M294069 八、新型說明： 【新型所屬之技術領域】 本案係為-種掌雜齡統的改良，其巾细細胞神經系統 網路(Cellular Neural Netw〇rks，以下簡稱⑽)的基本特性為依 ，· 據做為掌形辨識系統的影像前置處理。藉由輸入一張想要辨識身 份的掌形圖像，經過-連φ上的處理後，將此圖片轉換產生出有 意義的數值，再將這些數值和資料庫的資料進行比對，從中獲得 • 雜辨識的結果。而此構成的辨識系統分為三大部份：第一部份 為掌形影像前置處理及輪廓化；第二部份為f形特徵關取；第 二部份為掌形特徵點比對辨識。 【先前技術】 於1988年，被稱為細胞神經網路的嶄新資訊處理系統是由l 〇· hua〃L· Yang所共同提出的架構，此細胞神經網路系統擁有局部 連、、。的特|± ’目此可容易地實現單層或多層_路結構，由於具 _ 卩連、纟"轉性’所財常適合用在影像處理上且可在高速的 狀態下運作，細胞神經網路也發展出其它的應用，這包括信號的 處理、解決祕化的問題、雜物體的速度檢測及超大型積體電 路實現等等。 類神左網路疋人工智慧領域巾重要的技術之―，其靈感源自 於::經學’希望透過模擬人腦結構的方式，來建立一個可以主 動=習、主動思考等能力的電腦模式，使得電腦能像人類-樣具 …近年來，類神經網路在訊號處理中已經有廣泛的研究， 5 M294069 例：濾波(filtering)、訊號偵測(signal detection)、系統鑑別 (system identification)、參數估測（parameter estimation)、 訊號壓縮··等方面，所處理的訊號則包括影音、語音、影像、通 訊··等，而細胞神經系統網路Cellular Neural Networks(CM) 由LEON 0· CHUA所提出的架構，而依據其理論所建立的細胞神經 系統網路通用機器（CNN Universal Machine)具有低成本 '低功率 等優點，是一台比數位信號處理(1)3{))快1〇〇〇倍的超級電腦，適 合應用於純處理。基於如此優越之躲，在此將顧細胞神經 系統網路(CNN)理論來研發掌形辨識系統。 【新型内容】 本案係為-種掌形辨齡統驗良，其中利用細胞神經系統網 路(Cellular Neural Netw〇rks ’以下簡稱⑽)的基本特性為依據 做為掌形辨識系統的影像前置處理。藉由輸人—張想要辨識身份 的掌形圖像，經過—連串上的處理後，將此圖片轉換產生出有意 義的數值’再將這些數值和㈣庫的資料進行比對，從中 份辨識的結果。 & # 此構成的辨識系統分為三大部份·· ; 置處理至掌形輪廓化的掌形影像前置處理系統。 2.將旱形做特徵關取的掌形特徵關取處理系統。 進而達到本案創作以Matlab和 統網路架構為主軸的軟體，有效進行套以細胞神經系 负政進仃對掌形圖片的處理，並 M294069 從中模擬_'轉及驗證。 【實施方式】 請先參閱本案圖式 _ 流程圖，σ 圖所示為本案創作的掌形辨識處理 /、顺錢分為三大部份： 2. 频辄财形麟前纽理系統。 ,故特徵點褐取的掌形特徵關取處理系統。 肩徵耻對觸的掌轉徵點輯辨贼理系統。 ㈣\、°，111此必須先經過掌形影像前置處理祕的前置 =^仏^程序’分別是灰階處理、影像二值化、影像 、、、^並針對邊緣化後的掌形輪廓娜特難做比對的依據。 〔掌形影像前置處理系統的前置處理部分〕 其中灰階處理的部份是利用方程式〔如附件1所示〕將彩色 的輸入掌形圖片轉換為灰階形式。 衫像-值化部侧是將細處理後的灰階影像做像素值分佈 並做統計後如第2圖，第2圖巾橫軸代表灰階值；縱軸代表在某 灰階值的錄。其巾魏大部分的像素时佈在兩侧，所以很清 楚的發現_有祕鴨的峰舰域，分默灰雜較低的背景 (灰階值70以下），以及灰階值較高的手部(灰階值14〇以上），所以 灰階值70〜140的範圍就是手掌邊緣的部份，因此本案實施例選擇 70作為臨界值(Z)來對影像做二值化處理的參設定，而灰階影像 經由二值化處理後，其中灰階值大於70則趨向白色(手部），反之 M294069 趨向於黑色（背景），然後以細胞神經系統網路-門檻 (CNN-THRESHOLD)將影像轉為黑白圖片，其參數如第3圖以及 附件2之方程式所示。 灰階影像經過二值化處理後，影像已經變成了黑和白兩種顏 色。接下來就是利用邊緣偵測的觀念，在二值化影像結果中黑色 和白色的交接處取出。而此交界處就是想要得到的掌形輪廓。邊 緣偵測，顧名思義是描繪物體的邊緣並呈現在影像上，在此採用 的是細胞神經系統網路_灰階邊緣(CNN_EDGEGRAY)的方法取得 掌形輪廓，其參數如第4圖以及附件3之方程式所示。 在影像處理過程當中，可以從邊緣偵測後的曲線觀查其變 化，來找到曲線的特徵。曲線追蹤法的目的在於可以完整找出曲 線的走向，並且可以逐一記錄其座標及編號，在之後的特徵擷取 是非常重要的資訊。而在曲線追蹤的方法，由於掌形當中的拇指 疋右下方，因此可以設定所要掃瞄的點從最右下角開始掃瞄，所 以設計兩個遮罩結構掃瞄方向都是由右到左、下到上。丛^^ 益慧範，第5a圖代表一曲線所有點的集合；第5c圖、第5d圖是 定義出追蹤用的兩個3*3遮罩結構元素。利用第一個遮罩結構以 右到左、下到上的順序掃瞄到的第一點P，由第二點開始利用第二 個遮罩結構依顧終㈣線的所有像素點座標並作記錄，其結 果如第5b圖。 而整個曲線追蹤法的程序可分為以下的三個步驟： ⑴以第-個遮罩結構掃_第一點p為出發，再重複以第_個遮 .M294069 罩結構依著A、B、C、D順序找到第二點，記錄後並以〇取代中 心點。 (2)由於曲線的結構是八連通的關係，因此用第二個遮罩結構，依 著A、B、C、D、E、F、G、H的順序搜尋到下一點，記錄其座 標並以0取代中心點，再重複步驟(2) ⑶當A=B=C=D=E=F=G=H=0時，代表中心點是曲線的終點。因 此在記錄其座標後也就是整個曲線追蹤完成。M294069 VIII. New description: [New technical field] This case is an improvement of the type of palm-age system, and the basic characteristics of the Cellular Neural Netw〇rks (hereinafter referred to as (10)) are based on · Image pre-processing as a palm-shaped identification system. By inputting a palm-shaped image that you want to identify, after processing through - φ, the image is converted to produce meaningful values, and these values are compared with the data of the database to obtain The result of misidentification. The identification system consists of three parts: the first part is the palm-shaped image pre-processing and contouring; the second part is the f-shaped feature closing; the second part is the palm-shaped feature point comparison. Identification. [Prior Art] In 1988, a new information processing system called a cellular neural network was proposed by l〇· hua〃L·Yang, which has a local connection. The special |± 'this can easily achieve a single layer or multi-layer _ road structure, because _ 卩 纟, 纟 quot 转 转 转 所 所 所 所 所 所 所 所 所 所 所 所 所 所 所 所 所 所 所 所 所 所 所 所 影像 细胞 细胞 细胞 细胞 细胞 细胞 细胞Neural networks have also developed other applications, including signal processing, solving secret problems, speed detection of heterogeneous objects, and implementation of very large integrated circuits. The inspiration for the important technology of the genius left network 疋 artificial wisdom field towel is: - The scholastic ' hopes to build a computer model that can actively, learn, and think actively by simulating the structure of the human brain. In the recent years, neural networks have been extensively studied in signal processing. 5 M294069 Examples: filtering, signal detection, system identification, In terms of parameter estimation, signal compression, etc., the processed signals include video, audio, video, communication, etc., while Cellular Neural Networks (CM) is used by LEON 0·CHUA. The proposed architecture, and the CNN Universal Machine based on its theory, has the advantage of low cost 'low power, etc., which is one faster than digital signal processing (1)3{)). 〇 times the super computer, suitable for pure processing. Based on such superior hiding, the Cellular Neural Network (CNN) theory is used to develop the palm recognition system. [New content] This case is a kind of palm-shaped identification, which uses the basic characteristics of Cellular Neural Netw〇rks (hereinafter referred to as (10)) as the image front of the palm-shaped identification system. deal with. By inputting a palm-shaped image that wants to identify the identity, after a series of processing, the image is converted to produce a meaningful value, and then these values are compared with the data of the (four) library, from the middle The result of the identification. &# This identification system is divided into three parts: · Processing to the palm-shaped contoured palm image pre-processing system. 2. The palm-shaped feature of the dry shape is taken off and the processing system is closed. In turn, the software that uses the Matlab and the network architecture as the main axis of the case is created, and the processing of the palm-shaped picture is carried out with the negative of the cellular nervous system, and the M294069 is simulated and verified. [Embodiment] Please refer to the diagram of this case _ flow chart, σ diagram shows the palm shape recognition processing for this case /, the money is divided into three parts: 2. Frequency 辄 形 前 前 纽 纽 纽 纽 。 。. Therefore, the palm-shaped feature of the feature point is taken off and the processing system is closed. The shoulders of the shame and the touch of the palm of the hand to identify the thief system. (4) \, °, 111 This must first pass through the palm-shaped image pre-processing secret pre-set = ^ 仏 ^ program ' respectively grayscale processing, image binarization, image,, ^ and for the palm shape after marginalization Contour Nat is difficult to do the basis for comparison. [Pre-processing part of the palm-shaped image pre-processing system] The part of the gray-scale processing is to convert the color input palm-shaped picture into a gray-scale form by using the equation [as shown in Annex 1]. On the side of the shirt image-valued part, the fine-grained grayscale image is distributed as a pixel value and calculated as shown in Fig. 2. The horizontal axis of the second figure represents the grayscale value; the vertical axis represents the record of the grayscale value. . Most of the pixels of the towel are clothed on both sides, so it is very clear that there is a peak ship field of the secret duck, which has a low background (with a grayscale value of 70 or less) and a high grayscale value. The hand (the grayscale value is 14〇 or more), so the range of the grayscale value 70~140 is the part of the palm edge. Therefore, the embodiment of the present invention selects 70 as the critical value (Z) to set the image for binarization. And the grayscale image is processed by binarization, in which the grayscale value is greater than 70, it tends to white (hand), otherwise M294069 tends to black (background), and then the cell neural network - threshold (CNN-THRESHOLD) The image is converted to a black and white picture with parameters as shown in Figure 3 and the equation in Annex 2. After the grayscale image is binarized, the image has become black and white. The next step is to use the concept of edge detection to remove the black and white intersections in the binarized image results. And this junction is the palm shape you want to get. Edge detection, as the name suggests, is to depict the edge of the object and present it on the image. Here, the method of the cellular neural network _ gray edge (CNN_EDGEGRAY) is used to obtain the palm-shaped contour. The parameters are as shown in Figure 4 and Annex 3. The equation is shown. During the image processing process, the changes can be observed from the edge-detected curve to find the characteristics of the curve. The purpose of the curve tracing method is to find out the course of the curve completely, and to record its coordinates and number one by one. The subsequent feature extraction is very important information. In the curve tracking method, since the thumb of the palm shape is on the lower right side, it is possible to set the point to be scanned from the bottom right corner, so the scanning directions of the two mask structures are designed from right to left. Down to the top. Cong ^^ Yi Huifan, Figure 5a represents a set of all points of a curve; Figure 5c, Figure 5d are two 3*3 mask elements that define the tracking. Using the first mask structure to scan the first point P in right-to-left, bottom-up order, starting from the second point, using the second mask structure to follow all the pixel coordinates of the final (four) line Record, the result is shown in Figure 5b. The whole curve tracking method can be divided into the following three steps: (1) Starting with the first mask structure sweep _ first point p, and repeating the _th cover. M294069 cover structure according to A, B, C, D order to find the second point, after recording and replace the center point with 〇. (2) Since the structure of the curve is an eight-connected relationship, the second mask structure is used to search for the next point in the order of A, B, C, D, E, F, G, and H, and the coordinates are recorded. Replace the center point with 0, and repeat step (2). (3) When A=B=C=D=E=F=G=H=0, the center point is the end point of the curve. Therefore, after the coordinates are recorded, the entire curve is tracked.
〔掌形特徵點擷取處理系統之特徵點的萃取部份〕 經由前置處理系統取得掌形輪廓後，接著根據曲線追縱法， 將輪廓曲線的每個座標記錄起來^鱗各點座標變化關係中可 發現’當y S標的變化量由遞減轉為遞增時，其對應的點恰為指 尖；同理，y座標之變化#由遞增轉域減時，該點為指縫，根 據此賴便定義了舰卜2、3、4點。糾，由於以點無法以 座標變化絲狀，目此㈣戰物之輪廊 並取得交會點；陳，細是由第3點錄點之間轉段延^ 輪廓並取得交會點L發現這邊所定義出的指縫特徵H 3 4點並不_鱗根手指料它只缺義純綱轉的最左邊 的瞬間變化點，因此*適合於右邊的指指縫特徵 此問題’贱4職職右尋找#y向量_變化的―料2 侧手指的指縫，因此另外定義了特徵7、8、9點來輔助指縫義。 另外，發現第1點並不能夠與第7點做為食指的 決此問題’必須另外定義一點可與第7點做為食指指縫，此二發 9 M294069 削、姆純_斜轉跟食如_斜轉只差—個貞號，兩者 的料值是幾乎相同的。所以可以姻此—特性來找到特徵第1〇 點。取後’ _關特徵點做順序的_如細），以便往後的 ^擷取目此在賴點的部份取了 ι_特徵點，當然於本案實 ⑯上婦徵點亦可取敢複數㈣齡，但本案以 特徵點為例說明。 根，特徵闕相對位置關係，可以順利定義出手指寬度、面 •積手掌見度等幾何掌形特徵，於本案實施例中該掌形比對係以 特^為基準，擷取掌形19項幾何特徵作為辨識依據第i項至第 4項旱形特徵為食指至小姆指個別的指縫寬度；第5項至第8項掌形 概為食指至小姆指個別由指縫向上取特定高度的指寬；第9項至 第項為食指至小姆制指定面積；㈣項至㈣項為食指至小 姆I曰取其“縫的個別斜率；第1?項以掌形特徵點當中特定三點所 應，角ϋ周長大小來代表一個掌形的大小；第18項為大姆指指縫 第19項疋。十算構成整體掌开义的總像素〔_咖〕值。其掘 取方法如下： ^ 1·指縫距離 由^母個人的手指胖痩不一，而且每根手指指縫距離變化也沒 有一定的規律性，因此將五根手指個獅指縫距離來做為辨識特 徵之- ’如第7圖中白色虛線線段所*。其中姆指指縫距離是由^ 表示；食指指縫距^是由口表示；中指指縫距離是由碎表示無 名指指縫距離是由玩表示；小指指縫距離是由石表示，因此第； M294069 圖中白色虛線則代表手指指縫距離的特徵 2_指定手指寬度 當我們取完錄手指_鱗寬後，魏就聽定為手指的 特徵，因此除了擷取指縫寬度之外，還取了手指特定高度的指寬， 是從各個手指的指縫點向上偏移3〇像素〔㈣〕後的指寬，因為 手指指寬會因為不_高度㈣所不同，而且也不-定是對指縫 成-定的比率，因此’可以將其納入手指的特徵辨識當中。但這 項特徵並沒有取每-根手指頭，只雜取除了大姆指之外的四根 手指。如第8圖中所示，例如：食指指縫點—及點八向上偏移％ 像素〔pixels〕後的相對距離則定義為忑。 3 ·指縫斜率[Extraction of feature points of the palm-shaped feature point extraction processing system] After obtaining the palm-shaped contour through the pre-processing system, each coordinate of the contour curve is recorded according to the curve tracking method. In the relationship, it can be found that when the change amount of the y S standard is changed from decreasing to increasing, the corresponding point is just the fingertip; similarly, the change of the y coordinate is reduced by the incremental transition, and the point is the finger seam, according to this Lai defined the ship's 2, 3, and 4 points. Correction, because the point can not be changed by the coordinates of the filament, the purpose of this (four) warships and the landing point of the battle; Chen, the fine is from the third point between the record point extension ^ contour and get the intersection point L found this side The defined finger joint feature H 3 4 points is not _ scale root finger material, it only lacks the leftmost moment of change of the pure direction of the transition, so * suitable for the right finger joint feature of this problem '贱 4 job right Look for the #y vector_changed material 2 side finger finger joint, so the characteristics 7, 8, 9 points are additionally defined to assist the finger seam meaning. In addition, it is found that the first point is not able to use the seventh point as the index finger to solve this problem. 'An additional point must be defined with the seventh point as the index finger joint. The second hair 9 M294069 cut, m pure _ oblique turn to eat If the _ oblique turn is only a nickname, the material values of the two are almost the same. So you can use this feature to find the first point of the feature. After taking the ' _ off feature points to make the order _ as fine, so that the next ^ 撷 撷 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目 目(4) Age, but the case is characterized by a feature point. Root, feature 阙 relative position relationship, can smoothly define the geometric palm shape such as finger width, face and hand palm visibility. In the case of this case, the palm shape comparison is based on the special ^, taking the palm shape 19 items The geometric features are used as the identification basis. The dry shape of the i-th to the fourth item is the individual finger-slit width of the index finger to the small finger; the palms of the fifth to the eighth item are the index finger to the small finger. Height refers to the width; items 9 to 1 are the designated areas of the index finger to the Xiaomu system; (4) to (4) are the individual slopes of the “index” from the index finger to Xiaomu I; the first item is in the palm shape point. The specific three points should be that the size of the circumference of the corners represents the size of a palm shape; the 18th item is the 19th item of the thumb and the finger. The tenth count constitutes the total pixel [_coffee] value of the whole palm. The method of excavation is as follows: ^ 1·The distance between the finger and the finger of the mother is different, and there is no regularity in the change of the distance between the fingers and fingers. Therefore, the distance between the five fingers and the lion is used as the distance. Identify the characteristics - 'as in the white dotted line segment in Figure 7. * The thumb finger distance is ^ Representation; the index finger refers to the seam distance ^ is represented by the mouth; the middle finger refers to the seam distance is broken by the ring finger, the finger gap distance is represented by play; the little finger finger gap distance is represented by stone, so the first; M294069 the white dotted line represents the finger The characteristics of the finger gap distance 2_Specify the finger width. When we take the finger _ scale width, Wei will be the finger feature, so in addition to the finger width, the finger width of the finger is taken. It is the finger width that is offset from the finger point of each finger by 3 pixels [(4)], because the finger width is different because it is not the height (four), and it is not the ratio of the finger to the finger. Therefore, it can be included in the feature recognition of the finger. However, this feature does not take every finger, but only four fingers except the thumb. As shown in Figure 8, for example: index finger The finger-point offset—and the point eight offset upward. The relative distance after the pixel [pixels] is defined as 忑.
由於發現每個人手掌的指縫相對位置的變化方向並不相同， 而且只要錢齡置的不相對的解也會麵改變，因此可 以做為手指辨識的特徵之…但這—掃徵還是擷取姆指以外的 :指做為指縫斜率的特徵’將四指的個別的指縫點取斜率，例如： =9圖中可知道食指、中指、無名指、小指各個手指指縫的走向， *此可以根據此走絲縱各個指縫斜料正貞值。糾，可以 取得各個 指縫點的座標來算出指縫斜率的數值，其算式為 少2-％ X2 ^Xl ===走向都是大同小異，率的值就是作 4·手指指定面積 11 M294069 的縫的長度以及特定的指宽來表示手指 義手指的㈣的特徵’因此在這—部份定 的—維向置特徵：面積。此面積並不是指整個 :而是由指縫至特定高度之間所構成的面積：面 徵是不考慮高度… 特定ΐ度=指咖強烈變數。因此在計算手指面積則只取It is found that the direction of the relative position of the finger joints of each person's palm is not the same, and as long as the non-relative solution of the age of the money is changed, it can be used as a feature of finger recognition... but this is a sweep or a capture. Other than the thumb: refers to the feature of the slope of the finger joint. 'The slope of the individual finger points of the four fingers is taken. For example: =9 The direction of the fingers of the index finger, the middle finger, the ring finger and the little finger can be known. *This According to this, the longitudinal value of each finger joint oblique material can be determined. Correction, you can obtain the coordinates of each finger point to calculate the value of the finger seam slope, the formula is less 2-% X2 ^Xl === the trend is much the same, the value of the rate is the 4 finger specified area 11 M294069 seam The length and the specific width of the finger indicate the feature of the (four) of the finger's finger. Therefore, in this case, the dimension is dimensioned. This area does not refer to the whole: it is the area formed by the finger to the specific height: the face is regardless of the height... The specific twist = the strong variable of the coffee. Therefore, only the finger area is calculated.
特徵輕_ ’物+細化量大樹面積，影響到 1確度。而手細積求算麵首先侧顯㈣每 ^ΓΓε，t將各手指指縫點互相連接，而高度則是言博到 線，如二=:這個別的手指一 'τ顺騎崎當中像素〔pixels〕值為單位計 二=對：Γ當概色贿，侧面積算出並存 5·手掌大小 =於每個人的手掌大小不相同，而在擷取影像的時候也會因 :、、、手掌大小的關係，所拍攝到的手部範圍也不—定，因此手掌的 大J也可以列為掌形的特徵之一。由於手腕的位置無法精確定位 出來以致於無法叶算出整個手掌的面積。因此利用特徵點3小 、乾圍所不。料，可峨曲線追蹤法中記錄掌形每-點座標後 發現手掌的大小會影_記錄_數，因此將此記錄的點數也來 表示手掌大小的特徵之一。 12 M294069 - (掌形特徵點比對辨識處理系統之掌形比對部份） 本案創作特徵係以特徵點的比對為辨識的依據，並採用歐幾里 得距離(Euclidean Distance)計算差異值的方法，而比對的流程 如第12圖，來驗證前面所描述的掌型特徵摘取方法，以及所提出 的演算法則。 本案所使用的硬體設備包含lntel Pentium4 CPU 2 66GHz , 512MB RAM個人電腦以及EPSON PERFECTION 1250掃瞄器。作業系 •、统為WIND0WS 2000，辨識系統中主要演算法是用MATLAB 6.5版， 而在系統中之比對及介面部份演算法則是採用VisualThe characteristic light _ 'object + refine the amount of large tree area, affecting 1 accuracy. And the hand thinning calculation face first side display (four) each ^ ΓΓ ε, t will be the finger joint points of each other, and the height is the words to the line, such as two =: this other finger a 'τ 顺 崎 崎 像素 像素[pixels] is the unit of the two = pair: when the color is bribed, the side area is calculated and coexisted 5. The size of the palm = the size of each person's palm is different, and when the image is captured, it is also due to:,,, palm The size of the relationship, the range of the hand is not fixed, so the big J of the palm can also be listed as one of the characteristics of the palm. Since the position of the wrist cannot be accurately positioned so that the area of the entire palm cannot be calculated. Therefore, the feature points are small and dry. In the 峨 curve tracking method, after the palm-shaped coordinates of each point are recorded, it is found that the size of the palm will be _record_number, so the number of points recorded is also one of the characteristics of the palm size. 12 M294069 - (The palm-shaped comparison part of the palm-shaped feature point comparison processing system) The creative feature of this case is based on the comparison of feature points, and the difference value is calculated by Euclidean Distance. The method of comparison, as shown in Fig. 12, verifies the palm-shaped feature extraction method described above, and the proposed algorithm. The hardware used in this case consists of an Intel Pentium 4 CPU 2 66 GHz, a 512 MB RAM personal computer and an EPSON PERFECTION 1250 scanner. The operating system is the WIND0WS 2000. The main algorithm in the identification system is MATLAB version 6.5, and the comparison and interface algorithm in the system is based on Visual.
Basic 6. 0 版來元成介面視窗化的實現。在特徵比對測試和建立資料庫方 面’本案實施例實際測試拍攝了 1〇個人的手部影像，並且每個人 拍攝了5張手部影像，共5〇張1_1〇〇(>[象素*像素），以提供系統特 徵比對之測試。其中每個人的手部影像是分別在不同時間及同背 厅、下所拍攝完成的。 • 首先，資料庫建檔時，先取得每個人的前3張手部影像，經 像處理後，每張手部影像所得的19個特徵㈣張手部影像， 共57個特徵值(3組），然後把這3組特徵值取平均，並存入資料 庫建檔，來當作本案要特徵比對之依據。 為了順利進彳了特徵比對卫作，本案實施例制了歐幾里德距 離計算差異值的方法，敘述如下： 歐幾里德距離：歐幾里德距離在距離測量方面是最常用來計算差 異值的方法。此種相似程度的運算主要是直接計算兩向 13 M294069 3的=算後所得到的值越小代表兩向量差異 里越小，料公式如附件4所示。 :中’:稱為歐幾里得距離、為輪入的特徵向量的第i個 特徵值，苹為資料庫儲存的特徵向 出使用去於入〜m 里的弟1個特徵值’因此計算 出使用者輸入旱形特徵及資料庫内特徵比對後的誤差值。 本案提出了-套掌形特徵之身份辨識系統 =職編輪输成。 輪入的旱形影像當中取得19筆特徵向量作為整張影像的資訊因 此本案實施例使用歐幾里德距離之差異值來進行特徵向量的比 對，此方法不僅可以減少比對的資料量也可以維持不錯的辨識 率，而且確實具有辯識度的正確性及可行性。 本I]作系、、4係&廣泛應用在家庭保全、門禁身份辯識、簽到· · 等用途上。因此，本賴作「掌形觸系制改良」確實具有精 確、迅速的實収益與優異的產糊雜，且其所具有的創作特 徵與整體技術改良運用，皆為—般所無相同的^計運用，本案確 付新型專利之創作申請要件者。 【圖式簡單說明】 第1圖：本創作之掌形辨識處理流程示意圖。 第2圖：本創作之影像灰階值分佈統計圖。 第3圖：本創作之細胞神經系統網路一門檻(CNiTHRESH〇ld)參數 示意圖表。 14 M294069 第4圖·本創作之細胞神經系統網路_灰階邊緣(CNN_EDGEGRAY) 參數示意圖表。 第5a圖·本_之曲線追蹤法模版示賴（原始曲線）。 第5b圖·本創作之曲線追蹤法模版示意圖（曲線追蹤結果）。 乐D圃 第7圖 第8圖 第Q圖 第5c圖·本創作之曲線追蹤法模版示意圖（第一個遮罩結構）。 第5d圖·本創作之曲線追蹤法模版示意圖（第二個遮罩結構）。 本創作之特徵點擷取示意圖。 本創作之掌形特徵—指縫距離示意圖。 本創作之掌形特徵-指定手指寬度示意圖。 本創作之掌形特徵—指縫斜率示意圖。 第10圖：本創作之掌形特徵—手指指定面積示意圖 第11圖：本創作之掌形特徵-手掌大小示意圖。 第12圖：本創作之比對流程示意圖表。 附件1 :本創作實施例之灰階處理部份所利用的方程式。 附件2 :本創作實施例以細胞神經系統網路—門檻 (CNN一THRESHOLD)將影像轉為黑白圖片參數之方程 式。 附件3 :本創作實施例以細胞神經系統網路—灰階邊緣 (CNN一EDGEGRAY)的方法取得掌形輪廓參數之方程 式。 附件4:本創作實施例以歐幾里德距離測量來計算差異值的方 程式。 15 M294069 【主要元件符號說明】 1影像前置處理系統 11欲辨識掌形圖片輸入 12影像灰階化處理 13影像二值化處理 14影像邊緣化處理 2特徵點擷取轉換成數值 3掌形特徵點比對 5比對結果 4特徵資料庫 16Basic 6. 0 version of the realization of the interface into the window. In the feature comparison test and the establishment of the database, the actual test of the example of the case shot a personal hand image, and each person took 5 hand images, a total of 5 1 1_1 〇〇 (> [pixels *Pixel) to provide a test for system feature comparisons. Each of them's hand images were taken at different times and in the same hall. • First, when the database is filed, first obtain the first 3 hand images of each person, and after the image processing, 19 features (4) hand images obtained from each hand image, a total of 57 feature values (3 groups) Then, the three sets of eigenvalues are averaged and stored in the database to be used as the basis for the feature comparison of the case. In order to smoothly enter the feature comparison, the method of this case is to calculate the difference value of Euclidean distance calculation, which is described as follows: Euclidean distance: Euclidean distance is the most commonly used in distance measurement. The method of difference value. The operation of this similarity is mainly to directly calculate the two-direction 13 M294069 3 = the smaller the value obtained, the smaller the difference between the two vectors, the formula is shown in Annex 4. : Medium ': the i-th eigenvalue of the feature vector called the Euclidean distance, which is the wheeled feature, and the feature value stored in the database is used to enter the eigenvalue of the younger one in the ~m. The error value after the user inputs the dry shape feature and the feature comparison in the database. In this case, the identification system of the palm-shaped feature is proposed. In the rounded image, 19 feature vectors are obtained as the information of the whole image. Therefore, in this embodiment, the difference value of the Euclidean distance is used to compare the feature vectors. This method can not only reduce the amount of data of the comparison. Can maintain a good recognition rate, and indeed has the correctness and feasibility of the identification. This I] system, 4 series & is widely used in family security, access control identification, sign-in · · and other purposes. Therefore, this "realization of the palm-shaped system" does have accurate and rapid real income and excellent production, and its creative features and overall technical improvement and application are all the same. The application of the application, the case to ensure the creation of new patents. [Simple description of the diagram] Figure 1: Schematic diagram of the palm shape recognition process of this creation. Figure 2: Statistical map of the grayscale value distribution of the image of this creation. Figure 3: Schematic diagram of the parameters of the cell neural network of the creation (CNiTHRESH〇ld). 14 M294069 Figure 4 • The cell neural network _ grayscale edge (CNN_EDGEGRAY) parameter schematic table. Fig. 5a·The curve tracking method of this _ is based on the original curve. Figure 5b. Schematic diagram of the curve tracking method template for this creation (curve tracking results). Le D圃 Fig. 7 Fig. 8 Fig. Q Fig. 5c. Schematic diagram of the curve tracing template of the creation (first mask structure). Figure 5d. Schematic diagram of the curve tracking method template of this creation (second mask structure). A schematic diagram of the feature points of this creation. The palm shape of this creation—the schematic diagram of the finger distance. The palm shape feature of this creation - specifies the finger width diagram. The palm shape of this creation - a schematic diagram of the slope of the finger. Figure 10: The palm shape of the creation - a schematic diagram of the designated area of the finger Figure 11: The palm shape of the creation - the size of the palm. Figure 12: Schematic diagram of the comparison process of this creation. Annex 1: Equations utilized in the grayscale processing portion of the present creative embodiment. Annex 2: This creation example uses the cellular neural network - threshold (CNN-THRESHOLD) to convert the image into a black and white image parameter equation. Annex 3: This authoring example obtains the equation for the palm-shaped contour parameter by the method of the cellular nervous system network-the gray-scale edge (CNN-EDGEGRAY). Annex 4: This creation example uses a Euclidean distance measurement to calculate the difference value. 15 M294069 [Description of main component symbols] 1 Image pre-processing system 11 To identify palm-shaped image input 12 Image grayscale processing 13 Image binarization processing 14 Image marginalization processing 2 Feature point extraction into numerical 3 palm-shaped features Point comparison 5 comparison result 4 feature database 16