1291668 九、發明說明： 【务明所屬之技術領域】 本發明係有關於一種圖形比對搜尋方法，詳言之，係_ 種利甩梯度方向編碼技術及二階段之圖形比對搜尋方法。 【先前技術】 … 在曰#生活及工業檢測相關應用上，存在著許多需要分 類或辨識比對的動作，如辨識物體種類或其差異性、印刷 電路板（簡稱PCB)的電路檢測等。而這些動作所保有一個 特點疋重複性南，也因為這樣的重複性，我們能簡化這樣 勺動作方式，疋義一個標準樣版，重複對其資料群作一個 搜尋及比對的工作。 在電腦視覺、影像處理及圖形識別的領域内，由於影像 疋個光強度函數及反射函數兩個量的摺積，同時，具有 運算資料龐大的特性。我們對一個圖形作一所謂的前處 理、後處理及相關的圖形強化動作，這些動作的處理，無 非是要拿掉一些不合用、不相干的雜訊資料，只將相關合 用的資料保留下來。而這樣的流程，最後的目的，是為f 提供我們作一個物體識別、比對及分類。 先前技術之圖形比對搜尋方法，大略可分為下列幾種。 1·樣版比對法，其缺點在於進行全域搜尋時，由於影像函 數有著巨量資料，故比對識別時間將拉長。2•梯度下降 法，其固然有一定的效率，但無法確認可找到全域裡和目 標樣版最為匹配且變異最小的區塊樣本。3·正規化相關係 數法，在照明光源均勻分佈於整個全域區域的情況下，能 107494.doc 1291668 $免錯誤的匹配發生。不過，當部分物體被遮蔽、部分陰 =和背景產生變化、目標變形和合併上述情形同時存在 時，即無法解決這些情況下所發生的錯誤辨識。另外，當 目標物體旋轉時，自相關係數法只能偵測出大約5至1〇^ 的’又化情形，若大於此，則會發生辨識錯誤或偵測遺漏情 形。再者，計算時間的問題，係自相關係數法的嚴重缺 例如當進行印刷電路板上之電路之線寬檢測時，若運 算時間過久，則無法進行即時檢測。 因此’有必要提供一創新且富有進步性的圖形比對搜尋 方法’以解诀上述問題。 【發明内容】 不發明之目的係在於提供―種圖形比對搜尋方法，該方 =包括：⑷擷取—原始影像’該原始影像區分為複數個第 像素，⑻提供一第一樣版影像，用以與該原始影像比 對，該第-樣版影像區分為複數個第二像素，該第二像素 之^寸相同於該第-像素之尺寸；（c)將該原始影像及該第 一樣版影像進行灰階梯度編碼’並取得複數個第 度方向碼及複數個第二像♦ I婦 傻夸说洚古A 、又方向碼，（d)比對該等第一 像素梯度方向碼及該等第二像素梯度方向碼 數個該原始影像之不同位 开取侍啜 定義該原始影像與該第：版樣版影… — 樣版影像相似度最高之區域為— 弟-目標區域，.⑷定義一第二目標區域，該第 係由該第一目標區域外 “域 1擴浪口Ρ为區域，以大於兮 目“域之乾圍’胃第二目標區域區分為複數個第三像 107494.doc 1291668 二樣版影像區分為複數個 素定義為-第二樣版影像，該第四像辛之:：，四像 三像素之尺寸，蔣兮楚- ”之尺寸相同於該第 、以弟—目標區域及該第—媒 灰階梯度編碼，並取亥弟—樣版影像進行 個第四像素梯产方^ @弟二像素梯度方向碼及複數 “呆又方向碼，·及（f)比對該等第三 碼及該等第四像辛描 ”梯度方向 挪〜 I素梯度方向碼，計算取得複數個該第二曰 仏區域之不同位置盥 一目 一目俨F…〃 -樣版衫像之相似度，定義該第 -目W域與該第二樣版影像相似度最高 影像。 X巧目標 本表明圖料對搜尋方法，㈣用灰階梯度 = 寺徵的基礎，並採用兩階段搜尋比對方式，精確: 哥 目似度最大之影像物體。利用灰階梯度具有以下之 優點’當搜尋的目標物體與鄰域之灰階度變化差異大時， 灰階梯度方向反映在物體相對影像之變化關係不大，Z能 完整保留原目標物體的物理特性，可作為一個良好 2 辨識基礎。 ’& 再者，兩階段搜尋的方式，於第一階段先搜尋得該第一 目標區域之大約位置，再由該第一目標區域外圍向外擴張 部分區域，以大於該第一目標區域之範圍，並定義一第二 目標區域；而在第二階段搜尋中，係於該第二目標區域 中’以更小尺寸之該等第三像素及該等第四像素進行比對 更精細之比對，藉此更精確地找尋出該目標影像。 【實施方式】 參考圖1至圖7，其顯示本發明圖形比對搜尋方法之示音 107494.doc 1291668 圖。配合麥考圖1及圖7，首先，如步驟7〇1所示，利用一 影像擷取裝置（圖未示出）擷取一原始影像10(例如擷取自印 刷電路板（PCB)之電路或是液晶顯示器面板之電晶體等）。 配合蒼考圖2及圖7，如步驟7〇2所示，建立原始影像資 訊’该原始影像1〇具有複數個相同圖樣1〇1，及χι乘¥1個 第一像素102 ’且取得一原始灰階值g(x1，y1)，該原始影 像10具有9x9個第一像素1〇2，則χι=9，γι=9，共有81個第 . 一像素102,每一第一像素1〇2均具有一介於〇至255之灰 階值。 接著，如步驟703所示，提供一預先設定之第一樣版影 像20，用以與该原始影像1〇比對，該第一樣版影像別具有 一叹定圖樣201，該第一樣版影像2〇具有又2乘丫2個第二像 素202，且取得一原始灰階值G(X2，Y2)，該第一樣版影像 20具有3x3第二像素2〇2，則χ2 = 3，γ2==3，共有$個第二像 素202，每一個第二像素2〇2均具有一介於❹至]“之灰階 > 值。參考步驟7〇4,設定該等第二像素2〇2之尺寸相同於該 等第一像素1 02之尺寸。 接著如步驟705及步驟706所示，將該等第一像素1〇2 及该等第二像素202進行灰階梯度編碼，該灰階梯度編碼 係將每一第一像素102及該等第二像素2〇2之水平方向之灰 階值梯度及垂直方向之灰階值梯度設為心及办，取一反正 刀函數梯度方向角度則為沒=tan-〗（dy/dx)，該梯度方向 角度Θ係介於〇至2万間，再將該梯度方向角度θ分割為複 數個等份，使每一梯度方向角度0對應至其中之一個等 107494.doc 1291668 份’若其梯度方向大小（實務上，為一遮罩摺積運算，依 權重求得水平及垂直各自梯度大小，[如^ +办2]!/2，並計算 求得該處整體梯度大小）係大於一預定閥值時，則計算得 到一對應之梯度方向碼C(X，Y);反之，若梯度方向大小係 小於一預定閥值時，則表示無謂之微量變化，以一設定值 處理，不具特別之物理意義。依據上述之灰階梯度:碼方 法，可計算得複數個第-像素1〇2之梯度方向碼及複數個 第一像素202之梯度方向碼。 配合參考圖3及圖7，如步驟7〇7所示，比對該等第 素⑽之梯度方向碼及該等第二像素加之梯度方向碼 异^得複數個該原始影像1G之不同位置與該第—樣 =目似度：定義該原始影像1〇與該第一樣版影像2〇相似 度最间之區域為一第一目標區域％。 、、在該實施例中’在此對搜尋過程中，用「二分樹搜， 法」的概念，並不逐一去縣々 + 費，可分為兩個動作，第節點’徒增時間的耗 「根節點」比較的方式來二^ 路徑，—次過^半^^子樹或右子樹被選擇的 人也慮+的即點數量；第二，1 > 點的比較，大於該子節職右子樹走，拾棄 2即 樣過濾的動作反覆被進行時， 八 树，*這 點’其執行效率為0(Iog Ν), ^ /找出我們關切的節 以此節點為根節點作一個中戽此時，並不作-停止’而 起始’逐-拜訪其他節點、父r :動作’由樹左下葉節點 重複此拜訪動作，直 ：：：右節點或右葉節點’ 凡其根郎點展開的右子樹；細 107494.doc 1291668 過這個搜尋方4丰 ― 區塊及相對座到吻合相似度的範圍 +應的搜尋鳥座標位置’再作—自 區塊，便能各Λ· ^ 〃足寸比對 測圖…: 比對區塊數量，而不用對其檢 、回/ I >數或收斂因子作一窮舉搜尋，可以 旦 降低圖形比對的巨量曾士卜吊大里 入▲ 寸#使比對工作耗費時間縮短。 —配口苓考圖4及圖7，如步驟7〇8及步驟7〇9所示，定義一 第目‘區域40 ’該第二目標區域4〇由該第一目標區〇 外陶擴張部分區域，以大於該第一目標區域一 圍。/亥第二目標區域40區分為複數個第三像素401，該第 樣版’ν像20(圖2)則區分為複數個第四像素5〇1，而另定 義為一第二樣版影像50，該第二目標區域40之該等第三像 素4〇1之尺寸大小與該第二樣版影像5〇之該等第西像素^⑴ 之尺寸大小相同。如步驟71〇及步驟711所示，進行灰階梯 度編碼，計算取得每一第三像素4〇1及每一第四像素501之 水平方向及垂直方向之灰階梯度，再利用每一第三像素 4〇1及每一第四像素5〇1之水平方向及垂直方向之灰階梯 度汁异彳于该第二目標區域40之該等第三像素4〇1之梯度 方向碼及該第二樣版影像50之該等第四像素5〇1之梯度方 向碼。 该第二目標區域4〇之該等第三像素4〇1及該第二樣版影 像50之該等第四像素5〇1，係小於該原始影像1〇之該等第 一像素102及該第一樣版影像20之該第一像素2〇2。在該實 &例中’該第二目標區域4〇及該第二樣版影像5〇之切割 值’係為該原始影像1 0及該第一樣版影像2〇之切割值之二 107494.doc • 11 - 1291668 借 ’刀、即孩第四俊音$ n〗 9個，該第三像f4()1數1^，而該第二像素2〇1共有 如斗― 常401數置為該第一像素101數量之四俨。1291668 IX. Description of the invention: [Technical field to which the invention pertains] The present invention relates to a method for searching for a graphical comparison, in particular, a technique for grading a gradient direction and a method for searching for a two-stage graphical comparison. [Prior Art] ... In the application of 生活#Life and industrial inspection, there are many actions that require classification or identification comparison, such as identifying the type of object or its difference, circuit detection of printed circuit board (PCB). These actions retain a characteristic, repetitive south, and because of this repetitiveness, we can simplify the way of scooping, deprecating a standard pattern, and repeating a search and comparison of its data groups. In the field of computer vision, image processing and pattern recognition, due to the two products of the light intensity function and the reflection function, the data has a large amount of computational data. We do a so-called pre-processing, post-processing and related graphic enhancement actions on a graphic. The processing of these actions is to remove some unsuitable and irrelevant noise data and only retain the relevant data. The final purpose of such a process is to provide us with an object identification, comparison and classification. The prior art graphics comparison search methods can be roughly classified into the following types. 1·Pattern comparison method, the disadvantage is that when the global search is performed, since the image function has a huge amount of data, the comparison recognition time will be lengthened. 2 • Gradient descent method, although it has certain efficiency, it can not be confirmed that the block sample with the most matching and the smallest variation in the whole domain and the target sample can be found. 3. The normalized phase relationship method, in the case where the illumination source is evenly distributed throughout the entire area, can be matched with the error-free matching of 107494.doc 1291668 $. However, when part of the object is obscured, part of the negative = and the background changes, the target is deformed, and the above situation exists simultaneously, the misidentification occurring in these cases cannot be solved. In addition, when the target object is rotated, the autocorrelation coefficient method can only detect the recurrence of about 5 to 1 〇 ^. If it is larger than this, a recognition error or a missing omission will occur. Furthermore, the problem of calculating the time is a serious lack of the autocorrelation coefficient method. For example, when the line width detection of the circuit on the printed circuit board is performed, if the operation time is too long, the instant detection cannot be performed. Therefore, it is necessary to provide an innovative and progressive graphical comparison search method to solve the above problems. SUMMARY OF THE INVENTION The object of the invention is not to provide a method for searching for a graphic comparison, the method comprising: (4) capturing - the original image 'the original image is divided into a plurality of pixels, (8) providing a first version of the image, For comparing with the original image, the first template image is divided into a plurality of second pixels, the second pixel is the same size as the first pixel; (c) the original image and the first image The sample image is gray-graded coded' and obtains a plurality of first direction codes and a plurality of second images ♦ I idiots say 洚古 A and direction codes, and (d) compares the first pixel gradient direction codes And the second pixel gradient direction code number of the different positions of the original image open to define the original image and the first version of the sample image - the highest similarity of the pattern image is - the brother-target area, (4) defining a second target area, the first system is outside the first target area, and the domain 1 is expanded into a region, and the second target region of the stomach is divided into a plurality of thirds. Like the 107494.doc 1291668 two-page image is divided into The number of elements is defined as - the second version of the image, the fourth image is::, the size of the four-image three-pixel, Jiang Xiaochu - "the size is the same as the first, the younger - the target area and the first medium Gray step coding, and take the Haidi-sample image for a fourth pixel ladder production method @@弟二pixel gradient direction code and plural "stay and direction code, · and (f) than the third code and The fourth image is sinusoidal in the direction of the gradient, and the gradient direction code is calculated, and the plurality of different positions of the second region are calculated, and the similarity of the pattern is defined. - The visual field is the highest similarity image to the second sample image. The X-objective image indicates the image-to-search method, (4) the gray gradient = the basis of the temple sign, and uses a two-stage search comparison method, accurate: The most objective image object. The use of gray gradient has the following advantages: When the difference between the gray scale of the target object and the neighborhood is large, the gray gradient direction is reflected in the relative change of the object relative to the image, Z Can completely retain the physical characteristics of the original target object, As a good 2 identification basis. '& Again, the two-stage search method first searches for the approximate position of the first target area in the first stage, and then expands the partial area outward from the periphery of the first target area to Greater than the range of the first target area and defining a second target area; and in the second stage of searching, the third pixel and the fourth pixel are smaller in size in the second target area A finer alignment is performed, thereby finding the target image more accurately. [Embodiment] Referring to FIG. 1 to FIG. 7, the display of the graphical comparison search method of the present invention is shown in the figure 107494.doc 1291668. McCaw Figure 1 and Figure 7, first, as shown in step 7.1, an image capture device (not shown) is used to capture an original image 10 (such as a circuit taken from a printed circuit board (PCB) or It is a transistor of a liquid crystal display panel, etc.). In conjunction with FIG. 2 and FIG. 7, as shown in step 7〇2, the original image information 'the original image 1〇 has a plurality of identical patterns 1〇1, and χι times ¥1 first pixels 102' and obtains one The original grayscale value g(x1, y1), the original image 10 has 9x9 first pixels 1〇2, then χι=9, γι=9, a total of 81 first pixels 102, each first pixel 1〇 2 has a gray scale value between 〇 and 255. Next, as shown in step 703, a pre-set first version image 20 is provided for comparison with the original image, the first version image has an yaw pattern 201, the first version The image 2〇 has 2 times 2 second pixels 202, and an original gray scale value G(X2, Y2) is obtained. The first version image 20 has 3×3 second pixels 2〇2, then χ2=3, Γ2==3, there are a total of two second pixels 202, and each of the second pixels 2〇2 has a grayscale value of ❹ to ”. With reference to step 7〇4, the second pixels 2 are set. The size of 2 is the same as the size of the first pixels 102. Then, as shown in steps 705 and 706, the first pixels 1〇2 and the second pixels 202 are gray-graded, and the gray ladder is used. The degree coding system sets the grayscale value gradient in the horizontal direction and the grayscale value gradient in the vertical direction of each of the first pixels 102 and the second pixels 2〇2 as the heart and the center, and takes an inverse knife function gradient direction angle. For no = tan-〗 (dy/dx), the gradient direction angle Θ is between 〇 and 20,000, and the gradient direction angle θ is divided into a plurality of aliquots. , so that each gradient direction angle 0 corresponds to one of them, etc. 107494.doc 1291668 copies 'if the gradient direction size (actually, for a mask folding operation, according to the weight to find the horizontal and vertical respective gradient size, [such as ^ + 2]! / 2, and calculate the overall gradient size is greater than a predetermined threshold, then calculate a corresponding gradient direction code C (X, Y); conversely, if the gradient direction is When it is less than a predetermined threshold, it means that there is no unnecessary slight change, and it is processed by a set value, which has no special physical meaning. According to the gray step degree: code method, the gradient direction of the plurality of first-pixels 1〇2 can be calculated. a code and a plurality of gradient direction codes of the first pixels 202. Referring to FIG. 3 and FIG. 7, as shown in step 7〇7, the gradient direction codes of the first pixels (10) and the second pixels plus the gradient direction codes are used. The difference between the different positions of the original image 1G and the first sample = the degree of similarity: the area where the original image 1 〇 and the first version of the image 2 〇 similarity is the first target area % In this embodiment, 'search here' In the process of using the concept of "two-point tree search, method", I don't go to the county 々 + fee one by one. It can be divided into two actions. The node "the root node of the time-consuming increase" is used to compare the path. - The number of points that are selected by the second half ^^ subtree or the right subtree is also considered to be the number of points; the second, 1 > point comparison is greater than the subsection of the right subtree, and the 2 is discarded. When the filtered action is repeatedly performed, the eight tree, * this point' its execution efficiency is 0 (Iog Ν), ^ / find out the section we are concerned about, use this node as the root node for a lieutenant at this time, do not stop - stop 'And the start' - visit other nodes, parent r: action 'repeated this visit action by the left lower leaf node of the tree, straight ::: right node or right leaf node' right subtree whose roots are expanded; fine 107494. Doc 1291668 Through this search side 4 Feng - block and relative seat to the range of similarity + the search for the bird's coordinate position 're-doing - from the block, you can each Λ · ^ 〃 寸 inch comparison map...: Compare the number of blocks without an exhaustive search for their check, return / I > number or convergence factor, you can reduce the graph Compared to the huge amount of hanging Tali had Shibu into the # ▲ inch shorter than the time-consuming work. - Referring to Figure 4 and Figure 7, as shown in steps 7〇8 and 7〇9, defining an item 'area 40' the second target area 4〇 is expanded by the first target area The area is larger than the first target area. The second target area 40 is divided into a plurality of third pixels 401, and the first version 'ν image 20 (FIG. 2) is divided into a plurality of fourth pixels 5〇1, and is defined as a second pattern image. 50. The size of the third pixels 4〇1 of the second target area 40 is the same as the size of the second pixels (1) of the second template image. As shown in step 71 and step 711, gray step coding is performed, and gray levels in the horizontal direction and the vertical direction of each third pixel 4〇1 and each fourth pixel 501 are calculated, and then each third is used. The gray level gradient of the horizontal direction and the vertical direction of the pixel 4〇1 and each of the fourth pixels 5〇1 is different from the gradient direction code of the third pixels 4〇1 of the second target area 40 and the second The gradient direction code of the fourth pixel 5〇1 of the template image 50. The third pixel 4〇1 of the second target area 4〇 and the fourth pixels 5〇1 of the second template image 50 are smaller than the first pixels 102 of the original image 1〇 and the The first pixel 2〇2 of the first version of the image 20. In the real example, the 'cutting value of the second target area 4〇 and the second template image 5〇 is the cutting value of the original image 1 0 and the first version of the image 2〇 107494 .doc • 11 - 1291668 By the 'knife, that is, the child's fourth best sound $ n〗 9, the third image f4 () 1 number 1 ^, and the second pixel 2 〇 1 has a common hopper - often 401 number The number of the first pixels 101 is four.
在:亥貫施财’計算該第二目„域做料第三L 之梯之=向 1向碼及該第二樣版影像5G之該等第四像素501 又°碼之方法’係與上述之該原始影像之該等第一 像素102之梯度方向碼及該第一樣版影像 素202之梯度方向碼之方法相同，在此不加以㈣像 配合參考圖5及圖7 ’如步驟712所示，比對該等第三像 之梯度方向碼及該等第四像素501之梯度方向碼，計 二取仔複數個该第二目標區域4〇之不同位置與該第二樣版 影像50之相似度’定義該第二目標區域與該第二樣版影 像5〇相似度最高之區域為一目標影像60。 配合參考圖6及圖7’如步驟713所示，取得該目標影像 60之後’疋義該目標影像6〇之中心點6〇1為相對座標原 點，將該原始影像i 〇透過一電子元件樣式資料庫作一分 類，再將該目標影像60資料送至伺服機構，計算該目標影 像二之中心點6〇1之實際誤差並進行補償乂接著，讀取該 目標影像6G之資料，取得_包含複數個像素之掃描線資 料，再利用次像素的技術，求出該目標影像6〇之尺寸數 值0 本發明圖形比對搜尋方法，係利用灰階梯度方向作為描 述影像特徵的基礎，並採用兩階段搜尋比對方式，精確的 寻找出杻似度最大之影像物體。利用灰階梯度具有以下之 優點，當搜尋的目標物體與鄰域之灰階度雙化差異大時， 107494.doc •12· 1291668 閉丨尔+大，最能 ’可作為一個良好的影像 灰階梯度方向反映在物體相對影 完整保留原目標物體的物理特性 辨識基礎。 另，本發明係利用影像之灰階變化轉換成梯度方向碼， 已將其特徵保留下，故不受光源不均勻影響。另外，搭配 二元樹之相似度計算，更大幅縮短了繁叙計算時間。再 者，兩階段搜尋的方式’於第—階段先搜尋得該第—目伊In: Haiheng Shicai 'calculates the second item „ domain of the third L ladder = the 1-way code and the second pattern image 5G of the fourth pixel 501 and the code method The method of the gradient direction code of the first pixel 102 of the original image and the gradient direction code of the first version of the image element 202 is the same, and the image is not referred to as FIG. 5 and FIG. As shown, the gradient direction code of the third image and the gradient direction code of the fourth pixel 501 are taken to take a plurality of different positions of the second target area 4 and the second template image 50. The similarity 'is defined as the target image 60 with the highest similarity between the second target area and the second template image. The target image 60 is obtained after the target image 60 is obtained as shown in step 713 with reference to FIG. 6 and FIG. '疋疋 The target image 6〇's center point 6〇1 is the relative coordinate origin, the original image i 〇 is classified by an electronic component style database, and the target image 60 data is sent to the servo mechanism to calculate The actual error of the center point of the target image 2 is 6〇1 and compensated乂The data of the target image 6G is read, the scan line data of a plurality of pixels is obtained, and the size of the target image is determined by the technique of the sub-pixel. The graphic comparison search method of the present invention is utilized. The grey gradient direction is used as the basis for describing the image features, and the two-stage search comparison method is used to accurately find the image object with the greatest degree of similarity. The use of gray gradient has the following advantages when searching for the target object and the neighborhood. When the gray-scale doubled difference is large, 107494.doc •12· 1291668 Closed +++, the most can be used as a good image gray gradation direction reflected in the object relative to the shadow intact to retain the original target object physical characteristics identification basis In addition, the present invention converts the gray-scale change of the image into a gradient direction code, and has retained its characteristics, so that it is not affected by the unevenness of the light source. In addition, the similarity calculation of the binary tree is more greatly shortened. Calculate the time. In addition, the two-stage search method is to search for the first item in the first stage.
區域30之大約位置，再由該第—目標區域科圍向外擴^ 部分區域，以大於該第一目樣區域3〇之範圍，並定義一第 二目標區域40;而在第二階段搜尋中，係於 域财，以更小尺寸之料第三像素彻及料第四= 5〇1進行比對更精細之比對，藉此更精確地找尋出該目標 影像60。 ' 上述5鈿例僅為說明本發明之原理及其功效，並非限制 本奴明。因此習於此技術之人士對上述實施例進行修改及 艾化仍不脫本發明之精神。本發明之權利範圍應如後述之 申請專利範圍所列。 【圖式簡單說明】 圖1顯示本發明原始影像之示意圖； 圖2顯示本發明原始影像及第一樣版影像之示意圖； 圖3顯示本發明計算得第一目標區域之示意圖； 圖4顯不本發明第二目標區域及第二樣版影像之示意 圖； ® 5顯tf本發明計算得於第二目標區域中之目標影像之 107494.doc -13 -The approximate position of the area 30 is further expanded by the first target area to a larger area than the first target area, and a second target area 40 is defined; and the second stage is searched. In the middle, it is based on the domain, and the third pixel of the smaller size is the fourth=5〇1 to perform a finer alignment, thereby finding the target image 60 more accurately. The above 5 examples are only illustrative of the principles and effects of the present invention and are not intended to limit Benuming. Therefore, those skilled in the art can revise the above embodiments and the spirit of the present invention. The scope of the invention should be as set forth in the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a schematic view showing an original image of the present invention; FIG. 2 is a schematic view showing an original image and a first image of the present invention; FIG. 3 is a schematic view showing a first target region calculated by the present invention; A schematic diagram of the second target area and the second sample image of the present invention; the display image of the target image calculated in the second target area is 107494.doc -13 -
1291668 示意圖； 圖6顯示本發明目標影像及其中心點之示意圖；及 圖7顯示本發明圖形比對搜尋方法之流程圖。 【主要元件符號說明】 10 原始影像 20 第一樣版影像 30 第一目標區域 40 第二目標區域 50 第二樣版影像 60 目標影像 101 原始影像之圖樣 102 第一像素 201 第一樣版影像之圖樣 202 第二像素 401 第三像素 501 第四像素 601 目標影像之中心點 107494.doc -14-1291668 is a schematic diagram; FIG. 6 is a schematic diagram showing a target image of the present invention and its center point; and FIG. 7 is a flow chart showing a method of searching for a graphical comparison of the present invention. [Main component symbol description] 10 original image 20 first version image 30 first target area 40 second target area 50 second pattern image 60 target image 101 original image pattern 102 first pixel 201 first version image Pattern 202 second pixel 401 third pixel 501 fourth pixel 601 center point of the target image 107494.doc -14-