing normalization of a face based on. .

Therefore, it is possible to extract more accurately the face area, the face area - can greatly reduce the time required to extract.

If the face extraction apparatus according to claim 1 2, as the eye 镇補 position extracting means divides the 'extracted face region by any of the apparatus of claim 1 0 of claims 8 to the left and right, each segment region from a point corresponding to the brightness value since it is the you adopt what is extracted as eye candidate position, in addition to the function of the 請 claim 1 1, further possible to greatly reduce the time required to extract a face region in can. '·,

Minimum if the face extraction apparatus according to claim 1 3, as the eye candidate position extracting means divides the face region extracted by the method of any of claims 1 to 3 in the left and right, each segment area adopted to include analysis means for analyzing luminance position, the eye candidate location expression means for performing eye candidate location expression using the average luminance value by extracting eye candidate position by the template pine quenching from the analysis information, the eye as a position extracting means, from performing template matching of the eye is to employ those that extracts the position of the eye, in addition to the effect of claim 1 1 or claim 1 2, further increase the accuracy of extracting the face region this door can be.

If the face extraction apparatus according to claim 1 4, as the eye position extracting means, from the low luminance set is to employ what is extracted as eye position, the any one of claims 1 1 of claims 1 to 3 acting in addition to, it is possible to improve the extraction accuracy of the eye position, it is possible to increase the extraction accuracy of the turn face area.

If a storage medium according to claim 1 5, because the computer program for executing any of the processing procedure of claims 1 to 7 is stored, by executing a computer this Konbyu over data program, wherein can be achieved 'the action similar to those of one of the 請 Motomeko 7 from claim 1.

If claims 1 to 6 programs, since it is the computer of claims 1 to support execution what Re of processing procedures according to claim 7, to achieve the same as the action with any of claims 1 to 7 be able to. BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is a Proc diagram showing one embodiment of a face extraction apparatus of the present invention. FIG. 2, Ru Furochiya one Todea illustrating an embodiment of a face extraction method of the present invention.

Figure 3 is a flowchart for explaining the processing in step SP 2 of Furochiya Ichiboku of Figure 2 in detail.

Figure 4 is a Furochiya one you want to partially explain in detail the process of step SP 3 of the flowchart of Figure 2.

Figure 5 is a Furochiya one you want to explain in detail the remainder of the processing in step SP 3 of the flowchart of Figure 2.

6 is a diagram illustrating a specific example of order of processing to obtain a reduced image by applying the average luminance process the luminance value image.

7 is a diagram for explaining a specific example of a process for obtaining the first derivative image, and a specific example of the process to ignore the pixel that Ru exceeds the differential maximum value.

Figure 8 is Ru FIG der illustrating a specific example of processing for creating a first derivative face template.

9 is a diagram illustrating a process of scanning the first derivative face template primary differential image.

First, .0 is a drawing for explaining a practical example of an extended, division processing of the face outline area. The first 1 is a diagram showing an example of the minimum point group.

The first 2 is a diagram showing an example of a minimum luminance template of the eye.

The first 3 is a drawing for explaining a specific example of a process of scanning a minimum point group with minimum luminance template of the eye. The first 4 is a diagram illustrating a specific example of a process of extracting the eye regions.

The first 5 is a drawing for explaining a specific example of a process of extracting the eye regions more accurately. The first 6 is a diagram illustrating a specific example of extraction of the face region based on the distance between both eyes. BEST MODE FOR CARRYING OUT THE INVENTION

Hereinafter, with reference to the accompanying drawings, illustrating the status bar of the implementation of face extraction method and apparatus of the present invention in detail.

Figure 1 is a proc diagram showing an embodiment 觴様 face extraction apparatus of the present invention.

The face extraction apparatus includes an image input unit 1 such as a digital force camera, digital color image inputted by _ Ri in the image input unit 1 (eg, RGB image) the luminance value image by performing luminance value conversion on a luminance value image creating section 2 for creating, on the basis of the luminance value image, as one 'next differential image producing unit 3 for creating a first derivative image by calculating a brightness difference using the information between neighbor pixels, first derivative a first derivative face template holding portion 4 for holding the face template, a face outline area extraction unit 5 for extracting an outline region of the face by performing matching processing on the first derivative image using the first derivative face template , the minimum luminance value point extracting unit 6 for extracting a point of minimum luminance value from the outline area of ​​the face in the luminance value image, the eye extraction unit 7 for extracting a portion minimum luminance value point is concentrated as eye, It detects the distance between both eyes It has a interocular distance detecting section 8, and a 顏領 region extracting unit 9 for extracting a face region by performing normalization processing on the outline area of ​​the face based on the interocular distance. FIG. 2 is a Furochiya one Bok illustrating an embodiment of a face extraction method of the present invention schematically. 'In step SP 1, the image input unit such as a digital camera receives the digital color image (eg, RGB image), in step SP 2, input' outline of the face using a face template based on the digital color image was extract the area,

In 'SP 3, extract the exact position of both eyes from the coarse regions of the face, in Sutetsu flop SP 4, extracts the correct face area based on the distance between both eyes.

Figure 3 is a flow chart illustrating the process of step SP 2 of the flow chart of FIG. 2 in detail. In step SP 1, (as the brightness value, for example, employing a Y value in the YIQ conversion) RGB images to create a 'go-between luminance value image by the applying the luminance value conversion processing for, in step SP 2, obtain a reduced image by Rukoto to apply the average luminance process the luminance value image, in step SP 3, to create a first derivative image by calculating a luminance difference with the adjacent picture Motokan information in the reduced image. The 'The differential operator during use, for example, R 0 BERTS operator prior public can be exemplified.

. Then, in step SP 4, and extracts one previously created derivative face template or et derivative maximum value. It should be noted that the creation of the first-order differential face template, for example, to create by merging the primary differential face sample of multiple. Moreover, the first derivative face template may be preferably to keep lighten each pixel using a light color ratio, camera characteristics such as by the first derivative template image than the detected image image in order to absorb the sharp differences in the image the general pixel immobilized low position, it is possible to improve the comparison accuracy.

Then, in step SP 5, among all the pixels included in the first derivative image, ignoring the pixel corresponding by replacing the values ​​of the pixels exceeding the differential maximum value to 0 (by the carrying out this process' Ri , there is no high edge representation behind, it can be prevented 謌識 erroneous cormorants have a face outline area deviates from actual face by template matching processing to be described later), in step SP 6, first derivative face template the first derivative image take matching by scanning by, in step SP 7, the most consistent high degree of position detection as approximate position of the face (specifically, + for example, a simple scalar base-vector of the first derivative face template distance is calculated, and by the detection to) that the minimum value as the approximate position of the face, to extract the outline area of ​​the face, its series of remains The process is terminated.

Figure 4, Oyopi FIG. 5 is a flowchart for explaining the processing in step SP 3 of Furochiya Ichiboku second diagram details.

In step SP 1, extracts a rectangle corresponding luminance value image into the face outline area, in step SP 2, the extracted rectangular extended vertically and horizontally, and have contact to step SP 3, the extended rectangular area in a cross to select a compartment area of ​​the upper left is partitioned.

Then, in step SP 4, calculates a first luminance threshold value for detecting a typical brightness difference and want eye luminance values ​​and the eye at the center of the cross, Te step SP 5 smell, defined areas of the selected rectangle extracts the pixel (minimum luminance point) having a luminance value of 晕小 each column, among the extracted minimum brightness point, stores only minimum luminance point having a lower luminance value than the first luminance threshold value as the minimum point , to create a minimum point group (see white indicating portion during the first 1 Figure), in step S f 5 6, minimum luminance template of a typical eye previously prepared minimum point group created (e.g., the scanning with 1 reference 2 Figure). Here, as a scan method, for example, it is preferable to employ a method of using an AN D condition of the minimum point, simply glasses vines and eyebrows when determining the position of the eye on the basis of only the scan results, hair hairline it is possible to prevent a disadvantage that erroneous detection or the like. Then, in step SP 7, from the region above the threshold degree of match, which is preset for the scanned region, it extracts an area of ​​up to three from the upper area as eye candidate regions.

In step SP 8, the eye candidate region is judged whether there exist a plurality of, in the case where the eye candidate region is determined to be more present, in step SP 9, corresponding to each eye candidate region luminance value image image calculates the average luminance value in the region, 3/4 and 1/2 and the second average luminance value is set to the third luminance threshold, step SP Te 1 0 smell, each eye candidate region luminance value image second luminance value of each pixel of the corresponding region in, compared with the third luminance threshold and three values ​​represent, in step SP 1 1, and 3 values ​​expressed using ternary templates previously prepared eye interrupt down the best eye candidate region from the candidate region (most ternary template Bok eye more to selecting the eye candidate region loam to, preventing erroneous detection such as vine 'glasses), step SP 1 in 2, again binarized represent optimal eye candidate region, A portion low brightness points are densest extracted as the eye, in step SP 1 3, it determines whether binocular extraction has been completed, if the binocular extracted is determined not to be finished the, in step SP 1 4, select the top right corner of the divided area which is classified into a cross, the process of step SP 4 again. Conversely, the 塲合 it is determined that the extraction of both eyes in step SP 1 3 is completed, a series of the processes is terminated. '

Further, when the eye candidate region is determined that there is only one in step SP 8, it performs as Step SP 1 2 processing.

In step SP 4 of Furochiya Ichiboku of FIG. 2, to detect the distance between both eyes on the basis of both eyes, which is extracted as described above, a face area using typical aspect ratio of the face obtained by enlarging on the basis of the magnification setting this face area in advance, to extract the final face area.

Then, further described with reference to specific examples.

FIG. 6 is a diagram illustrating a specific example of a process for obtaining a reduced image by applying the average luminance process the luminance value image.

As shown in FIG. 6 (A), the luminance value image with pixels of 4 X 4, p is the respective luminance values ​​of the second X 2 pixels under the left, q, r, in the case of s, taking the average of these luminance values ​​{specifically, performs a calculation of y = (p + q + r + s) / 4}, and the luminance value y of the corresponding pixel of the reduced image the resulting luminance values, by applying the same processing for other pixels, obtaining a reduced image.

By this process is performed, as shown in FIG. 6 (B), it is possible to reduce the luminance value image.

'It should be noted that FIG. 6, it is assumed that the case of reducing the 1/2, if necessary: ​​^ Z n (n is an integer of 3 or more) can be reduced to. Thus reducing the luminance value image, it is possible to reduce the number of pixels of interest in subsequent processing, it is possible to shorten the processing time required.

Of course, if it has sufficient processing speed, it can be omitted to reduce the luminance value image 'processing.

FIG. 7 is a diagram for explaining a specific example of a process for obtaining the first derivative image, and a specific example of the process to ignore the pixel that exceeds the differential maximum value.

The Rukoto to apply the R 0 BERTS operator for the luminance value image shown in FIG. 7 (A), it is possible to obtain a primary # of image shown in the seventh (B).

Then, the first derivative image shown in FIG. 7 (B) ', by a value of a pixel exceeding the threshold 徼分 maximum value acquired et one derivative face template as a threshold is set to 0, 7 as shown in FIG. (C), it is possible to obtain an image which eliminated the miscellaneous signal background. In this case, as described above, since the high edge representation disappears behind, it is possible to prevent a disadvantage that extracts a region outside 顏以 order to obtain a face outline area using a template.

FIG. 8 is a diagram illustrating a specific example of processing for creating a first derivative face template. First, as shown in FIG. 8 (A) ~ (D), Tokushi collected four primary differential face sample, merge these first derivative face sample {FIG. 8 in (A) ~ (D) in each pixel of the total value of the luminance values ​​of the same position, by a} this is divided by its number, to create a first derivative face template shown in FIG. 8 (E).

The pale rate (1. Value less than 0) the multiplication with the pixels of a certain in FIG. 8 using the light color index 'first derivative face template shown in (E) to the luminance values ​​of {each pixel lightening image by uniformly to lighter} a and creates a eighth lightening first derivative divided face template shown in FIG. (F). -

By adopting this manner lightening first derivative 顏 template created in order to absorb the sharp characteristic of the first derivative image due camera characteristics, from the detected image

'Immobilized whole pixel values ​​in the low position of the rate images can and this improving the comparison precision. Further, even if the face image size is larger than the size of the face template, it is possible to determine the position around the nose as the approximate position of the face.

Figure 9 is a diagram that describes the process of Sukiyan by the primary differential face template primary differential image.

When scanning by the primary differential face template T the first derivative image, as shown by an arrow in FIG. 9, shifting the pixels from the upper left end to the lower right, a simple scalar vector of the primary differential face template distance {a Euclidean distance (norm) is the square root of the sum of the squares of the differences for each element of the solid torr} determines calculates the position corresponding to the minimum value as a face outline position F. ..

Further described.

n X m V 1 1 the luminance value of each pixel of the pixel of the image, · · ·, V n 1, · ·

•, V im, · · ·, the vector of this image in the case of V nm (V 1 1, - ·

•, V n 'l, · · ·, V im, · · ·, and be represented as V nm) tau, the first derivative face template as well as solid torr of, to the right lower end from the upper left end of the first derivative image the vector of base and remove the image of the same size as the primary differential face template in order.

Then, the first derivative face template base-vector, to obtain a scalar distance by calculating the square root of the sum of the squares of the differences for each element of each image base-vector, the position of the image corresponding to the minimum scalar distance determining a 顏概 shown position F.

The first 0 Figure is a diagram illustrating a specific example of an extended, division processing of the face outline area.

First, is the region {first 0 in the drawing (Alpha) see} obtained by the face outline position acquisition fit into corresponding position of the luminance values, images, real number to enlarge the area of ​​this region from the center in four directions (first 0 FIG obtain a medium (beta) reference}. Then, for subsequent processing to divide the enlarged area to the cross {first 0 in the figure (C) see}.

The first 3 figure is a diagram illustrating a specific example of a process of scanning a minimum point group with minimum luminance template of the eye.

- If the minimum point group as indicated by the first 3 in the drawing (Alpha) is given, as indicated by the arrow in the first 3 in the drawing (beta), shifting the pixels from the upper left end to the right lower end, the minimum point perform a scan by using the aND conditions. Scan using the AND condition, instead of obtaining a perfect match positions, portions having high coincidence degree (result of the AND, the portion more points are Zantsu) detects as the position of the eye. Of course, perfect match location is also Rukoto sought. As a result, as shown by D in FIG. 13 (C), it is possible to detect the position of the eye. Note that the minimum luminance template E of the eye in FIG. 13 (B) shows.

Figure 14 is a diagram illustrating a specific example of a process of extracting the eye regions. -

If the partitioned regions is given as shown in FIG. 14 (A), to obtain a minimum point group, as shown in FIG. 14 (B), using the minimum brightness template of the eye, and the degree of matching threshold . provided value, eye candidate region D 1 from the upper position to the third candidate exceeding the threshold, and extracts as a .D 2 {in FIG. 14 (C) see}. Of the luminance value area, referring to extract an area corresponding to the eye candidate regions Dl, D2 {FIG. 14 in (El) (E2)), 3/4 and 1 of the average luminance values ​​of Meko capturing area / 2 for 3 values ​​expressed using {FIG. 14 in (F l) (F 2) reference}, using the first 14 ternary representation the template of the eye as shown in FIG. (G) matching by performing the process, as shown in FIG. 14 (H), selects one eye candidate region as the eye region.

FIG. 15 is a diagram illustrating a specific example of a process to more accurately extract the eye area. If the eye region is selected as shown in FIG. 15 (A), and the area again binarized representation {in FIG. 15 (B) see}, a portion low luminance point is most concentrated D the extract more accurate eye area by {in FIG. 15 (C) see} is extracted as the eye ball position.

FIG. 16 Ru FIG der illustrating a specific example of a process of extracting the face region based on the distance between both eyes.

Based on the distance d between the both eyes, to give a face region A 1 using typical aspect ratio of the face, by expanding using a predetermined magnification the center of the face 'region A 1 as a reference, the final obtaining a facial region a 2.

By employing the above 顏抽 out method and apparatus, it is Rukoto to Sosu the following effects. .

(1) Since the focusing on the luminance difference between the face of the part performing the template matching, Individuals vary, the amount of light, regardless of the influence of light source, it is possible to accurately extract a face region, the processing time required it can be shortened.

(2) by ignoring the detected image (luminance value image) significantly differentiated value at the stage of obtaining the force et first derivative image large point (point adjacent difference luminance value is greater), ignoring the ^ multi background can be, it is possible to more accurately extract a face region, it is possible to further shorten the processing time required.

(3) By lighter by rate in each pixel of the face template, to be able to absorb the difference of the primary differential caused by the sharpness of the image due to camera characteristics together, the size of the face included in the detected image it is exactly the extracted child face areas regardless. ,

(4) Noting that the eyeball absorbs light, by performing template matching on information minimum luminance shown to point at a peripheral region, individual differences, regardless quantity, the influence of light, the eye can be the position accurately extracted, hence, it is possible to accurately extract a face region, it is possible to shorten the processing time required.

(5) using the color information 'that no extraction of outline areas of the face using only the luminance value information, extraction of positions of both eyes, the extraction of the face region based on the distance between both eyes in stages, and test can be performed to narrow down the 查領 zone is omitted and thus the redundant process, it is possible to shorten the processing plant essential time.

(6) because of performing the extraction for each eye, be inclined face to some extent can you to deal with.

Incidentally, it is possible to store a computer program for executing each processing procedure of the face extraction method described above floppy disk, CD- R OM, in a storage medium such as MO, in this case, by executing the computer program stored in a storage medium a computer can achieve the effect similar to the above. Of course, more than the described face extracted wired computer program for running each processing procedure of the method, or it is also possible to transmit the combi Yuta through a wireless data channel, the computer program 'transmitted by Combi Yuta by executing, in addition to the operation and effect similar to the above. The invention of claim 1, individual differences, the influence of the light source, it is possible to extract a face region greatly eliminate highly accurately the effects such as color adjustment, such as digital cameras, in order to extract the face area ' so the specific effect that it is possible to significantly reduce the time required.

Specific that the invention of claim 2, by eliminating the influence of miscellaneous background image it is possible to further enhance the accuracy of extracting the face area, it is possible to greatly reduce the time required to extract a face region achieve the effect.

A third aspect of the present invention, in addition to the effect of claim 1 or claim 2, exhibits the characteristic effect of being able to compare the predetermined direction with respect to the comparison.

Of claim 4, it is possible to extract more accurately the face area, so the unique advantageous effect that it is possible to greatly reduce the time required to extract a face region.

The invention of claim 5, in addition to the effects of 請 ^ ¾ claim 4, exhibits a unique effect that the time required to extract a face region can be more greatly reduced.

The invention of claim 6, in addition to the effect of claim 4 or claim 5, exhibits a unique advantageous effect that it is possible to further increase the extraction accuracy of the face area.

The invention of claim 7, in addition claims 4 to any of the effects of claims 6, it is possible to enhance the 'extraction accuracy of eye position, a unique advantageous effect that it is possible to improve the extraction accuracy of the turn face area unlikely to. · Invention of claim 8, individual differences, the influence of the light source, it is possible to extract a face region to influence significantly eliminated to high precision, such as color adjustment, such as a digital camera., To extract a face region so the specific effect that it is possible to significantly reduce the time required to.

Specific that the invention of claim 9 is to eliminate the influence of miscellaneous background image it is possible to further enhance the accuracy of extracting the face area, it is possible to greatly reduce the time required to extract a face region achieve the effect.

The invention of claim 1 0, in addition to claim 8 or effect of claim 9, provides the 'unique advantage of being able to compare the fixed direction with respect to comparison.

The invention of claim 1 1 achieves it is possible to extract more accurately the face area, a unique advantageous effect that it is possible to greatly reduce the time required to extract a face region.

The invention of claim 1 2, in addition to the effect of claim 1 1, provides the unique advantageous effect that the time required to extract a face region can be more greatly reduced.

The invention of claim 1 3, in addition to claim 1 1 or claim 1 second effect achieves the specific effect of the Extraction accuracy of the face region can be further 髙Meru.

Specific that the invention of claim 1 4, in addition claims 1 1 to any of the effects of claims 1 to 3, it is possible to improve the extraction accuracy of the eye position., It is possible to increase the extraction accuracy of the turn face area achieve the effect.

The invention of claim 1 5, the same effects as any of claims 1 to 7. The invention of claim 1 6, the same effect as any of claims 1 to 7.

Claims

The scope of the claims
1. Create a converted image luminance value by performing luminance value conversion on the color image including a face region, "
5 luminance value converted image, to create a first derivative image using adjacent pixel information,
Face extraction method characterized by extracting the face area by performing a matching processing using the first-order derivative face template.
2. In preparing the first derivative image, face extraction method according to claim 1 to ignore the following points higher luminance value of the primary differential face template.
0 3. To extract the face region by using the first derivative face template lighter expressed dimmed rate. Ru face extraction method according to claim 1 or claim 2.
4. From the face region extracted by the method of any of claims 1 to 3, most. Extracts a point corresponding to a low luminance value as eye candidate positions,
Based on the extracted eye candidate position, extracted 5 out eye position by performing template matching,
Extracting a face subjected to the normalization of the face image based on the distance between the eyes and 'extracted
Face extraction wherein the.
5. The 顏領 area extracted by the method of any of claims 1 to 3 divided into right and left, the extraction to 0 Ru claim 4 points corresponding to the lowest luminance value in each segmented region as eye candidate position face extraction method described. -
6. The face region extracted by the method of any of claims 1 to 3 divided into right and left, to analyze the minimum luminance position in each partitioned area,
Perform eye candidate location expression using the luminance value average. Extracts eye candidate position by the template matching from the analysis information,
5 by performing template matching of the eye extracts the position of the eye
Face extraction method according to claim 4.
7. Face extraction method according to claim 6 claim 4 for extracting a low luminance set as eye position.
8. Luminance value converted image generating means for generating luminance values ​​converted image by performing luminance value conversion on the color image including a face region and (2),
From the luminance value converted image, the first derivative divided image creating means for creating a first derivative images using information between P 粦接 pixels (3),.
Face region extraction means for extracting a Matsuchingu process using the first derivative face template the means pursuant face region and (5).
Face extraction apparatus which comprises a.
9. The first derivative image generating means (3), the primary differential image bears stand to create a face extraction apparatus according to claim 8 in which ignores the following points higher luminance value of the primary differential face template .
1 0. The face region extraction means (5), the face extraction device according to claim 8 or claim 9 extracts a face region using first derivative face template was lighter expressed dimmed rate.
1 1. From the extracted face region by any of the apparatus of claim 1 0 of claims 8, with the eye climate capturing position extraction means for extracting a point corresponding to the minimum luminance value as eye candidate positions (6),
Based on the extracted eye candidate position, the eye position extraction means to extract the eye position by performing template matching (7),
Face extraction means for extracting a face by performing normalization of a face based on the extracted to the distance between both eyes (8), and (9).
Face extraction apparatus which comprises a.
1 2. The eye candidate position extraction means (6), that divides the face region extracted by any device of claim 1 0 of claims 8 to the left and right, corresponding to the lowest luminance value in each partitioned area the face extraction of claim 1 1 and extracts as eye candidate position
1 3. The eye candidate position extraction means (6), the analysis means for dividing the face region extracted by the method of any of claims 1 to 3 on the left and right, to analyze the minimum luminance position in each partitioned area When, the eye candidate location expression means for performing eye candidate location expression using the luminance value average extract eye candidate position 5 by the template matching from the analysis information is Dressings containing '
The eye position extracting means (7) is to perform the template matching of the eye extracts the position of the eye
Face extraction apparatus of claim 1 1.
10 1 4. The eye position extracting means (7) is face extraction device according to claim '1 3 low luminance set of claims 1 1 is. Those extracted as the eye position.
1 5. Storage medium Combi Yuta program is stored for executing any of the processing procedure of claims 1 to 7.
. 1 6 to the computer from claim 1 to perform any of the processing procedures according to claim 7:15 because of the program. -
20
■ 25
PCT/JP2001/001541 2001-02-28 2001-02-28 Face extracting method, device therefor, storage medium, and program WO2002069266A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2001/001541 WO2002069266A1 (en) 2001-02-28 2001-02-28 Face extracting method, device therefor, storage medium, and program

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2001/001541 WO2002069266A1 (en) 2001-02-28 2001-02-28 Face extracting method, device therefor, storage medium, and program

Publications (1)

Publication Number Publication Date
WO2002069266A1 true WO2002069266A1 (en) 2002-09-06

Family

ID=11737079

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2001/001541 WO2002069266A1 (en) 2001-02-28 2001-02-28 Face extracting method, device therefor, storage medium, and program

Country Status (1)

Country Link
WO (1) WO2002069266A1 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1703480A3 (en) * 2005-03-17 2007-02-14 Delphi Technologies, Inc. System and method to determine awareness
US8363957B2 (en) 2009-08-06 2013-01-29 Delphi Technologies, Inc. Image classification system and method thereof

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5982912A (en) * 1996-03-18 1999-11-09 Kabushiki Kaisha Toshiba Person identification apparatus and method using concentric templates and feature point candidates

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5982912A (en) * 1996-03-18 1999-11-09 Kabushiki Kaisha Toshiba Person identification apparatus and method using concentric templates and feature point candidates

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1703480A3 (en) * 2005-03-17 2007-02-14 Delphi Technologies, Inc. System and method to determine awareness
US7697766B2 (en) 2005-03-17 2010-04-13 Delphi Technologies, Inc. System and method to determine awareness
US8363957B2 (en) 2009-08-06 2013-01-29 Delphi Technologies, Inc. Image classification system and method thereof

Similar Documents

Publication Publication Date Title
US5933527A (en) Facial image processing method and apparatus
US7415165B2 (en) Red-eye detection device, red-eye detection method, and red-eye detection program
US6247813B1 (en) Iris identification system and method of identifying a person through iris recognition
US6151403A (en) Method for automatic detection of human eyes in digital images
KR0130962B1 (en) Apparatus for extracting facila characteristic points
US8254691B2 (en) Facial expression recognition apparatus and method, and image capturing apparatus
JP4161659B2 (en) Image recognition system and recognition method, and program
EP1296279B1 (en) Method and computer program product for locating facial features
US6895103B2 (en) Method for automatically locating eyes in an image
US6792134B2 (en) Multi-mode digital image processing method for detecting eyes
US20090169064A1 (en) Multi-scale Variable Domain Decomposition Method and System for Iris Identification
US6661907B2 (en) Face detection in digital images
US7239726B2 (en) System and method for effectively extracting facial feature information
US20040146187A1 (en) Iris extraction method
JP4217664B2 (en) Image processing method, image processing apparatus
Gllavata et al. A robust algorithm for text detection in images
CN1894703B (en) Pattern recognition method and device
US7092554B2 (en) Method for detecting eye and mouth positions in a digital image
Gómez-Moreno et al. Goal evaluation of segmentation algorithms for traffic sign recognition
US20110142355A1 (en) Apparatus, method, and program for discriminating subjects
US4903312A (en) Character recognition with variable subdivisions of a character region
JP4604439B2 (en) Image processing apparatus and method, and recording medium
US20120068917A1 (en) System and method for dynamic gesture recognition using geometric classification
JP4628882B2 (en) The method of learning classifiers, face detection method and apparatus, and program
US20060008145A1 (en) Image data processing apparatus and method, and image data processing program

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): AT CA CN IL IN JP KR NO RU SG US

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE TR

121 Ep: the epo has been informed by wipo that ep was designated in this application
122 Ep: pct application non-entry in european phase
NENP Non-entry into the national phase in:

Ref country code: JP