TWI419060B - Human face recognition method using a pyramid sampling algorithm - Google Patents
Human face recognition method using a pyramid sampling algorithm Download PDFInfo
- Publication number
- TWI419060B TWI419060B TW98134869A TW98134869A TWI419060B TW I419060 B TWI419060 B TW I419060B TW 98134869 A TW98134869 A TW 98134869A TW 98134869 A TW98134869 A TW 98134869A TW I419060 B TWI419060 B TW I419060B
- Authority
- TW
- Taiwan
- Prior art keywords
- sampling
- pyramid
- image
- array
- sample
- Prior art date
Links
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Description
本發明係關於一種採用金字塔取樣演算法之人臉辨識方法；特別是關於利用一第一陣列取樣組[例如：菱形取樣組]及一第二陣列取樣組[例如：方形取樣組]之間以交替取樣方式進行金字塔取樣演算法之人臉辨識方法。The present invention relates to a face recognition method using a pyramid sampling algorithm; in particular, using a first array sampling group [eg, a diamond sampling group] and a second array sampling group [eg, a square sampling group] The face recognition method of pyramid sampling algorithm is performed by alternate sampling method.
一般而言，人臉影像辨識常因光線變化、人臉姿態差異、複雜背景及影像擷取距離差異等因素，導致相同類別的人臉影像在空間中的分佈過於分散，而不同類別的人臉影像在空間中的分佈形成重疊，因而降低辨識成功率。特別是人臉姿態變化是影響人臉類別分佈的主要原因。In general, face image recognition is often caused by light changes, differences in face poses, complex backgrounds, and differences in image capture distances, resulting in the distribution of face images of the same category in space, and different types of faces. The distribution of images in space forms an overlap, thus reducing the recognition success rate. In particular, the change of face pose is the main reason that affects the distribution of face categories.
近年來的人臉影像辨識技術常採用對數座標取樣法，以獲得大小不變、平移不變及旋轉不變的特性，以有效改善人臉姿態變異問題，且提升人臉影像辨識的成功率。雖然採用對數座標取樣法確實能有效提升人臉影像辨識的成功率，但是該取樣法過度依賴實際人臉正中心點的精確定位，方能成功達成人臉影像辨識作業。In recent years, the face image recognition technology often adopts the logarithmic coordinate sampling method to obtain the characteristics of constant size, translation invariance and rotation invariance, so as to effectively improve the face pose variation problem and improve the success rate of face image recognition. Although the logarithmic coordinate sampling method can effectively improve the success rate of face image recognition, the sampling method relies too much on the precise positioning of the actual center point of the face to successfully achieve the face image recognition operation.
舉例而言，在取樣定位作業中，若相對於實際人臉中心點存在一小偏移距離的變化量時，可產生非線性放大該變化量，並導致人臉影像辨識作業的失敗。另外，在取樣定位作業中，人臉影像相對於實際人臉平移亦造成影像取樣平均值異動。For example, in the sampling and positioning operation, if there is a small amount of change in the offset distance from the actual face center point, the amount of change can be nonlinearly amplified, and the face image recognition operation fails. In addition, in the sampling and positioning operation, the translation of the face image relative to the actual face also causes the average value of the image sample to be changed.
附照1揭示將人臉影像進行放大、縮小、平移及旋轉之各影像圖。附照1包含將人臉T字部位之原始影像分別進行放大1.25倍及1.5倍，縮小0.75倍及0.5倍，向左上、左下、右上及右下分別平移10個像素[pixel]，向左及右分別旋轉30度。Attachment 1 discloses each image of a face image that is enlarged, reduced, translated, and rotated. Attachment 1 includes magnifying the original image of the T-shaped part of the face by 1.25 times and 1.5 times, 0.75 times and 0.5 times, and shifting 10 pixels [pixel] to the left upper left, lower left, upper right and lower right respectively, to the left and Rotate 30 degrees to the right.
附照2揭示將附照1之人臉影像之放大、縮小、平移及旋轉影像圖利用對數極座標取樣方式進行取樣之取樣影像圖。附照2顯示放大1.25倍、放大1.5倍、縮小0.75倍及縮小0.5倍之取樣影像與取樣原始影像之間內容具有明顯差異；向左上平移、向左下平移、向右上平移及向右下平移之取樣影像與取樣原始影像之間具有紋路左右反轉之差異；向左旋轉30度及右旋轉30度之取樣影像與取樣原始影像之間內容具有差異。Attachment 2 discloses a sampled image image in which the enlarged, reduced, translated, and rotated image of the face image of the attached image is sampled by the logarithmic coordinate sampling method. Attachment 2 shows that there is a significant difference between the sampled image enlarged by 1.25 times, 1.5 times enlarged, 0.75 times smaller, and 0.5 times smaller than the sampled original image; panning to the left, panning to the left, panning to the right, and panning to the right. There is a difference between the sampled image and the sampled original image with the left and right inversion of the line; the content between the sampled image rotated 30 degrees to the left and 30 degrees to the right has a difference between the sampled image and the sampled original image.
因此，該人臉影像辨識技術之成功與否係取決於在取樣定位作業時，是否能成功精確定位在人臉的中心點。若該取樣定位作業無法精確定位在人臉的中心點時，必然導致後續人臉影像辨識作業的失敗結果。是以，習用對數座標取樣法仍存在有必要進一步改善其需要精確定位在人臉中心點的問題。Therefore, the success of the face image recognition technology depends on whether it can be accurately positioned at the center point of the face during the sampling and positioning operation. If the sampling and positioning operation cannot be accurately located at the center point of the face, it will inevitably lead to the failure of the subsequent face image recognition operation. Therefore, the conventional logarithmic coordinate sampling method still needs to further improve the problem that it needs to be accurately positioned at the center point of the face.
關於人臉辨識技術，其亦揭示於部分國內專利之技術內容。舉例而言，中華民國發明專利公開第200923800號之〝結合投影與主成分分析之二維人臉辨識方法〞專利案、第200915216號之〝以適應性資料高斯核心為基礎進行臉部特徵擷取的人臉辨識特徵擷取方法〞專利案、第200707310號之〝以人臉五官辨識為基礎之人臉辨識方法〞專利案及第200725433號之〝三維人臉辨識系統及其方法〞專利案，前述中華民國專利公開案僅為本發明技術背景之參考及說明目前技術發展狀態而已，其並非用以限制本發明之範圍。Regarding the face recognition technology, it is also disclosed in the technical content of some domestic patents. For example, the Republic of China Patent Publication No. 200923800 combines the two-dimensional face recognition method of projection and principal component analysis, the patent case, No. 200915216, based on the adaptive data Gaussian core for facial feature extraction. Face recognition feature extraction method 〞 Patent Case, No. 200707310, face recognition method based on face facial features recognition, patent case and No. 200725433 〝 three-dimensional face recognition system and its method 〞 patent case, The foregoing disclosure of the Patent of the Republic of China is merely a reference to the technical background of the present invention and the state of the art is not intended to limit the scope of the present invention.
關於人臉辨識技術，其亦揭示於許多美國專利之技術內容。舉例而言，美國專利第7,426,292號、第7,406,184號、第7,391,889號、第7,369,686號、第7,308,133號、第7,295,687號、第7,221,809號、第7,203,346號、第7,177,450號、第7,095,879號、第6,975,750號、第6,108,437號及第5,699,449號等；美國專利公開案第20090238419號、第20090232365號、第20090195638號、第20090180673號、第20090180671號、第20090059061號、第20090041309號、第20090016575號、第20080273766號、第20080247611號、第20080166026號、第20080159601號、第20080144941號、第20080122943號、第20080107311號、第20070297652號、第20070258627號、第20070200916號、第20070172099號、第20070160296號、第20070152037號、第20070127787號、第20070122009號、第20070104362號、第20070071288號、第20070041644號、第20060280344號、第20060280341號、第20060158307號、第20060120571號、第20060104504號、第20060093238號、第20060018522號、第20050147280號、第20050123202號、第20050117783號、第20050105806號、第20050084137號、第20050063569號、第20050036649號、第20050031196號、第20040170305號、第20040151347號、第20040119851號、第20040034611號、第20040005086號、第20030215115號、第20030123713號、第20030063796號、第20030063780號、第20020113687號、第20020106114號及第20010031073號。前述諸美國專利及專利公開案僅為本發明技術背景之參考及說明目前技術發展狀態而已，其並非用以限制本發明之範圍。Regarding face recognition technology, it is also disclosed in the technical content of many U.S. patents. For example, U.S. Patent Nos. 7,426,292, 7,406,184, 7,391,889, 7,369,686, 7,308,133, 7,295,687, 7,221,809, 7,203,346, 7,177,450, 7,095,879, 6,975,750, No. 6,108,437 and 5,699,449, etc.; U.S. Patent Publication Nos. 20080238419, 20090232365, 20090195638, 20090180673, 20090180671, 20090059061, 20090041309, 20090016575, 20080273766, 20080247611, No. 20080166026, No. 20080159601, No. 20080144941, No. 20080122943, No. 20080107311, No. 20070297652, No. 20070258627, No. 20070200916, No. 20070172099, No. 20070160296, No. 20070152037, No. 20070127787 , 20070122009, No. 20070104362, No. 20070071288, No. 20070041644, No. 20060280344, No. 20060280341, No. 20060158307, No. 20060120571, No. 20060104504, No. 20060093238, No. 20060018522, No. 20050147280, 20050123202, the number of 20050117783 , No. 20050105806, No. 20050084137, No. 20050053569, No. 20050036649, No. 20050511196, No. 20040170305, No. 20040151347, No. 20040119851, No. 2004034611, No. 2004005086, No. 20030215115, No. 20030123713, No. 20030063796, No. 20020033780, No. 2001023687, No. 20060106114, and No. 20010031073. The above-mentioned U.S. patents and patent publications are only for the purpose of the present invention and are not intended to limit the scope of the present invention.
有鑑於此，本發明為了滿足上述需求，其提供一種採用金字塔取樣演算法之人臉辨識方法，其在一人臉影像座標上利用一第一陣列取樣組及一第二陣列取樣組進行交替取樣，以達成提升人臉影像辨識成功率之目的。In view of the above, the present invention provides a face recognition method using a pyramid sampling algorithm, which uses a first array sampling group and a second array sampling group for alternate sampling on a face image coordinate. In order to achieve the purpose of improving the success rate of face recognition.
本發明之主要目的係提供一種採用金字塔取樣演算法之人臉辨識方法，其在一人臉影像座標上利用一第一陣列取樣組及一第二陣列取樣組進行交替取樣，以達成提升人臉影像辨識成功率之目的。The main object of the present invention is to provide a face recognition method using a pyramid sampling algorithm, which uses a first array sampling group and a second array sampling group to alternately sample on a human face image coordinate to achieve an enhanced face image. Identify the purpose of success.
為了達成上述目的，本發明之採用金字塔取樣演算法之人臉辨識方法包含步驟：提供一人臉影像座標；利用一第一陣列取樣組進行取樣數個第一取樣影像；利用一第二陣列取樣組進行取樣數個第二取樣影像；利用該第一取樣影像及第二取樣影像依序組成一金字塔取樣影像。In order to achieve the above object, the face recognition method using the pyramid sampling algorithm of the present invention comprises the steps of: providing a face image coordinate; using a first array sampling group to sample a plurality of first sample images; using a second array sampling group And sampling a plurality of second sample images; and sequentially forming a pyramid sample image by using the first sample image and the second sample image.
本發明較佳實施例之人臉辨識方法另包含步驟：將該金字塔取樣影像進行擷取特徵及辨識。The face recognition method of the preferred embodiment of the present invention further includes the step of: capturing the feature and identifying the pyramid sample image.
本發明較佳實施例之該人臉影像座標係採用卡笛爾座標系統。In the preferred embodiment of the present invention, the face image coordinate system adopts a Cartesian coordinate system.
本發明較佳實施例之該人臉影像座標係對應於一T型人臉影像。In the preferred embodiment of the present invention, the face image coordinate system corresponds to a T-type face image.
本發明較佳實施例之該第一陣列取樣組係屬一菱形取樣組。In the preferred embodiment of the present invention, the first array sampling group is a diamond sampling group.
本發明較佳實施例之該菱形取樣組具有四個取樣點。The diamond shaped sampling set of the preferred embodiment of the invention has four sampling points.
本發明較佳實施例之該第二陣列取樣組係屬一方形取樣組。In the preferred embodiment of the present invention, the second array sampling group is a square sampling group.
本發明較佳實施例之該方形取樣組具有四個取樣點。The square sampling set of the preferred embodiment of the invention has four sampling points.
為了充分瞭解本發明，於下文將例舉較佳實施例並配合所附圖式作詳細說明，且其並非用以限定本發明。In order to fully understand the present invention, the preferred embodiments of the present invention are described in detail below and are not intended to limit the invention.
本發明較佳實施例之採用金字塔取樣演算法之人臉辨識方法適用於各種人臉影像辨識系統之任何辨識方法，且適用於T字部位人臉影像[以下簡稱：T型人臉影像]之人臉影像辨識作業，但其並非用以限定本發明之適用範圍。The face recognition method using the pyramid sampling algorithm in the preferred embodiment of the present invention is applicable to any identification method of various face image recognition systems, and is applicable to a face image of a T-shaped part (hereinafter referred to as a T-type face image). The face image recognition operation is not intended to limit the scope of application of the present invention.
本發明較佳實施例之金字塔取樣演算法亦可稱為梅花型金字塔取樣演算法[quincunx pyramid sampling algorithm]，但其並非用以限定本發明。The pyramid sampling algorithm of the preferred embodiment of the present invention may also be referred to as a quincunx pyramid sampling algorithm, but it is not intended to limit the present invention.
本發明之採用金字塔取樣演算法之人臉辨識方法包含第一步驟：提供一人臉影像座標。本發明較佳實施例之該人臉影像座標係採用卡笛爾座標系統[Cartesian coordinate]，且該人臉影像座標係對應於一T型人臉影像，即該人臉影像座標之原始點對應於該T型人臉影像之中心點。The face recognition method using the pyramid sampling algorithm of the present invention comprises the first step of providing a face image coordinate. In the preferred embodiment of the present invention, the face image coordinate system adopts a Cartesian coordinate system, and the face image coordinate system corresponds to a T-type face image, that is, the original point corresponding to the face image coordinate At the center of the T-shaped face image.
接著，第1圖揭示本發明較佳實施例之採用金字塔取樣演算法之人臉辨識方法在人臉影像座標上利用第一陣列取樣組及第二陣列取樣組進行交替取樣之示意圖。Next, FIG. 1 is a schematic diagram showing a face recognition method using a pyramid sampling algorithm for alternately sampling a first array sampling group and a second array sampling group on a face image coordinate according to a preferred embodiment of the present invention.
本發明較佳實施例之採用金字塔取樣演算法需要適當決定金字塔取樣密度，該金字塔取樣密度之取得係利用方程式(1)如下：The pyramid sampling algorithm of the preferred embodiment of the present invention needs to appropriately determine the pyramid sampling density. The pyramid sampling density is obtained by using equation (1) as follows:
其中γ為梅花型金字塔取樣點間距，T w 為T型人臉影像寬度，T h 為T型人臉影像高度，N 為梅花金字塔數量。Where γ is the spacing of the sampling points of the plum-shaped pyramid, T w is the width of the T-shaped face image, T h is the height of the T-shaped face image, and N is the number of plum pyramids.
請參照第1圖所示，本發明之採用金字塔取樣演算法之人臉辨識方法包含第二步驟：利用一第一陣列取樣組1在該T型人臉影像上決定數個第一取樣點10，以便依數個該第一取樣點10進行取樣數個第一取樣影像T 。本發明較佳實施例之該第一陣列取樣組1係屬一菱形取樣組，如第1圖之箭頭所示，或其它幾何形狀取樣組，其並非用以限制本發明。本發明較佳實施例之該菱形取樣組具有四個取樣點，即該第一陣列取樣組1具有四個該第一取樣點10。Referring to FIG. 1 , the face recognition method using the pyramid sampling algorithm of the present invention includes a second step of determining a plurality of first sampling points 10 on the T-type face image by using a first array sampling group 1 . So that a plurality of first sample images T are sampled according to the plurality of first sampling points 10. The first array sampling group 1 of the preferred embodiment of the present invention is a diamond sampling group, as indicated by the arrows in FIG. 1, or other geometric sampling groups, which are not intended to limit the present invention. In the preferred embodiment of the present invention, the diamond sampling group has four sampling points, that is, the first array sampling group 1 has four of the first sampling points 10.
請再參照第1圖所示，本發明之採用金字塔取樣演算法之人臉辨識方法包含第三步驟：利用一第二陣列取樣組2在該T型人臉影像上決定數個第二取樣點20，以便依數個該第二取樣點20進行取樣數個第二取樣影像T 。本發明較佳實施例之該第二陣列取樣組2係屬一方形取樣組，如第1圖之箭頭所示，或其它幾何形狀取樣組，其並非用以限制本發明。本發明較佳實施例之該方形取樣組具有四個取樣點，即該第二陣列取樣組2具有四個該第二取樣點20。本發明之採用金字塔取樣演算法之人臉辨識方法在第二步驟及第三步驟之間可適當進行兩者相互對調或兩者組合執行，因此該第二步驟及第三步驟之順序並非用以限制本發明。Referring to FIG. 1 again, the face recognition method using the pyramid sampling algorithm of the present invention includes a third step of determining a plurality of second sampling points on the T-type face image by using a second array sampling group 2 20, so as to sample a plurality of second sample images T according to the plurality of second sampling points 20. The second array of sample sets 2 of the preferred embodiment of the present invention is a square sample set, as indicated by the arrows in Figure 1, or other geometric sample sets, which are not intended to limit the invention. In the preferred embodiment of the present invention, the square sampling group has four sampling points, that is, the second array sampling group 2 has four of the second sampling points 20. The face recognition method using the pyramid sampling algorithm of the present invention can perform the mutual intermodulation or the combination of the two between the second step and the third step, so the order of the second step and the third step is not used. Limit the invention.
此時，本發明較佳實施例之金字塔取樣演算法在取得金字塔取樣影像Q 之第一取樣點及第二取樣點上利用方程式(2)、(3)、(4)如下：At this time, the pyramid sampling algorithm of the preferred embodiment of the present invention uses equations (2), (3), and (4) to obtain the first sampling point and the second sampling point of the pyramid sampling image Q as follows:
其中Q (n ,k )為金字塔影像，n 為第幾個金字塔，k 為在金字塔中第幾個取樣點，c 為取樣密度變化週期，其值設為10，m 為整數。Where Q ( n , k ) is the pyramid image, n is the first pyramid, k is the first sampling point in the pyramid, c is the sampling density change period, and its value is set to 10, m is an integer.
請再參照第1圖所示，本發明之金字塔取樣演算法利用方程式(3)計算每個該第一陣列取樣組[菱形取樣組]1之四個該第一取樣點10，其中該第一陣列取樣組1依序為第0個菱形取樣組至第49個菱形取樣組，該第一取樣點10依序為第0個取樣點至第3個取樣點[如第1圖之星形符號所示]。Referring again to FIG. 1, the pyramid sampling algorithm of the present invention calculates four of the first sampling points 10 of each of the first array sampling group [diamond sampling group] 1 using equation (3), wherein the first The array sampling group 1 is sequentially from the 0th diamond sampling group to the 49th diamond sampling group, and the first sampling point 10 is sequentially from the 0th sampling point to the 3rd sampling point [such as the star symbol of FIG. 1 Shown].
請再參照第1圖所示，本發明之金字塔取樣演算法利用方程式(4)計算每個該第二陣列取樣組[方形取樣組]2之四個該第二取樣點20，其中該第二陣列取樣組2依序為第0個菱形取樣組至第49個菱形取樣組，該第二取樣點20依序為第0個取樣點至第3個取樣點[如第1圖之星形符號所示]。Referring again to FIG. 1, the pyramid sampling algorithm of the present invention calculates four of the second sampling points 20 of each of the second array sampling group [square sampling group] 2 using equation (4), wherein the second sampling point 20 The array sampling group 2 is sequentially from the 0th diamond sampling group to the 49th diamond sampling group, and the second sampling point 20 is sequentially from the 0th sampling point to the 3rd sampling point [such as the star symbol of FIG. 1 Shown].
請再參照第1圖所示，本發明之金字塔取樣演算法係將該第一陣列取樣組1及第二陣列取樣組2相對於該人臉影像座標之原始參考點(0,0)進行堆疊排列而形成一基本金字塔狀單元。再者，每個該第一陣列取樣組1及第二陣列取樣組2之組合共同提供四個該第一取樣點10及四個該第二取樣點20，且該第一陣列取樣組1及第二陣列取樣組2之組合相對於人臉影像中心點可依序向外排列，因而在預定金字塔數量[例如：0≦n＜49(N )]範圍內不斷將該第一陣列取樣組1及第二陣列取樣組2進行交替重疊。Referring again to FIG. 1, the pyramid sampling algorithm of the present invention stacks the first array sampling group 1 and the second array sampling group 2 with respect to the original reference point (0, 0) of the face image coordinate. Arranged to form a basic pyramidal unit. Furthermore, each combination of the first array sampling group 1 and the second array sampling group 2 provides four first sampling points 10 and four second sampling points 20, and the first array sampling group 1 and The combination of the second array sampling group 2 can be sequentially arranged outward with respect to the center point of the face image, and thus the first array sampling group 1 is continuously continued within a predetermined number of pyramids [eg, 0≦n<49( N )] And the second array of sampling groups 2 are alternately overlapped.
第2圖揭示本發明較佳實施例之採用金字塔取樣演算法之人臉辨識方法利用第一陣列取樣組及第二陣列取樣組進行交替取樣後，取得金字塔取樣影像之示意圖。請參照第1及2圖所示，利用該第一陣列取樣組1之第一取樣點10及該第二陣列取樣組2之第二取樣點20決定該第一取樣影像及第二取樣影像T ；再利用方程式(3)及(4)計算後，獲得每個該第一取樣點10之第一取樣影像Q 及每個該第二取樣點20之第二取樣影像Q ，如第2圖所示。該第一陣列取樣組1之第一取樣點10及該第二陣列取樣組2之第二取樣點20之值，其值係由3×3遮罩像素之平均值[mean value]取得。FIG. 2 is a schematic diagram showing a pyramid sampling image obtained by alternately sampling the first array sampling group and the second array sampling group by using the pyramid sampling algorithm in the face recognition method according to the preferred embodiment of the present invention. The first sampling image and the second sampling image T are determined by using the first sampling point 10 of the first array sampling group 1 and the second sampling point 20 of the second array sampling group 2, as shown in FIGS. 1 and 2. ; then using equation (3) and (4) are calculated and obtained for each sampling point a first sample of the first image and a Q 10 of each of the second sampling point of the second sample image Q 20, FIG. 2 as the Show. The values of the first sampling point 10 of the first array sampling group 1 and the second sampling point 20 of the second array sampling group 2 are obtained by the mean value of the 3×3 mask pixels.
請再參照第2圖所示，本發明之採用金字塔取樣演算法之人臉辨識方法包含第四步驟：利用該第一取樣影像Q 及第二取樣影像Q 組成一金字塔取樣影像。本發明較佳實施例之金字塔取樣影像之橫軸依金字塔n 之序列進行排列，其縱軸依在金字塔中取樣點k 序列進行排列，以便依序組成該金字塔取樣影像。Referring to FIG. 2 again, the face recognition method using the pyramid sampling algorithm of the present invention includes a fourth step of forming a pyramid sample image by using the first sample image Q and the second sample image Q. In the preferred embodiment of the present invention, the horizontal axis of the pyramid sample image is arranged according to the sequence of the pyramid n , and the vertical axis thereof is arranged according to the sequence of sample points k in the pyramid to sequentially form the pyramid sample image.
舉例而言，請再參照第1及2圖所示，本發明之金字塔取樣演算法利用方程式(3)及(4)計算後，在金字塔0獲得取樣影像Q 為Q (0,0)、Q (0,1)、Q (0,2)、Q (0,3)、Q (0,4)、Q (0,5)、Q (0,6)、Q (0,7)，並將該取樣影像Q 依序組合形成該金字塔取樣影像，如第2圖所示。For example, referring to FIGS. 1 and 2, the pyramid sampling algorithm of the present invention calculates the sample image Q in the pyramid 0 by using equations (3) and (4), and Q is Q (0, 0), Q. (0,1), Q (0,2), Q (0,3), Q (0,4), Q (0,5), Q (0,6), Q (0,7), and The sampled images Q are sequentially combined to form the pyramid sample image, as shown in FIG.
附照3揭示將附照1之人臉影像之放大、縮小、平移及旋轉影像圖利用本發明較佳實施例之金字塔取樣方式進行取樣之取樣影像圖。附照3顯示放大1.25倍、放大1.5倍、縮小0.75倍及縮小0.5倍之取樣影像與取樣原始影像之間不具有明顯差異；向左上平移、向左下平移、向右上平移及向右下平移之取樣影像相對於取樣原始影像產生變化，但其仍維持主要紋路；向左旋轉30度及右旋轉30度之取樣影像相對於取樣原始影像產生變化，但其仍維持主要紋路。Attachment 3 discloses a sampled image map in which the enlarged, reduced, translated, and rotated image of the human face image of the attached image is sampled by the pyramid sampling method of the preferred embodiment of the present invention. Attachment 3 shows that there is no significant difference between the sampled image that is magnified by 1.25 times, magnified by 1.5 times, reduced by 0.75 times, and reduced by 0.5 times, and the sampled original image; panning to the left, panning to the left, panning to the right, and panning to the right. The sampled image changes relative to the sampled original image, but it still maintains the main texture; the sampled image rotated 30 degrees to the left and 30 degrees to the right produces a change relative to the sampled original image, but it still maintains the main texture.
本發明之採用金字塔取樣演算法之人臉辨識方法另包含步驟：將該金字塔取樣影像進行擷取特徵及辨識，但其並非用以限制本發明。The face recognition method using the pyramid sampling algorithm of the present invention further comprises the steps of: capturing the feature and identification of the pyramid sample image, but it is not intended to limit the present invention.
第3圖揭示本發明較佳實施例之採用金字塔取樣演算法之人臉辨識方法及採用對數極座標取樣演算法達成辨識成功率之走勢圖。在人臉辨識實驗上，使用彩色FERET人臉資料庫，並選取100人，且分為5人、50人及100人組合。將本發明之金字塔取樣影像與對數極座標取樣影像饋入主能量分析[PCA]或自我主能量分析[Self-PCA]，利用該兩方法先行擷取特徵。接著，利用基因演算法[genetic algorithms]結合留一法[leave-one-out method]，再藉由適應函數[fitness function]的引導，經不斷的遞迴演化重組計算，最後求得最佳的特徵組合應用於辨識。FIG. 3 is a diagram showing a face recognition method using a pyramid sampling algorithm and a logarithmic coordinate sampling algorithm for achieving a recognition success rate according to a preferred embodiment of the present invention. In the face recognition experiment, the color FERET face database was used, and 100 people were selected and divided into 5, 50 and 100 combinations. The pyramid sample image and the log polar coordinate sample image of the present invention are fed into a main energy analysis [PCA] or a self-primary energy analysis [Self-PCA], and the two methods are used to extract features first. Then, using the genetic algorithms to combine the leave-one-out method, and then through the guidance of the fitness function, the evolutionary reorganization and recalculation are continuously regressed, and finally the best is obtained. Feature combinations are applied to the identification.
其中本發明之金字塔取樣訓練辨識結果為表1，對數極座標取樣訓練辨識結果為表2。The pyramid sampling training identification result of the present invention is shown in Table 1, and the logarithmic coordinate sampling training identification result is Table 2.
請參照第3圖所示，本發明之金字塔取樣與對數極座標取樣顯示於表1及2之比較結果，其包含四條走勢曲線[標示符號為圓形、星形、三角形、菱形]；其中圓形走勢曲線及三角形走勢曲線分別代表利用對數極座標取樣進行主能量分析[PCA]及自我主能量分析[Self-PCA]之結果；其中星形走勢曲線及菱形走勢曲線分別代表利用本發明之金字塔取樣取樣進行主能量分析[PCA]及自我主能量分析[Self-PCA]之結果。Referring to FIG. 3, the pyramid sampling and the logarithmic polar coordinate sampling of the present invention are shown in the comparison results of Tables 1 and 2, which include four trend curves [marked symbols are circles, stars, triangles, diamonds]; The trend curve and the triangle trend curve respectively represent the results of the main energy analysis [PCA] and the self-primary energy analysis [Self-PCA] using the log polar coordinates sampling; wherein the star trend curve and the diamond trend curve respectively represent the pyramid sampling sample using the present invention. The results of primary energy analysis [PCA] and self-primary energy analysis [Self-PCA] were performed.
請再參照第3圖所示，由表1及2之比較結果可發現本發明之金字塔取樣與對數極座標取樣在資料庫成員人數越來越多的情況下，本發明之金字塔取樣的辨識率下降幅度相對比較小。Referring to FIG. 3 again, it can be found from the comparison results of Tables 1 and 2 that the pyramid sampling and the log polar coordinate sampling of the present invention are reduced in the number of members of the database, and the recognition rate of the pyramid sampling of the present invention is decreased. The magnitude is relatively small.
上述實驗數據為在特定條件之下所獲得的初步實驗結果，其僅用以易於瞭解或參考本發明之技術內容而已，其尚需進行其他實驗。該實驗數據及其結果並非用以限制本發明之權利範圍。The above experimental data is preliminary experimental results obtained under specific conditions, which are only used to easily understand or refer to the technical content of the present invention, and other experiments are still required. The experimental data and its results are not intended to limit the scope of the invention.
前述較佳實施例僅舉例說明本發明及其技術特徵，該實施例之技術仍可適當進行各種實質等效修飾及/或替換方式予以實施；因此，本發明之權利範圍須視後附申請專利範圍所界定之範圍為準。The foregoing preferred embodiments are merely illustrative of the invention and the technical features thereof, and the techniques of the embodiments can be carried out with various substantial equivalent modifications and/or alternatives; therefore, the scope of the invention is subject to the appended claims. The scope defined by the scope shall prevail.
1．．．第一陣列取樣組1. . . First array sampling group
10．．．第一取樣點10. . . First sampling point
2．．．第二陣列取樣組2. . . Second array sampling group
20．．．第二取樣點20. . . Second sampling point
Q ．．．取樣影像 Q . . . Sampled image
第1圖：本發明較佳實施例之採用金字塔取樣演算法之人臉辨識方法在人臉影像座標上利用第一陣列取樣組及第二陣列取樣組進行交替取樣之示意圖。FIG. 1 is a schematic diagram of a face recognition method using a pyramid sampling algorithm according to a preferred embodiment of the present invention for alternate sampling using a first array sampling group and a second array sampling group on a face image coordinate.
第2圖：本發明較佳實施例之採用金字塔取樣演算法之人臉辨識方法利用第一陣列取樣組及第二陣列取樣組進行交替取樣後，取得金字塔取樣影像之示意圖。2: A face recognition method using a pyramid sampling algorithm according to a preferred embodiment of the present invention uses a first array sampling group and a second array sampling group to perform alternate sampling, and then obtains a schematic diagram of a pyramid sampling image.
第3圖：本發明較佳實施例之採用金字塔取樣演算法之人臉辨識方法及採用對數極座標取樣演算法達成辨識成功率之走勢圖。FIG. 3 is a diagram showing a face recognition method using a pyramid sampling algorithm and a logarithmic coordinate sampling algorithm for achieving a recognition success rate according to a preferred embodiment of the present invention.
附照1：將人臉影像進行放大、縮小、平移及旋轉之各影像圖。Attachment 1: Image of the face image is enlarged, reduced, translated, and rotated.
附照2：將附照1之人臉影像之放大、縮小、平移及旋轉影像圖利用對數極座標取樣方式進行取樣之取樣影像圖。Attachment 2: A sample image of a sample that is sampled by a logarithmic coordinate sampling method by enlarging, reducing, panning, and rotating the image of the face of the attached image 1.
附照3：將附照1之人臉影像之放大、縮小、平移及旋轉影像圖利用本發明較佳實施例之金字塔取樣方式進行取樣之取樣影像圖。Attachment 3: Amplified, reduced, translated, and rotated image of the face image of the attached image 1 is sampled by sampling in a pyramid sampling manner according to a preferred embodiment of the present invention.
1．．．第一陣列取樣組1. . . First array sampling group
10．．．第一取樣點10. . . First sampling point
2．．．第二陣列取樣組2. . . Second array sampling group
20．．．第二取樣點20. . . Second sampling point
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW98134869A TWI419060B (en) | 2009-10-15 | 2009-10-15 | Human face recognition method using a pyramid sampling algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW98134869A TWI419060B (en) | 2009-10-15 | 2009-10-15 | Human face recognition method using a pyramid sampling algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| TW201113820A TW201113820A (en) | 2011-04-16 |
| TWI419060B true TWI419060B (en) | 2013-12-11 |
Family
ID=44909804
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW98134869A TWI419060B (en) | 2009-10-15 | 2009-10-15 | Human face recognition method using a pyramid sampling algorithm |
Country Status (1)
| Country | Link |
|---|---|
| TW (1) | TWI419060B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI618027B (en) * | 2016-05-04 | 2018-03-11 | 國立高雄應用科技大學 | 3d hand gesture image recognition method and system thereof with ga |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW466452B (en) * | 1998-11-18 | 2001-12-01 | Sintec Co Ltd | Method and apparatus for photographing/recognizing a face |
| TW200707310A (en) * | 2005-08-08 | 2007-02-16 | Chunghwa Telecom Co Ltd | Facial recognition method based on recognition of facial features |
| US7391889B2 (en) * | 2002-10-15 | 2008-06-24 | Samsung Electronics Co., Ltd. | Method and apparatus for extracting feature vector used for face recognition and retrieval |
| US20090238419A1 (en) * | 2007-03-05 | 2009-09-24 | Fotonation Ireland Limited | Face recognition training method and apparatus |
-
2009
- 2009-10-15 TW TW98134869A patent/TWI419060B/en not_active IP Right Cessation
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW466452B (en) * | 1998-11-18 | 2001-12-01 | Sintec Co Ltd | Method and apparatus for photographing/recognizing a face |
| US7391889B2 (en) * | 2002-10-15 | 2008-06-24 | Samsung Electronics Co., Ltd. | Method and apparatus for extracting feature vector used for face recognition and retrieval |
| TW200707310A (en) * | 2005-08-08 | 2007-02-16 | Chunghwa Telecom Co Ltd | Facial recognition method based on recognition of facial features |
| US20090238419A1 (en) * | 2007-03-05 | 2009-09-24 | Fotonation Ireland Limited | Face recognition training method and apparatus |
Also Published As
| Publication number | Publication date |
|---|---|
| TW201113820A (en) | 2011-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107154023B (en) | Based on the face super-resolution reconstruction method for generating confrontation network and sub-pix convolution | |
| CN106716450B (en) | Image-based feature detection using edge vectors | |
| US8861800B2 (en) | Rapid 3D face reconstruction from a 2D image and methods using such rapid 3D face reconstruction | |
| CN108491786B (en) | Face detection method based on hierarchical network and cluster merging | |
| CN110458898A (en) | Camera calibration plate, nominal data acquisition method, distortion correction method and device | |
| CN101458763B (en) | Automatic human face identification method based on image weighting average | |
| CN104361314B (en) | Based on infrared and transformer localization method and device of visual image fusion | |
| WO2015139574A1 (en) | Static object reconstruction method and system | |
| CN108009222B (en) | Three-dimensional model retrieval method based on better view and deep convolutional neural network | |
| CN104008542B (en) | A kind of Fast Corner matching process for specific plane figure | |
| CN105069746A (en) | Video real-time human face substitution method and system based on partial affine and color transfer technology | |
| CN106919944A (en) | A kind of wide-angle image method for quickly identifying based on ORB algorithms | |
| CN105654421B (en) | Based on the projective transformation image matching method for converting constant low-rank texture | |
| CN111274944A (en) | Three-dimensional face reconstruction method based on single image | |
| CN107507263B (en) | Texture generation method and system based on image | |
| CN103733225B (en) | Characteristic point peer system, characteristic point counterpart method and record medium | |
| CN112418250B (en) | Optimized matching method for complex 3D point cloud | |
| TWI419060B (en) | Human face recognition method using a pyramid sampling algorithm | |
| US20200250878A1 (en) | Method of displaying an object | |
| CN111047513B (en) | Robust image alignment method and device for cylindrical panorama stitching | |
| CN113379899B (en) | Automatic extraction method for building engineering working face area image | |
| CN113222990B (en) | Chip counting method based on image data enhancement | |
| CN108537855A (en) | A kind of ceramic marble paper method for generating pattern and device that sketch is consistent | |
| CN108268533A (en) | A kind of Image Feature Matching method for image retrieval | |
| CN114463534A (en) | Target key point detection method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| MM4A | Annulment or lapse of patent due to non-payment of fees |