JP2008287399A

JP2008287399A - Personal authentication device and method

Info

Publication number: JP2008287399A
Application number: JP2007130347A
Authority: JP
Inventors: Hideaki Yamamoto; 英明山本
Original assignee: Toshiba Corp
Current assignee: Toshiba Corp
Priority date: 2007-05-16
Filing date: 2007-05-16
Publication date: 2008-11-27

Abstract

PROBLEM TO BE SOLVED: To provide a personal authentication device and a method for it, capable of preventing increase in processing quantity and time in personal authentication execution even if feature point extraction fails. SOLUTION: In the personal authentication device 1, a face area extraction part 11 extracts a face area of a current frame, a feature point candidate extraction part 12 extracts feature point candidates from the face area, a similarity determination part 13 determines whether there are feature candidates having similarity higher than a reference similarity or not, a predictive feature point area generation part 16 reads feature points, which are extracted by a feature point extraction part 14 and stored in a feature point storage part 14, in the previous frame to generate a predictive feature point area, a face area continuous output instruction part 18 instructs the face area extraction part 11 to continuously output the face area of the current frame to the following frame screen input when a feature point candidate determination part 17 determines that at least one feature candidate is included in the predictive feature point area. COPYRIGHT: (C)2009,JPO&INPIT

Description

本発明は、人物認証装置および方法に関する。 The present invention relates to a person authentication apparatus and method.

従来、個人認証に関し、さまざまな人物認証装置および方法が提案されている。その１つに、動画像を用いて顔認証を行う人物認証装置および方法がある。 Conventionally, various person authentication devices and methods have been proposed for personal authentication. One of them is a person authentication apparatus and method for performing face authentication using moving images.

この動画像を用いた個人認証の基本的な処理の流れは次のとおりである。
１．動画のフレームごとの入力画像から顔領域を抽出
２．抽出された領域から判定部位ごとの特徴点を抽出
３．抽出された特徴点から顔の正規化画像を生成
４．正規化画像を予め登録されている個人辞書の画像と比較して個人を認証
この中で、特徴点抽出処理では、顔領域抽出処理で抽出された顔領域から、分離度フィルタを用いて、瞳、鼻孔などの判定部位ごとの特徴点候補を抽出し、予め登録してある判定部位ごとの領域のテンプレートとの類似度を計算して、左右瞳点、左右鼻孔点などの特徴点を抽出する方法が提案されている（例えば、特許文献１参照。）。 The basic processing flow of personal authentication using this moving image is as follows.
1. 1. Extract face area from input image for each frame of video 2. Extract feature points for each determination part from the extracted area. 3. Generate a normalized image of the face from the extracted feature points Individuals are authenticated by comparing the normalized image with a pre-registered personal dictionary image. Among them, in the feature point extraction process, a pupil is extracted from the face area extracted by the face area extraction process using a separability filter. Extract feature points such as nostrils and right and left nostril points by extracting feature point candidates for each determination site such as nostrils and calculating the similarity with the template of the region for each determination site registered in advance. A method has been proposed (see, for example, Patent Document 1).

従来、このような特徴点抽出処理により現フレームの動画像から総ての特徴点の抽出に成功した場合、抽出された特徴点から次フレームの動画像における顔領域を予測し、その予測領域を次フレームの特徴点抽出に用いることにより、上述の顔領域抽出処理を省略することが行われている。 Conventionally, when all feature points have been successfully extracted from the moving image of the current frame by such feature point extraction processing, the face region in the moving image of the next frame is predicted from the extracted feature points, and the predicted region is By using the feature point extraction of the next frame, the face area extraction process described above is omitted.

また、上述の認証処理の方法として制約相互部分空間法が提案されている（例えば、特許文献２参照。）。この制約相互部分空間法では、予め規定された枚数の正規化画像が蓄積されたら入力部分空間を生成し、その入力部分空間を制約部分空間へ射影した入力射影部分空間と、予め制約部分空間へ射影して登録してある個人辞書部分空間との類似度を算出することにより、対象人物の同定を行う。 Further, a constrained mutual subspace method has been proposed as a method for the above-described authentication processing (see, for example, Patent Document 2). In this constrained mutual subspace method, an input subspace is generated when a predetermined number of normalized images are accumulated, and the input subspace is projected to the constrained subspace, The target person is identified by calculating the similarity with the personal dictionary subspace registered by projection.

この制約相互部分空間法を用いた認証処理では、連続するフレームで連続して特徴点の抽出に成功し、それにより生成された正規化画像が予め規定された枚数に達したときに認証処理を開始する。しかし、従来、特徴点の抽出に１フレームでも失敗すると、それまでに蓄積した正規化画像を総て破棄し、次フレームから新たに正規化画像を蓄積し直すことが行われている。 In the authentication process using the constrained mutual subspace method, feature points are successfully extracted continuously in consecutive frames, and the authentication process is performed when the number of normalized images generated thereby reaches a predetermined number. Start. However, conventionally, if even one frame fails in feature point extraction, all of the normalized images accumulated so far are discarded and the normalized image is newly accumulated from the next frame.

このように、従来の人物認証装置および方法では、特徴点抽出処理が成功するかどうかにより、個人認証実行時の処理量および処理時間が左右される。すなわち、特徴点抽出処理に成功すると、次フレームの顔領域抽出処理を省略することができ、また、正規化画像の蓄積を継続することができる。 As described above, in the conventional person authentication device and method, the processing amount and the processing time at the time of executing the personal authentication depend on whether or not the feature point extraction process is successful. That is, if the feature point extraction process is successful, the face area extraction process for the next frame can be omitted, and the accumulation of the normalized image can be continued.

ところが、従来の特徴点抽出処理では、実際に瞳、鼻孔が画像中に存在し、特徴点候補が抽出されるにも拘らず、画像処理の影響などで瞳点、鼻孔点という特徴点の抽出に失敗することがある。この場合、顔領域の予測が可能であるにも拘らず、次フレームでの顔領域抽出処理の実行が行われ、その分、処理量が増加するという問題があった。 However, in the conventional feature point extraction process, although pupils and nostrils are actually present in the image and feature point candidates are extracted, feature points such as pupil points and nostril points are extracted due to the influence of image processing. May fail. In this case, although the face area can be predicted, the face area extraction process is executed in the next frame, and the processing amount increases accordingly.

また、制約相互部分空間法における正規化画像の蓄積においては、同一人物の動画像のフレームが連続する場合に、たまたま、その中の１フレームでも特徴点抽出に失敗すると、その後の処理に利用できるにも拘らず、それまでに蓄積した画像を破棄してしまい、正規化画像の再蓄積から認証処理に至るまでの処理時間が増加するという問題があった。
特開平９−２５１５３４号公報（第４ページ、図３）特開２０００−３００６５号公報（第８−９ページ、図１） In addition, in the accumulation of normalized images in the constrained mutual subspace method, when moving image frames of the same person continue, if one of the frames happens to fail to extract feature points, it can be used for subsequent processing. Nevertheless, there is a problem in that the image accumulated so far is discarded, and the processing time from the re-accumulation of the normalized image to the authentication process increases.
JP-A-9-251534 (4th page, FIG. 3) JP 2000-30065 (page 8-9, FIG. 1)

そこで、本発明の目的は、特徴点抽出に失敗しても、個人認証実行時の処理量および処理時間の増加を抑制することのできる人物認証装置および方法を提供することにある。 Therefore, an object of the present invention is to provide a person authentication apparatus and method that can suppress an increase in processing amount and processing time when executing personal authentication even if feature point extraction fails.

本発明の一態様によれば、入力された動画像のフレーム画面から個人認証を行うための特徴点を抽出する人物認証装置であって、現フレームの画面から人物の顔領域を抽出する顔領域抽出手段と、前記顔領域から複数の判定部位ごとに特徴点候補を抽出する特徴点候補抽出手段と、前記特徴点候補と予め登録されているテンプレートとを比較し、基準の類似度を超える特徴点候補があるかどうかを判定する類似度判定手段と、前記類似度判定手段により前記複数の判定部位の総てで前記基準の類似度を超える特徴点候補があると判定されたときに、前記判定部位ごとに、最も類似度の高い特徴点候補を特徴点として抽出する特徴点抽出手段と、前記特徴点抽出手段により抽出された前記特徴点をフレームごとに記憶する特徴点記憶手段と、前記類似度判定手段により前記複数の判定部位のいずれかで前記基準の類似度を超える特徴点候補がないと判定されたときに、前記特徴点記憶手段から前フレームの特徴点を読み出し、その特徴点を中心とした予測特徴点領域を生成する予測特徴点領域生成手段と、前記予測特徴点領域が生成されたときに、前記予測特徴点領域に前記特徴点候補が少なくとも１つは含まれているかどうかを判定する特徴点候補判定手段と、前記特徴点候補判定手段により前記複数の判定部位総ての前記予測特徴点領域に前記特徴点候補が少なくとも１つは含まれていると判定されたときに、前記顔領域抽出手段へ、次フレーム画面入力に対する顔領域抽出を行わず、現フレームの前記顔領域を継続して出力するよう指示する顔領域継続出力指示手段と、を備えることを特徴とする人物認証装置が提供される。 According to one aspect of the present invention, there is provided a person authentication device that extracts feature points for performing personal authentication from a frame screen of an input moving image, the face region extracting a human face region from a screen of a current frame. A feature point candidate extraction unit that extracts a feature point candidate for each of a plurality of determination sites from the face region, a feature point candidate and a template registered in advance, and a feature that exceeds a reference similarity A similarity determination unit that determines whether or not there is a point candidate; and when the similarity determination unit determines that there is a feature point candidate that exceeds the reference similarity in all of the plurality of determination sites, Feature point extraction means for extracting the feature point candidate with the highest similarity as a feature point for each determination part; feature point storage means for storing the feature points extracted by the feature point extraction means for each frame; When the similarity determination unit determines that there is no feature point candidate exceeding the reference similarity in any of the plurality of determination parts, the feature point of the previous frame is read from the feature point storage unit, and the feature point A predicted feature point region generating means for generating a predicted feature point region centered at the center, and whether the predicted feature point region includes at least one feature point candidate when the predicted feature point region is generated When it is determined by the feature point candidate determination unit that determines whether or not the feature point candidate determination unit includes at least one of the feature point candidates in the predicted feature point region of all of the plurality of determination portions. And a face area continuous output instruction means for instructing the face area extraction means to continuously output the face area of the current frame without performing face area extraction for the next frame screen input. People authentication apparatus is provided, characterized in that.

また、本発明の別の一態様によれば、個人認証を行うための正規化画像を、入力された動画像のフレーム画面から生成して蓄積する人物認証装置であって、現フレームの画面から人物の顔領域を抽出する顔領域抽出手段と、前記顔領域から複数の判定部位ごとに特徴点候補を抽出する特徴点候補抽出手段と、前記特徴点候補と予め登録されているテンプレートとを比較し、基準の類似度を超える特徴点候補があるかどうかを判定する類似度判定手段と、前記類似度判定手段により前記複数の判定部位の総てで前記基準の類似度を超える特徴点候補があると判定されたときは、前記判定部位ごとに、最も類似度の高い特徴点候補を特徴点として抽出する特徴点抽出手段と、前記特徴点にもとづいて正規化画像を生成する正規化画像生成手段と、前記特徴点の抽出に成功したフレームの前記正規化画像を蓄積する正規化画像蓄積手段と、前記特徴点抽出手段により抽出された前記特徴点をフレームごとに記憶する特徴点記憶手段と、前記特徴点の抽出に失敗したときに、前記特徴点記憶手段から前フレームの特徴点を読み出し、その特徴点を中心とした予測特徴点領域を生成する予測特徴点領域生成手段と、前記予測特徴点領域が生成されたときに、前記予測特徴点領域に前記特徴点候補が少なくとも１つは含まれているかどうかを判定する特徴点候補判定手段と、前記特徴点候補判定手段により前記複数の判定部位総ての前記予測特徴点領域に前記特徴点候補が少なくとも１つは含まれていると判定されたときに、前記正規化画像蓄積手段へ、蓄積している前記正規化画像を保持するよう指示する正規化画像蓄積保持指示手段と、を備えることを特徴とする人物認証装置が提供される。 According to another aspect of the present invention, there is provided a person authentication device for generating and storing a normalized image for performing personal authentication from a frame screen of an input moving image, the screen being displayed from a screen of a current frame. Comparison is made between a face area extracting means for extracting a person's face area, a feature point candidate extracting means for extracting a feature point candidate for each of a plurality of determination sites from the face area, and a template registered in advance. A similarity determination unit that determines whether or not there is a feature point candidate that exceeds a reference similarity, and a feature point candidate that exceeds the reference similarity in all of the plurality of determination sites by the similarity determination unit. When it is determined that there is a feature point extraction unit that extracts a feature point candidate having the highest similarity as a feature point for each determination part, and a normalized image generation that generates a normalized image based on the feature point Means, Normalized image storage means for storing the normalized image of a frame for which feature points have been successfully extracted, feature point storage means for storing the feature points extracted by the feature point extraction means for each frame, and the features A prediction feature point region generation unit that reads out a feature point of the previous frame from the feature point storage unit and generates a prediction feature point region centered on the feature point when the point extraction fails; and the prediction feature point region Is generated, a feature point candidate determination unit that determines whether or not at least one of the feature point candidates is included in the predicted feature point region, and the plurality of determination site totals by the feature point candidate determination unit When it is determined that at least one of the feature point candidates is included in all the predicted feature point regions, the normalized image accumulated is held in the normalized image accumulation unit. People authentication apparatus is provided, characterized in that it comprises, a normalized image storing holding instruction means for instructing cormorants.

また、本発明のさらに別の一態様によれば、入力された動画像のフレーム画面を用いて個人認証を行う人物認証方法であって、前フレームからの指示に従って、入力された現フレームの画面に対して、前フレームで抽出された顔領域を継続使用するか、顔領域を新規抽出するかを決定するステップと、前記決定が新規抽出であったときに現フレームの画面から人物の顔領域を抽出するステップと、前記決定に応じて、継続使用または新規抽出の顔領域から複数の判定部位ごとに特徴点候補を抽出するステップと、前記特徴点候補と予め登録されているテンプレートとを比較し、基準の類似度を超える特徴点候補があるかどうかを判定するステップと、前記複数の判定部位の総てで前記基準の類似度を超える特徴点候補があると判定されたときに、前記判定部位ごとに、最も類似度の高い特徴点候補を特徴点として抽出するステップと、抽出された前記特徴点をフレームごとに記憶手段に記憶するステップと、前記複数の判定部位のいずれかで前記基準の類似度を超える特徴点候補がないと判定されたときに、前記記憶手段から前フレームの特徴点を読み出し、その特徴点を中心とした予測特徴点領域を生成するステップと、前記予測特徴点領域が生成されたときに、前記予測特徴点領域に前記特徴点候補が少なくとも１つは含まれているかどうかを判定するステップと、前記複数の判定部位総ての前記予測特徴点領域に前記特徴点候補が少なくとも１つは含まれていると判定されたときに、次フレームの画面入力に対して、現フレームの前記顔領域を継続して使用するよう指示するステップと、を備えることを特徴とする人物認証方法が提供される。 According to still another aspect of the present invention, there is provided a person authentication method for performing personal authentication using a frame screen of an input moving image, wherein the screen of the current frame input according to an instruction from the previous frame Determining whether to continue using the face area extracted in the previous frame or to newly extract the face area; and when the determination is a new extraction, the face area of the person from the screen of the current frame A step of extracting a feature point candidate for each of a plurality of determination sites from a face region that is continuously used or newly extracted according to the determination, and a comparison between the feature point candidate and a template registered in advance Determining whether there is a feature point candidate exceeding the reference similarity, and determining that there is a feature point candidate exceeding the reference similarity in all of the plurality of determination parts A step of extracting a feature point candidate having the highest similarity as a feature point for each determination portion, a step of storing the extracted feature point in a storage unit for each frame, and any one of the plurality of determination portions When it is determined that there is no feature point candidate exceeding the reference similarity, the feature point of the previous frame is read from the storage means, and a predicted feature point region centered on the feature point is generated; and Determining whether or not at least one feature point candidate is included in the predicted feature point region when the predicted feature point region is generated; and the predicted feature point region for all of the plurality of determination portions When it is determined that at least one of the feature point candidates is included, the screen input of the next frame is instructed to continue using the face area of the current frame. People authentication method characterized by comprising the step, is provided.

また、本発明のさらに別の一態様によれば、入力された動画像のフレーム画面から生成した正規化画像を複数枚蓄積し、その蓄積された複数枚の正規化画像を用いて個人認証を行う人物認証方法であって、現フレームの画面から人物の顔領域を抽出するステップと、前記顔領域から複数の判定部位ごとに特徴点候補を抽出するステップと、前記特徴点候補と予め登録されているテンプレートとを比較し、基準の類似度を超える特徴点候補があるかどうかを判定するステップと、前記複数の判定部位の総てで前記基準の類似度を超える特徴点候補があると判定されたときに、前記判定部位ごとに、最も類似度の高い特徴点候補を特徴点として抽出するステップと、前記特徴点にもとづいて正規化画像を生成するステップと、前記特徴点の抽出に成功したフレームの前記正規化画像を蓄積するステップと、抽出された前記特徴点をフレームごとに記憶手段に記憶するステップと、前記特徴点の抽出に失敗したときに、前記記憶手段から前フレームの特徴点を読み出し、その特徴点を中心とした予測特徴点領域を生成するステップと、前記予測特徴点領域が生成されたときに、前記予測特徴点領域に前記特徴点候補が少なくとも１つは含まれているかどうかを判定するステップと、を備え、前記複数の判定部位総ての前記予測特徴点領域に前記特徴点候補が１つも含まれていないと判定されたときは、蓄積していた前記正規化画像を破棄し、前記複数の判定部位総ての前記予測特徴点領域に前記特徴点候補が少なくとも１つは含まれていると判定されたときは、蓄積している前記正規化画像を保持することを特徴とする人物認証方法が提供される。 According to still another aspect of the present invention, a plurality of normalized images generated from a frame screen of an input moving image are accumulated, and personal authentication is performed using the accumulated normalized images. A person authentication method to be performed, the step of extracting a face area of a person from the screen of the current frame, the step of extracting feature point candidates for each of a plurality of determination parts from the face area, and the feature point candidates registered in advance And determining whether there is a feature point candidate that exceeds the reference similarity, and determining that there is a feature point candidate that exceeds the reference similarity in all of the plurality of determination portions. And extracting the feature point candidate with the highest similarity as a feature point for each determination part, generating a normalized image based on the feature point, and extracting the feature point. Storing the normalized image of successful frames; storing the extracted feature points in a storage unit for each frame; and when the extraction of the feature points fails, from the storage unit A step of reading out feature points and generating a predicted feature point region centered on the feature points; and when the predicted feature point region is generated, at least one feature point candidate is included in the predicted feature point region A step of determining whether the feature point candidates are not included in the predicted feature point region of all of the plurality of determination parts, When the normalized image is discarded and it is determined that at least one of the feature point candidates is included in the predicted feature point region of all of the plurality of determination portions, the accumulated normalized image People authentication method characterized by retaining is provided.

本発明によれば、特徴点抽出に失敗しても、個人認証実行時の処理量および処理時間の増加を抑制することができる。 According to the present invention, even if feature point extraction fails, an increase in the processing amount and processing time when executing personal authentication can be suppressed.

以下、本発明の実施例を図面を参照して説明する。 Embodiments of the present invention will be described below with reference to the drawings.

図１は、本発明の実施例１に係る人物認証装置の構成の例を示すブロック図である。 FIG. 1 is a block diagram illustrating an example of the configuration of the person authentication device according to the first embodiment of the present invention.

本実施例の人物認証装置１は、入力された動画像のフレーム画面から個人認証を行うために、左右瞳点や左右鼻孔点など、個人の特定に用いる判定部位を顔面の特徴点として抽出する装置である。 In order to perform personal authentication from the input moving image frame screen, the person authentication device 1 according to the present embodiment extracts determination parts used for individual identification, such as left and right pupil points and right and left nostril points, as facial feature points. Device.

人物認証装置１は、入力された現フレームの画面から人物の顔領域を抽出する顔領域抽出部１１と、顔領域から左右瞳点や左右鼻孔点などの判定部位ごとに特徴点候補を抽出する特徴点候補抽出部１２と、抽出された特徴点候補と予め登録されているテンプレートとを比較し、基準の類似度を超える特徴点候補があるかどうかを判定する類似度判定部１３と、類似度判定部１３により左右瞳点や左右鼻孔点などの判定部位の総てで基準の類似度を超える特徴点候補があると判定されたときに、左右瞳点や左右鼻孔点などの判定部位ごとに、最も類似度の高い特徴点候補を特徴点として抽出する特徴点抽出部１４と、その抽出された特徴点をフレームごとに記憶する特徴点記憶部１５と、類似度判定部１３により左右瞳点や左右鼻孔点などの判定部位のいずれかで基準の類似度を超える特徴点候補がないと判定されたときに、特徴点記憶部１５から前フレームの特徴点を読み出し、その特徴点を中心とした予測特徴点領域を生成する予測特徴点領域生成部１６と、予測特徴点領域が生成されたときに、この予測特徴点領域に上述の特徴点候補が少なくとも１つは含まれているかどうかを判定する特徴点候補判定部１７と、特徴点候補判定部１７により左右瞳点や左右鼻孔点などの判定部位総ての予測特徴点領域に上述の特徴点候補が少なくとも１つは含まれていると判定されたときに、顔領域抽出部１１へ、次フレーム画面入力に対する顔領域抽出を行わず、現フレームの前記顔領域を継続して出力するよう指示する顔領域継続出力指示部１８と、を備える。 The person authentication device 1 extracts a face area extraction unit 11 that extracts a person's face area from the input current frame screen, and extracts feature point candidates for each determination part such as left and right pupil points and right and left nostril points from the face area. The feature point candidate extraction unit 12 compares the extracted feature point candidate with a pre-registered template, and determines whether there is a feature point candidate exceeding the reference similarity, and a similarity determination unit 13 When it is determined by the degree determination unit 13 that there are feature point candidates that exceed the reference similarity in all of the determination parts such as the left and right pupil points and the right and left nostril points, The feature point extraction unit 14 that extracts the feature point candidate with the highest similarity as the feature point, the feature point storage unit 15 that stores the extracted feature point for each frame, and the similarity determination unit 13 Point and right and left nostrils When it is determined that there is no feature point candidate exceeding the reference similarity in any of the parts, the feature point of the previous frame is read from the feature point storage unit 15 and a predicted feature point region centered on the feature point is generated And a feature point candidate determination unit that determines whether or not at least one feature point candidate is included in the predicted feature point region when the predicted feature point region is generated. 17 and when the feature point candidate determination unit 17 determines that at least one of the above-described feature point candidates is included in the predicted feature point regions of all the determination sites such as the left and right pupil points and the left and right nostril points, And a face area continuous output instructing unit 18 that instructs the face area extracting unit 11 to continuously output the face area of the current frame without extracting the face area for the next frame screen input.

このように、人物認証装置１は、現フレームの画面から抽出された特徴点候補に基準の類似度を超える特徴点候補がないと判定したとき、すなわち、現フレームの画面からの特徴点の抽出に失敗したときに、予測特徴点領域を生成し、この予測特徴点領域に特徴点候補が少なくとも１つ含まれていれば、次フレーム画面入力の際に、新たな顔領域抽出を行わず、現フレームの顔領域を継続して使用するようにする。 As described above, when it is determined that the feature point candidate extracted from the screen of the current frame has no feature point candidate exceeding the reference similarity, the person authentication device 1 extracts feature points from the screen of the current frame. If the prediction feature point region is generated, and if at least one feature point candidate is included in the predicted feature point region, a new face region is not extracted when the next frame screen is input, Continue to use the face area of the current frame.

すなわち、本実施例の人物認証装置１では、入力されたフレームの画像データに対して顔領域を抽出するか否かの指示が、前のフレームから次のフレームへと、順次、出されて行く。 That is, in the person authentication device 1 of the present embodiment, an instruction as to whether or not to extract a face area from the input frame image data is sequentially issued from the previous frame to the next frame. .

そこで、人物認証装置１における処理の冒頭では、顔領域抽出に関する前フレームの指示にもとづいて、新たな顔領域抽出を行うか、前フレームの顔領域を継続して使用するかの決定を行う。 Therefore, at the beginning of the processing in the person authentication device 1, it is determined whether to extract a new face area or to continue using the face area of the previous frame based on an instruction of the previous frame related to face area extraction.

次に、このような人物認証装置１における特徴点抽出の処理手順の例について図２〜図５を用いて説明する。 Next, an example of the processing procedure of feature point extraction in the person authentication apparatus 1 will be described with reference to FIGS.

図２は、人物認証装置１における特徴点抽出処理の手順の例を示すフロー図である。 FIG. 2 is a flowchart showing an example of the procedure of the feature point extraction process in the person authentication device 1.

現フレームの画像データが入力されると、人物認証装置１は、まず、顔領域継続出力指示部１８から出力される顔領域抽出に関する前フレームの指示を確認し（ステップＳ０１）、その指示が前フレームの顔領域の継続使用であるときは、前フレームの顔領域の継続使用することとし（ステップＳ０２）、その指示が顔領域の新規抽出であるときは、顔領域抽出部１１が、現フレームの画面から人物の顔領域を新たに抽出する（ステップＳ０３）。 When the image data of the current frame is input, the person authentication apparatus 1 first confirms the instruction of the previous frame related to face area extraction output from the face area continuous output instruction unit 18 (step S01), and the instruction is When the face area of the frame is to be used continuously, the face area of the previous frame is continuously used (step S02). When the instruction is a new extraction of the face area, the face area extracting unit 11 performs the current frame. The face area of the person is newly extracted from the screen (step S03).

次に、この継続使用または新規抽出の顔領域から、特徴点候補抽出部１２が、左右瞳点や左右鼻孔点などの判定部位ごとに特徴点候補を抽出し（ステップＳ０４）、類似度判定部１３が、予め登録されているテンプレートとの類似度の比較を行い（ステップＳ０５）、基準の類似度を超える特徴点候補があるかどうかを判定する（ステップＳ０６）。 Next, the feature point candidate extraction unit 12 extracts feature point candidates for each determination site such as left and right pupil points and right and left nostril points from the continuously used or newly extracted face region (step S04), and the similarity determination unit 13 compares the similarity with a template registered in advance (step S05), and determines whether there is a feature point candidate exceeding the reference similarity (step S06).

この判定で、基準の類似度を超える特徴点候補があると判定されたときは、特徴点抽出部１４が、判定部位ごとに、最も類似度の高い特徴点候補を特徴点として抽出する（ステップＳ０７）。 When it is determined in this determination that there is a feature point candidate exceeding the reference similarity, the feature point extraction unit 14 extracts the feature point candidate with the highest similarity as a feature point for each determination part (step) S07).

抽出された特徴点は、記憶手段に記憶され（ステップＳ０８）、認証処理へ引き渡される。 The extracted feature points are stored in the storage means (step S08) and delivered to the authentication process.

図３に、抽出された特徴点の例を模式的に示す。ここでは、個人の特定に用いる判定部位を左右瞳点および左右鼻孔点としたときに抽出された特徴点の例を示す。 FIG. 3 schematically shows an example of extracted feature points. Here, an example of feature points extracted when the determination sites used for specifying an individual are the left and right pupil points and the left and right nostril points is shown.

図３では、入力されたフレーム画面の画像データの顔領域１００から、左右瞳点を示す特徴点１０１、１０２、および、左右鼻孔点を示す特徴点１０３、１０４が抽出された例を示す。 FIG. 3 shows an example in which feature points 101 and 102 indicating left and right pupil points and feature points 103 and 104 indicating left and right nostril points are extracted from the face area 100 of the image data of the input frame screen.

一方、図２に示すステップＳ０６における判定で、基準の類似度を超える特徴点候補がないと判定されたときは、予測特徴点領域生成部１６が、フレームごとの特徴点が記憶されている記憶手段から前フレームの特徴点を読み出し（ステップＳ０９）、その特徴点を中心とした予測特徴点領域を生成する（ステップＳ１０）。 On the other hand, when it is determined in step S06 shown in FIG. 2 that there is no feature point candidate exceeding the reference similarity, the predicted feature point region generation unit 16 stores the feature points for each frame. The feature point of the previous frame is read from the means (step S09), and a predicted feature point region centered on the feature point is generated (step S10).

図４に、生成された予測特徴点領域の例を示す。ここでは、予測特徴点領域を矩形とした例を示すが、予測特徴点領域を、円形、楕円形としてもよい。 FIG. 4 shows an example of the generated predicted feature point region. Here, an example in which the predicted feature point region is rectangular is shown, but the predicted feature point region may be circular or elliptical.

記憶手段から読み出した前フレームの特徴点１０１、１０２、１０３、１０４に対して、それぞれの特徴点を中心として、予測特徴点領域１１１、１１２、１１３、１１４が生成される。この予測特徴点領域の大きさは、実験などにより最適値を求めて設定するなど、任意に設定することができる。 For the feature points 101, 102, 103, and 104 of the previous frame read from the storage unit, predicted feature point regions 111, 112, 113, and 114 are generated around the feature points. The size of the predicted feature point region can be arbitrarily set, for example, by obtaining an optimum value through experiments or the like.

図２のフローに戻って、次に、特徴点候補判定部１７が、それぞれの予測特徴点領域に、ステップＳ０４で抽出した特徴点候補が少なくとも１つは含まれているかどうかを判定する（ステップＳ１１）。 Returning to the flow of FIG. 2, next, the feature point candidate determination unit 17 determines whether or not at least one feature point candidate extracted in step S 04 is included in each predicted feature point region (step S 04). S11).

このとき、図５（ａ）に示すように、予測特徴点領域１１１、１１２、１１３、１１４に対して、×で示す特徴点候補が抽出されているときは、特徴点候補判定部１７が、総ての判定部位の予測特徴点領域に特徴点候補が少なくとも１つは含まれていると判定する。この判定を受けて、顔領域継続出力指示部１８は、顔領域抽出部１１へ、次フレーム画面入力に対して、現フレームの顔領域を継続して出力するよう指示する（ステップＳ１２）。 At this time, as shown in FIG. 5A, when feature point candidates indicated by x are extracted for the predicted feature point regions 111, 112, 113, and 114, the feature point candidate determination unit 17 It is determined that at least one feature point candidate is included in the predicted feature point regions of all determination parts. In response to this determination, the face area continuous output instructing unit 18 instructs the face area extracting unit 11 to continuously output the face area of the current frame in response to the next frame screen input (step S12).

一方、予測特徴点領域１１１、１１２、１１３、１１４に対して、図５（ｂ）に×で示すような特徴点候補が抽出されているときは、特徴点候補判定部１７が、総ての判定部位の予測特徴点領域のいずれにも特徴点候補が１つも含まれていないと判定する。この判定を受けると、顔領域継続出力指示部１８は、顔領域抽出部１１へ、次フレーム画面入力に対して、顔領域を新規抽出するよう指示する（ステップＳ１３）。 On the other hand, when the feature point candidates as indicated by x in FIG. 5B are extracted for the predicted feature point regions 111, 112, 113, 114, the feature point candidate determination unit 17 It is determined that none of the feature point candidates is included in any of the predicted feature point regions of the determination part. Upon receiving this determination, the face area continuation output instruction unit 18 instructs the face area extraction unit 11 to newly extract a face area in response to the next frame screen input (step S13).

上述のステップＳ１２またはステップＳ１３の処理が行われると、現フレームに対する処理が終了し、次フレームの画面データ入力へと処理は進む。次フレームの画面データが入力されると、図２のフローの先頭から処理が行われる。すなわち、図２に示すフローの処理が、入力されるフレームの画面データごとに実行される。 When the process of step S12 or step S13 described above is performed, the process for the current frame ends, and the process proceeds to screen data input for the next frame. When the screen data of the next frame is input, processing is performed from the top of the flow of FIG. That is, the process of the flow shown in FIG. 2 is executed for each screen data of the input frame.

このような本実施例によれば、現フレームの画像データから特徴点の抽出に失敗しても、前フレームの特徴点を中心に形成した予測特徴点領域に現フレームの特徴点候補が１つでも含まれていれば、次フレームの画像データ入力に対する顔領域抽出を行わなくてもよいので、個人認証実行時の処理量の増加を抑制することができる。 According to this embodiment, even if feature point extraction fails from image data of the current frame, one feature point candidate of the current frame is present in the predicted feature point region formed around the feature point of the previous frame. However, if it is included, it is not necessary to extract the face area for the image data input of the next frame, so that it is possible to suppress an increase in the processing amount when executing personal authentication.

図６は、本発明の実施例２に係る人物認証装置の構成の例を示すブロック図である。 FIG. 6 is a block diagram illustrating an example of the configuration of the person authentication device according to the second embodiment of the present invention.

本実施例の人物認証装置２は、例えば制約相互部分空間法による個人認証を行うために必要な正規化画像を、入力された動画像のフレーム画面から生成して蓄積する人物認証装置である。なお、図６において、図１に示した実施例１のブロックと同様の機能を有するブロックには図１と同一の符号を付し、ここではその詳細な説明を省略する。 The person authentication device 2 according to the present embodiment is a person authentication device that generates and stores a normalized image necessary for performing personal authentication by a constrained mutual subspace method, for example, from a frame screen of an input moving image. In FIG. 6, blocks having the same functions as those in the first embodiment shown in FIG. 1 are denoted by the same reference numerals as those in FIG. 1, and detailed description thereof is omitted here.

本実施例の人物認証装置２は、実施例１の人物認証装置１と同様に、顔領域抽出部１１と、特徴点候補抽出部１２と、類似度判定部１３と、特徴点抽出部１４と、特徴点記憶部１５と、予測特徴点領域生成部１６と、特徴点候補判定部１７と、を備える。 Similar to the person authentication device 1 of the first embodiment, the person authentication device 2 of the present embodiment includes a face area extraction unit 11, a feature point candidate extraction unit 12, a similarity determination unit 13, and a feature point extraction unit 14. A feature point storage unit 15, a predicted feature point region generation unit 16, and a feature point candidate determination unit 17.

さらに、人物認証装置２は、特徴点抽出部１４により抽出された特徴点にもとづいて正規化画像を生成する正規化画像生成部２１と、上述の特徴点の抽出に成功したフレームの正規化画像を蓄積する正規化画像蓄積部２２と、特徴点候補判定部１７により左右瞳点や左右鼻孔点などの判定部位総ての予測特徴点領域に上述の特徴点候補が少なくとも１つは含まれていると判定されたときに、正規化画像蓄積部２２へ、蓄積している正規化画像を保持するよう指示する正規化画像蓄積保持指示部２３と、を備える。 Furthermore, the person authentication device 2 includes a normalized image generation unit 21 that generates a normalized image based on the feature points extracted by the feature point extraction unit 14, and a normalized image of the frame that has been successfully extracted. At least one of the above-mentioned feature point candidates is included in the predicted feature point regions of all the determination sites such as the left and right pupil points and the right and left nostril points by the normalized image storage unit 22 that stores And a normalized image accumulation / holding instruction unit 23 for instructing the normalized image accumulation unit 22 to retain the accumulated normalized image.

このように、人物認証装置２は、現フレームの画面から抽出された特徴点候補に基準の類似度を超える特徴点候補がないと判定したとき、すなわち、現フレームの画面からの特徴点の抽出に失敗したときに、予測特徴点領域を生成し、この予測特徴点領域に特徴点候補が少なくとも１つ含まれていれば、正規化画像蓄積部２２に蓄積している正規化画像を破棄せず、そのまま保持するようにする。 As described above, when it is determined that the feature point candidate extracted from the screen of the current frame has no feature point candidate exceeding the reference similarity, the person authentication device 2 extracts feature points from the screen of the current frame. If the predicted feature point region is generated, and if at least one feature point candidate is included in the predicted feature point region, the normalized image stored in the normalized image storage unit 22 is discarded. Instead, keep it as it is.

次に、このような人物認証装置２における正規化画像蓄積の処理手順の例について図７を用いて説明する。 Next, an example of a normalized image accumulation processing procedure in the person authentication device 2 will be described with reference to FIG.

図７は、人物認証装置２における正規化画像蓄積処理の手順の例を示すフロー図である。 FIG. 7 is a flowchart showing an example of a procedure of normalized image accumulation processing in the person authentication device 2.

現フレームの画像データが入力されると、顔領域抽出部１１が人物の顔領域を抽出し（ステップＳ２１）、特徴点候補抽出部１２が左右瞳点や左右鼻孔点などの判定部位ごとに特徴点候補を抽出し（ステップＳ２２）、類似度判定部１３が、予め登録されているテンプレートとの類似度の比較を行い（ステップＳ２３）、基準の類似度を超える特徴点候補があるかどうかを判定する（ステップＳ２４）。 When the image data of the current frame is input, the face area extracting unit 11 extracts the face area of the person (step S21), and the feature point candidate extracting unit 12 is characterized for each determination site such as the left and right pupil points and the left and right nostril points. A point candidate is extracted (step S22), and the similarity determination unit 13 compares the similarity with a template registered in advance (step S23), and determines whether there is a feature point candidate exceeding the reference similarity. Determination is made (step S24).

この判定で、基準の類似度を超える特徴点候補があると判定されたときは、特徴点抽出部１４が、判定部位ごとに、最も類似度の高い特徴点候補を特徴点として抽出する（ステップＳ２５）。抽出された特徴点は、記憶手段に記憶される（ステップＳ２６）。 When it is determined in this determination that there is a feature point candidate exceeding the reference similarity, the feature point extraction unit 14 extracts the feature point candidate with the highest similarity as a feature point for each determination part (step) S25). The extracted feature points are stored in the storage means (step S26).

次に、正規化画像生成部２１が、抽出された特徴点にもとづいて正規化画像を生成し（ステップＳ２７）、正規化画像蓄積部２２が、生成された正規化画像を蓄積する（ステップＳ２８）。 Next, the normalized image generation unit 21 generates a normalized image based on the extracted feature points (step S27), and the normalized image storage unit 22 stores the generated normalized image (step S28). ).

さらに、正規化画像蓄積部２２は、この蓄積により、正規化画像の蓄積された枚数が、規定の枚数に達したかどうかを判定し（ステップＳ２９）、正規化画像が規定枚数あるときは、蓄積している正規化画像を認証処理へ引き渡す。 Furthermore, the normalized image accumulating unit 22 determines whether or not the accumulated number of normalized images has reached a prescribed number by this accumulation (step S29). When the normalized image has a prescribed number, The accumulated normalized image is transferred to the authentication process.

一方、正規化画像が規定枚数ないときは、蓄積している正規化画像を正規化画像蓄積保持指示部２３の指示に従って処理する。 On the other hand, when there is no specified number of normalized images, the accumulated normalized images are processed in accordance with an instruction from the normalized image accumulation / holding instruction unit 23.

この正規化画像蓄積保持指示部２３の指示は、次に述べるような処理を行って出される。 The instruction of the normalized image accumulation / holding instruction unit 23 is issued by performing the following processing.

すなわち、ステップＳ２４において、基準の類似度を超える特徴点候補がないと判定されたときに、予測特徴点領域生成部１６が、フレームごとの特徴点が記憶されている記憶手段から前フレームの特徴点を読み出し（ステップＳ３０）、その特徴点を中心とした予測特徴点領域を生成し（ステップＳ３１）、特徴点候補判定部１７が、それぞれの予測特徴点領域に、ステップＳ２２で抽出した特徴点候補が少なくとも１つは含まれているかどうかを判定する（ステップＳ３２）。 That is, when it is determined in step S24 that there are no feature point candidates exceeding the reference similarity, the predicted feature point region generation unit 16 performs the feature of the previous frame from the storage unit storing the feature points for each frame. A point is read (step S30), a predicted feature point region centered on the feature point is generated (step S31), and the feature point candidate determination unit 17 extracts the feature point extracted in step S22 to each predicted feature point region. It is determined whether at least one candidate is included (step S32).

このとき、特徴点候補判定部１７が、総ての判定部位の予測特徴点領域に特徴点候補が少なくとも１つは含まれていると判定すると、正規化画像蓄積保持指示部２３は、蓄積している正規化画像をそのまま保持することを正規化画像蓄積部２２へ指示する。 At this time, if the feature point candidate determination unit 17 determines that at least one feature point candidate is included in the predicted feature point regions of all the determination parts, the normalized image storage and retention instruction unit 23 stores the points. The normalized image storage unit 22 is instructed to retain the normalized image as it is.

一方、特徴点候補判定部１７が、総ての判定部位の予測特徴点領域に特徴点候補が１つも含まれていないと判定すると、正規化画像蓄積保持指示部２３は、蓄積している正規化画像を破棄することを正規化画像蓄積部２２へ指示する。 On the other hand, if the feature point candidate determination unit 17 determines that no feature point candidates are included in the predicted feature point regions of all the determination parts, the normalized image storage and retention instruction unit 23 stores the normal The normalized image storage unit 22 is instructed to discard the normalized image.

正規化画像蓄積部２２は、この正規化画像蓄積保持指示部２３からの指示を受けて、蓄積している正規化画像をそのまま保持する（ステップＳ３３）か、破棄する（ステップＳ３４）か、のいずれかの処理を行う。 In response to the instruction from the normalized image accumulation / holding instruction unit 23, the normalized image accumulation unit 22 retains the accumulated normalized image as it is (step S33) or discards it (step S34). Do one of the processes.

これらの処理により現フレームの画像データへの処理は終了し、次フレームの画像データ入力へと、処理は進む。次フレームの画面データが入力されると、図７のフローの先頭から処理が行われる。すなわち、図７に示すフローの処理が、入力されるフレームの画面データごとに実行される。 With these processes, the process for the image data of the current frame is completed, and the process proceeds to the image data input of the next frame. When screen data of the next frame is input, processing is performed from the top of the flow of FIG. That is, the processing of the flow shown in FIG. 7 is executed for each screen data of the input frame.

このような本実施例によれば、現フレームの画像データから特徴点の抽出に失敗しても、前フレームの特徴点を中心に形成した予測特徴点領域に現フレームの特徴点候補が１つでも含まれていれば、それまでに蓄積していた画像をそのまま保持するので、従来の特徴点の抽出に失敗したときに直ちに蓄積していた画像を破棄して正規化画像の蓄積を最初からやり直していたものに比べて、個人認証実行時の処理時間の増加を抑制することができる。 According to this embodiment, even if feature point extraction fails from image data of the current frame, one feature point candidate of the current frame is present in the predicted feature point region formed around the feature point of the previous frame. However, if it is included, the image that has been accumulated up to that point is retained as it is, so when the conventional feature point extraction fails, the image that was accumulated immediately is discarded and the accumulation of the normalized image is started from the beginning. Compared to what has been redone, an increase in processing time when executing personal authentication can be suppressed.

図８は、本発明の実施例３に係る人物認証装置の構成の例を示すブロック図である。 FIG. 8 is a block diagram illustrating an example of the configuration of the person authentication device according to the third embodiment of the present invention.

本実施例の人物認証装置３は、実施例２の人物認証装置２に、実施例１の顔領域継続出力指示部１８を追加したものである。 The person authentication device 3 of this embodiment is obtained by adding the face area continuous output instruction unit 18 of the first embodiment to the person authentication device 2 of the second embodiment.

すなわち、本実施例の人物認証装置３は、実施例２の人物認証装置２の機能と、実施例１の人物認証装置１の機能と、を併せ持つ。これにより、本実施例では、特徴点候補判定部１７の判定にもとづき、正規化画像蓄積保持指示部２３が正規化画像蓄積部２２へ指示を出すとともに、顔領域継続出力指示部１８が顔領域抽出部１１への指示を出す。 That is, the person authentication device 3 of the present embodiment has both the function of the person authentication device 2 of the second embodiment and the function of the person authentication device 1 of the first embodiment. Thus, in this embodiment, the normalized image accumulation / holding instruction unit 23 issues an instruction to the normalized image accumulation unit 22 based on the determination of the feature point candidate determination unit 17, and the face area continuation output instruction unit 18 An instruction to the extraction unit 11 is issued.

このような本実施例によれば、個人認証実行時の処理時間の増加を抑制することができるとともに、処理量の増加も抑制することができる According to such a present Example, while being able to suppress the increase in the processing time at the time of personal authentication execution, the increase in a processing amount can also be suppressed.

本発明の実施例１に係る人物認証装置の構成の例を示すブロック図。1 is a block diagram showing an example of the configuration of a person authentication device according to Embodiment 1 of the present invention. 実施例１の人物認証装置における処理の手順の例を示すフロー図。FIG. 3 is a flowchart illustrating an example of a processing procedure in the person authentication device according to the first embodiment. 実施例１の人物認証装置における特徴点抽出の例を示す図。FIG. 3 is a diagram illustrating an example of feature point extraction in the person authentication device according to the first embodiment. 実施例１の人物認証装置における予想特徴点領域生成の例を示す図。FIG. 6 is a diagram illustrating an example of expected feature point region generation in the person authentication device according to the first embodiment. 実施例１の人物認証装置における特徴点候補判定の様子を示す図。The figure which shows the mode of the feature point candidate determination in the person authentication apparatus of Example 1. FIG. 本発明の実施例２に係る人物認証装置の構成の例を示すブロック図。The block diagram which shows the example of a structure of the person authentication apparatus which concerns on Example 2 of this invention. 実施例２の人物認証装置における処理の手順の例を示すフロー図。FIG. 9 is a flowchart illustrating an example of a processing procedure in the person authentication device according to the second embodiment. 本発明の実施例３に係る人物認証装置の構成の例を示すブロック図。The block diagram which shows the example of a structure of the person authentication apparatus which concerns on Example 3 of this invention.

Explanation of symbols

１、２、３人物認証装置
１１顔領域抽出部
１２特徴点候補抽出部
１３類似度判定部
１４特徴点抽出部
１５特徴点記憶部
１６予想特徴点領域生成部
１７特徴点候補判定部
１８顔領域継続出力指示部
２１正規化画像生成部
２２正規化画像蓄積部
２３正規化画像蓄積指示部 1, 2, 3 Person authentication device 11 Face region extraction unit 12 Feature point candidate extraction unit 13 Similarity determination unit 14 Feature point extraction unit 15 Feature point storage unit 16 Expected feature point region generation unit 17 Feature point candidate determination unit 18 Face region Continuous output instruction unit 21 Normalized image generation unit 22 Normalized image storage unit 23 Normalized image storage instruction unit

Claims

A person authentication device for extracting feature points for performing personal authentication from a frame screen of an input moving image,
Face area extraction means for extracting a person's face area from the screen of the current frame;
Feature point candidate extraction means for extracting feature point candidates for each of a plurality of determination sites from the face area;
A similarity determination unit that compares the feature point candidate with a template registered in advance and determines whether there is a feature point candidate that exceeds a reference similarity;
When it is determined by the similarity determination means that there is a feature point candidate that exceeds the reference similarity in all of the plurality of determination parts, the feature point candidate with the highest similarity is characterized for each determination part. Feature point extraction means for extracting as points;
Feature point storage means for storing the feature points extracted by the feature point extraction means for each frame;
When the similarity determination unit determines that there is no feature point candidate exceeding the reference similarity in any of the plurality of determination parts, the feature point of the previous frame is read from the feature point storage unit, and the feature Predicted feature point region generating means for generating a predicted feature point region centered on a point;
Feature point candidate determination means for determining whether or not at least one of the feature point candidates is included in the predicted feature point region when the predicted feature point region is generated;
When it is determined by the feature point candidate determination means that at least one of the feature point candidates is included in the predicted feature point areas of all of the plurality of determination portions, the next frame is sent to the face area extraction means. Face area continuous output instruction means for instructing to continuously output the face area of the current frame without performing face area extraction for screen input;
A person authentication device comprising:

A person authentication device that generates and stores a normalized image for performing personal authentication from a frame screen of an input moving image,
Face area extraction means for extracting a person's face area from the screen of the current frame;
Feature point candidate extraction means for extracting feature point candidates for each of a plurality of determination sites from the face area;
A similarity determination unit that compares the feature point candidate with a template registered in advance and determines whether there is a feature point candidate that exceeds a reference similarity;
When the similarity determination unit determines that there is a feature point candidate that exceeds the reference similarity in all of the plurality of determination parts, the feature point candidate with the highest similarity is characterized for each determination part. Feature point extraction means for extracting as points;
Normalized image generation means for generating a normalized image based on the feature points;
Normalized image storage means for storing the normalized image of the frame that has succeeded in extracting the feature points;
Feature point storage means for storing the feature points extracted by the feature point extraction means for each frame;
A prediction feature point region generation unit that reads out a feature point of the previous frame from the feature point storage unit when the feature point extraction fails, and generates a prediction feature point region centered on the feature point;
Feature point candidate determination means for determining whether or not at least one of the feature point candidates is included in the predicted feature point region when the predicted feature point region is generated;
When the feature point candidate determination unit determines that at least one of the feature point candidates is included in the predicted feature point region of all of the plurality of determination sites, the feature point candidate is stored in the normalized image storage unit. Normalized image storage and holding instruction means for instructing to hold the normalized image
A person authentication device comprising:

When it is determined by the feature point candidate determination unit that at least one of the feature point candidates is included in the predicted feature point region of all of the plurality of determination portions, the next frame is transmitted to the face region extraction unit. Face area continuous output instruction means for instructing to continuously output the face area of the current frame without performing face area extraction for screen input;
The person authentication device according to claim 2, further comprising:

A person authentication method for performing personal authentication using a frame screen of an input moving image,
In accordance with an instruction from the previous frame, determining whether to continue using the face area extracted in the previous frame or newly extract the face area for the screen of the input current frame;
Extracting a human face area from the screen of the current frame when the decision is a new extraction;
In response to the determination, extracting feature point candidates for each of a plurality of determination sites from a continuously used or newly extracted face region;
Comparing the feature point candidate with a pre-registered template and determining whether there is a feature point candidate exceeding a reference similarity;
Extracting a feature point candidate having the highest similarity as a feature point for each determination portion when it is determined that there is a feature point candidate exceeding the reference similarity in all of the plurality of determination portions; ,
Storing the extracted feature points in a storage means for each frame;
When it is determined that there is no feature point candidate exceeding the reference similarity in any of the plurality of determination parts, the feature point of the previous frame is read from the storage unit, and the predicted feature point centered on the feature point Generating a region;
Determining whether or not at least one feature point candidate is included in the predicted feature point region when the predicted feature point region is generated;
When it is determined that at least one of the feature point candidates is included in the predicted feature point regions of all of the plurality of determination regions, the face region of the current frame is determined for screen input of the next frame. Instructing them to continue using it,
A person authentication method comprising:

A person authentication method for accumulating a plurality of normalized images generated from a frame screen of an input moving image and performing personal authentication using the accumulated normalized images.
Extracting a human face area from the current frame screen;
Extracting feature point candidates for each of a plurality of determination sites from the face region;
Comparing the feature point candidate with a pre-registered template and determining whether there is a feature point candidate exceeding a reference similarity;
Extracting a feature point candidate having the highest similarity as a feature point for each determination portion when it is determined that there is a feature point candidate exceeding the reference similarity in all of the plurality of determination portions; ,
Generating a normalized image based on the feature points;
Accumulating the normalized image of a frame in which the feature points have been successfully extracted;
Storing the extracted feature points in a storage means for each frame;
When extraction of the feature points fails, reading the feature points of the previous frame from the storage means, and generating a predicted feature point region centered on the feature points;
Determining whether or not at least one feature point candidate is included in the predicted feature point region when the predicted feature point region is generated;
With
When it is determined that none of the feature point candidates is included in the predicted feature point region of all the plurality of determination parts, the accumulated normalized image is discarded,
When it is determined that at least one feature point candidate is included in the predicted feature point region of all of the plurality of determination parts, the accumulated normalized image is retained. Person authentication method.