201250235 六、發明說明： 【發明所屬之技術領域】 本發明是有關於一種自動光學檢測方法（Automated Optical Inspection，A0I)’特別是指一種三維光學膜（3D optical film )的瑕疵（defect)檢測方法及系統。 【先前技術】 現有的光學膜瑕疵檢測技術，多半是針對表面不具紋 理（texture)特性之產品進行檢測。請參閱圖！，以表面不 具紋理特性之偏光片（polarized film)為例，根據偏光片分 別對應的影像11、12在色彩空間（c〇i〇I· Space )上的特性 ’即可直接判斷出何者具有瑕疵；進一步來說，顏色變異 微小的該影像11所對應的偏光片為無瑕疵（defect_free) 的偏光片，顏色變異較大的該影像12所對應的偏光片為具 有瑕疵的偏光片。 然而，三維光學膜屬於表面具有結構性紋理之產品， 故無法利用其對應的影像在色彩空間上的特性來進行瑕疵 檢測。 【發明内容】 因此，本發明之目的，即在提供一種三維光學膜的瑕 疵檢測方法。 於是，本發明三維光學膜的瑕疵檢測方法，係利用一 個處理單元配合-個影像操取單元來執行，該影像擁取單 元用以擷取對應於一個待檢測的三維光學膜的一個待檢測 影像；該方法包含下列步驟： 201250235 (A) 接收該待檢測影像； (B) 根據該待檢測影像t的至少__個區域，求得該區 域於一個第一方向上的複數個第一投影量； (C) 根據步驟⑻所求得之該區域的該等第-投影 直，求得對應於該區域的-個第—瑕㈣測參數； (D )根據步驟（C )所求得之對應於該區域的該第一 瑕疫檢測參數，判斷該待檢測影像中的該區域是否具有瑕 庇；及 (E)輸出步驟（D)的判斷結果。 本發明之另-目的，即在提供一種三維光學膜的瑕疵 檢測系統。 於疋，本發明二維光學膜的瑕麻檢測系統包含一個影 像榻取單元’及—個處理單元。該影像掏取單元用以操取 對應於一個待檢測的三維光學膜的一個待檢測影像。該處 理單元用以進行·接收該待檢測影像；根據該待檢測影像 中的至y個區域，求得該區域於一個第一方向上的複數 個第投影量，根據該區域的該等第一投影量，求得對應 於該區域的一個第一瑕疵檢測參數；及根據對應於該區域 的該第一瑕疵檢測參數，判斷該待檢測影像中的該區域是 否具有瑕疵’並輸出判斷結果。 本發明之功效在於：藉由根據待檢測影像中的該區域 的該等第一投影量所求得的該第一瑕疵檢測參數，可對具 有紋理特性的三維光學膜進行瑕疵檢測。 【實施方式】 201250235 有關本發明之前述及其他技術内容、特點與功效，在 以下配合參考圖式之一個較佳實施例的詳細說明中，將可 清楚的呈現。 參閱圖2 ’本發明三維光學膜的瑕疵檢測系統2之較佳 實施例包含一個光學鏡頭21、一個影像擷取單元22，及一 個處理單元23。其中’該影像擷取單元22用以擷取對應於 該光學鏡頭21下的一個三維光學膜的一個影像；該處理單 元23用以根據對應於該三維光學膜的該影像進行瑕疵檢測 〇 以下配合本發明三維光學膜的瑕疵檢測方法的一個較 佳實施例，說明該處理單元23所執行的步驟。其中，該三 維光學膜的瑕疵檢測方法分為系統初始化階段，及系統檢 測階段。 【系統初始化階段】 首先，參閱圖2〜圖5，該影像擷取單元22擷取對應於 一個無瑕疵的三維光學膜的一個第一訓練（training )影像 31，其中，該第一訓練影像31具有如圖3所示的結構性紋 理特性。 繼而，該處理單元23執行以下步驟，以得到一個最適 區域大小（size )。 在步驟41中’該處理單元23接收該第一訓練影像31 〇 在步驟42中，該處理單元23將該第一訓練影像31進 行二值化處理（thresholding )’以得到二值化的該第一訓練 201250235 衫像31，其中，二值化的該第一訓練影像3丨為一個二元（ binary)影像。 • 在步驟43中’該處理單元23根據二值化的該第一訓 練衫像3 1，求得至少一個水平紋理週期，及至少一個垂直 紋理週期。 該水平紋理週期及該垂直紋理週期配合圖5所示的一 個二元範例影像51進一步說明如後；其中，該二元範例影 像51包括其二元值為〇的複數個晝素（ρίχει) 511，及其 一元值為1的複數個晝素5丨2。在本較佳實施例中，該處理 單元23係根據式子（1)〜(2)的計算結果，以得到該水平紋理 ' 週期及該垂直紋理週期。 々為' —元紅例影像5 1的大小以m X η表示；且令該二元 範例影像51中每一個晝素511、512的二維座標以（χ，β表示 其中’ §玄一元範例影像51中最左上角的畫素511的二維座 標為（U) ’最右下角的畫素511的二維座標為（w，„);户(χ，少)代 表其二維座標為（x，j；)的畫素511、512之二元值。 m-i η (ο=Σ ΣΙρ(χ+^ y) ~ ρ(χ^ y)\............................(ι) χ=1 夕=1 SUMv ^ U) = Σ y+J)~ ρ(χ^ y)\............................(2) jC=l y=\ 其中，i = l，2，...，m-l，_/ = 1，2，·..，《 —1。 為了便於說明，將式子（1)〜(2)的計算結果分別以一個第 一曲線圖52及一個第二曲線圖53來示意；由該第一曲線 圖52可付知邮⑷峨⑻= 邮(12) = 〇，其對應的 物理意義為：該二元範例影像51於水平方向上紋理的重複 週期為4、8、12 ;類似地，由該第二曲線圖5 3可得知 7 201250235 ⑷St/Λ^邮⑻以伽匕嘲(12)=()，其對應的物理意義為： β 一兀範例影像51於垂直方向上紋理的重複週期為4、8、 7水平紋理週期以//_々咖〇ί/表示，垂直紋理週期以 Ci-ZM表示，則在本範例中，得到三個水平紋理週期與三 ， 個垂直紋理;即U_ = 4812 υ_ = 48,ι2。 更廣義。之，只要分別找出該第一曲線圖52及該第二 曲線圖53中的波谷’即可得到水平紋理週期 及垂直紋理週期〇__0ί〇 ;換句話說，根據式子⑴〜⑺的 計算結果，配合預設的一個第一門檻值（以咖表示）及一 個第二門檻值（以私表示）’即可得到付一㈣及乙一； 其中’ SUM”讲一period)與岬，及SUMyff(v — peri〇(〇與牦之關 係式進一步表示如式子（3)〜(4)。 SUMH_diff (Η _ period) < ..............................................201250235 VI. Description of the Invention: [Technical Field] The present invention relates to an automated optical inspection (A0I)', particularly to a three-dimensional optical film (Dan) method for detecting defects And system. [Prior Art] Existing optical film detection techniques are mostly for detecting products having no texture characteristics on the surface. Please see the picture! Taking a polarized film with no surface texture as an example, according to the characteristics of the corresponding images 11 and 12 of the polarizer in the color space (c〇i〇I·Space), it can be directly judged which has the flaw. Further, the polarizer corresponding to the image 11 having a small color variation is a defect-free polarizer, and the polarizer corresponding to the image 12 having a large color variation is a polarizer having a flaw. However, the three-dimensional optical film belongs to a product having a structural texture on the surface, and therefore it is impossible to detect the 瑕疵 by the characteristics of the corresponding image in the color space. SUMMARY OF THE INVENTION Accordingly, it is an object of the present invention to provide a method for detecting a three-dimensional optical film. Therefore, the method for detecting the flaw of the three-dimensional optical film of the present invention is performed by using a processing unit and an image capturing unit for capturing a to-be-detected image corresponding to a three-dimensional optical film to be detected. The method includes the following steps: 201250235 (A) receiving the image to be detected; (B) determining a plurality of first projections of the region in a first direction according to at least __ regions of the image to be detected t (C) according to the first-projection straightness of the region obtained in step (8), obtain a -first (four) measurement parameter corresponding to the region; (D) corresponding to the step (C) The first plague detection parameter in the area determines whether the area in the image to be detected has a smear; and (E) the judgment result of the output step (D). Another object of the present invention is to provide a helium detection system for a three-dimensional optical film. In 疋, the ramie detection system of the two-dimensional optical film of the present invention comprises an image taking unit and a processing unit. The image capturing unit is configured to acquire a to-be-detected image corresponding to a three-dimensional optical film to be detected. The processing unit is configured to: receive and receive the image to be detected; and determine, according to the y regions in the image to be detected, a plurality of first projection amounts of the region in a first direction, according to the first regions of the region a projection quantity is obtained, and a first detection parameter corresponding to the area is obtained; and according to the first detection parameter corresponding to the area, determining whether the area in the image to be detected has 瑕疵' and outputting a determination result. The effect of the present invention is that flaw detection can be performed on a three-dimensional optical film having texture characteristics by determining the first flaw detection parameter based on the first projection amounts of the region in the image to be detected. The above and other technical contents, features, and advantages of the present invention will be apparent from the following detailed description of the preferred embodiments. Referring to Fig. 2, a preferred embodiment of the flaw detection system 2 of the three-dimensional optical film of the present invention comprises an optical lens 21, an image capture unit 22, and a processing unit 23. The image capturing unit 22 is configured to capture an image corresponding to a three-dimensional optical film under the optical lens 21; the processing unit 23 is configured to perform flaw detection according to the image corresponding to the three-dimensional optical film. A preferred embodiment of the method for detecting flaws in a three-dimensional optical film of the present invention illustrates the steps performed by the processing unit 23. Among them, the detection method of the three-dimensional optical film is divided into a system initialization phase and a system detection phase. [System Initialization Phase] First, referring to FIG. 2 to FIG. 5, the image capturing unit 22 captures a first training image 31 corresponding to a flawless three-dimensional optical film, wherein the first training image 31 It has the structural texture characteristics shown in Figure 3. Then, the processing unit 23 performs the following steps to obtain an optimum area size (size). In step 41, the processing unit 23 receives the first training image 31. In step 42, the processing unit 23 performs a binarization process on the first training image 31 to obtain a binarized number. A training 201250235 shirt image 31, wherein the binarized first training image 3 is a binary image. • In step 43 the processing unit 23 determines at least one horizontal texture period and at least one vertical texture period based on the binarized first training shirt image 3 1,. The horizontal texture period and the vertical texture period are further described in conjunction with a binary example image 51 shown in FIG. 5; wherein the binary example image 51 includes a plurality of elements (ρίχ ει) 511 whose binary value is 〇. And a plurality of elements 5丨2 whose unitary value is 1. In the preferred embodiment, the processing unit 23 obtains the horizontal texture 'period and the vertical texture period according to the calculation results of the equations (1) to (2). The size of the image is 5 x η; and the two-dimensional coordinates of each of the pixels 511, 512 in the binary sample image 51 are (χ, β represents the example of the § 玄元元The two-dimensional coordinates of the pixel in the upper left corner of the image 51 are (U) 'The two-dimensional coordinates of the pixel 511 at the bottom right corner are (w, „); the household (χ, less) represents its two-dimensional coordinates ( x, j;) The binary value of pixels 511, 512. mi η (ο=Σ ΣΙρ(χ+^ y) ~ ρ(χ^ y)\............. ...............(ι) χ=1 夕=1 SUMv ^ U) = Σ y+J)~ ρ(χ^ y)\........ ....................(2) jC=ly=\ where i = l,2,...,ml,_/ = 1,2,· .., "-1. For convenience of explanation, the calculation results of the equations (1) to (2) are respectively indicated by a first graph 52 and a second graph 53; from the first graph 52, Fu Zhimail (4) 峨 (8) = postal (12) = 〇, the corresponding physical meaning is: the repetition period of the texture of the binary example image 51 in the horizontal direction is 4, 8, 12; similarly, by the second graph 5 3 can know 7 201250235 (4) St / Λ ^ mail (8) with gamma ridicule (12) = (), its right The physical meanings are as follows: β 兀 影像 影像 影像 51 51 51 51 纹理 纹理 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀 兀In this example, three horizontal texture periods and three vertical textures are obtained; that is, U_ = 4812 υ _ = 48, ι2. More broadly, as long as the first graph 52 and the second graph 53 are respectively found. The trough can get the horizontal texture period and the vertical texture period 〇__0ί〇; in other words, according to the calculation results of the equations (1) to (7), with a preset first threshold (in coffee) and a second The threshold value (in private) can be paid for one (four) and one for one; where 'SUM' speaks a period and 岬, and SUMyff(v — peri〇 (the relationship between 〇 and 进一步 is further expressed as equation (3) ~(4). SUMH_diff (Η _ period) < ..................................... .........
w J SUMV diff{V period) < ............................................. (4) 在步驟44中，該處理單元23根據該水平紋理週期及 該垂直紋理週期，決定該最適區域大小。令該最適區域大 小以WXA表示；在本較佳實施例中，係以最小的該水平紋理 週期，及最小的該垂直紋理週期來決定該最適區域大小； 即，w = min(开;7eriW)，A = min(厂 。 然後，參閱圖2、圖3及圖6，該影像擷取單元22擷 取對應於一個具有瑕疵的三維光學膜的一個第二訓練影像 32，該處理單元23接著執行以下步驟，以得到一個第一瑕 疵檢測參數上限值、一個第一瑕疵檢測參數下限值、一個 第二瑕疵檢測參數上限值，及一個第二瑕疵檢測參數下限 8 201250235 值。 在步驟61中，該處理單元23接收該第二訓練影像32 〇 在步驟62中，該處理單元23將該第二訓練影像32進 行二值化處理，以得到二值化的該第二訓練影像32。 在步驟63中，該處理單元23根據預先求得的該最適 區域大小（，將二值化的該第二訓練影像32分割為複 數個訓練區域·，其中，每一訓練區域的大小等於該最適區 域大小。 在步驟64中，該處理單元23根據二值化的該第二訓 練衫像32中的該等訓練區域，求得每一訓練區域於一第一 方向上的複數個第一訓練投影量，及於一第二方向上的複 數個第二訓練投影量。在本較佳實施例中，該第一方向為 一水平方向，該第二方向為一垂直方向，該第二方向係垂 直於該第一方向；每一訓練區域的該等第一、二訓練投影 量係利用式子（5)〜(6)來計算。 ... . ........................................................(5) vP(x) = ^P(x,y)....................................................(6) 其中，對於每一訓練區域，母⑴代表該訓練區域的該 等第-訓練投影量，阶)代表該訓練區域的該等第二訓練投 影ΐ ; 3； = 1，2，··.’Α，x = l，2，...，w;户(以代表在該訓練區域中， 其二維座標為（AJV)的畫素之二元值。 在步驟65中，該處理單元23根據每一訓練區域的該 等第一訓練投影量，及該等第二訓練投影量，計算對應於 201250235 每一訓練區域的一個第一訓練標準差，及一個第二訓練標 準差；其計算如式子（7)〜(8)所示。 HP _ std=...................................(7)w J SUMV diff{V period) < ........................................ (4) In step 44, the processing unit 23 determines the optimal region size based on the horizontal texture period and the vertical texture period. Let the optimum region size be represented by WXA; in the preferred embodiment, the optimal region size is determined by the minimum horizontal texture period and the minimum vertical texture period; that is, w = min (open; 7 eriW) , A = min (factory. Then, referring to FIG. 2, FIG. 3 and FIG. 6, the image capturing unit 22 captures a second training image 32 corresponding to a three-dimensional optical film having a cymbal, and the processing unit 23 then performs The following steps are performed to obtain a first 瑕疵 detection parameter upper limit value, a first 瑕疵 detection parameter lower limit value, a second 瑕疵 detection parameter upper limit value, and a second 瑕疵 detection parameter lower limit 8 201250235 value. The processing unit 23 receives the second training image 32. In step 62, the processing unit 23 performs binarization processing on the second training image 32 to obtain the binarized second training image 32. In step 63, the processing unit 23 divides the binarized second training image 32 into a plurality of training regions according to the optimal region size obtained in advance, wherein the size of each training region, etc. The optimum area size. In step 64, the processing unit 23 determines a plurality of first regions of each training region in a first direction according to the training regions in the second training shirt image 32 that are binarized. The training projection amount and the plurality of second training projections in a second direction. In the preferred embodiment, the first direction is a horizontal direction, and the second direction is a vertical direction, the second direction The first and second training projections of each training region are calculated by using equations (5) to (6). ........... ........................................(5) vP (x) = ^P(x,y).......................................... (6) wherein, for each training region, the parent (1) represents the first-training projection amount of the training region, and the step represents the second of the training region. Training projection ΐ ; 3; = 1,2,··.'Α,x = l,2,...,w; household (to represent the two-dimensional coordinates (AJV) in the training area The binary value. In step 65, the processing unit 23 is based on each training area. The first training projection amount, and the second training projection amount, calculating a first training standard deviation corresponding to each training region of 201250235, and a second training standard deviation; the calculation is as in equation (7)~ (8). HP _ std=.....................................(7)
Vp_Std = \-Xmx)~vPmgy x-\ ⑻ 其中’對於每一訓練區域，母-版代表對應於該訓練區 域的該第一訓練標準差，印—细代表對應於該訓練區域的該 第一訓練標準差，母^代表該訓練區域的該等第一訓練投影 量的平均值，印W代表該訓練區域的該等第二訓練投影量 的一平均值。 ，驟66中，该處理單元23根據分別對應於該等 練區域的該等第—訓練標準差及該等第H丨練標準差， :該第-瑕疲檢測參數上、下限值，及該第二瑕庇檢測 上、下限值；其計算如式子（9)〜(12)所示。Vp_Std = \-Xmx)~vPmgy x-\ (8) where 'for each training area, the mother-print represents the first training standard deviation corresponding to the training area, and the print-fine representation corresponds to the first of the training area The training standard deviation, the parent ^ represents the average of the first training projections of the training region, and the print W represents an average of the second training projections of the training region. In step 66, the processing unit 23 is configured according to the first training standard deviation corresponding to the training regions and the standard deviation of the H-th training, the upper and lower limit values of the first-weekness detecting parameter, and The second shelter detects upper and lower limits; the calculation is as shown in equations (9) to (12).
Hp-thr» =HP_Stdavs +kxHp_Std d............... „ , ~ ..............(9) P-Stdavg kxHp_Stdild ................................... P_thru =Vp^Stdavg +kxVp_Stdstd ................. ..............(11)Hp-thr» =HP_Stdavs +kxHp_Std d............... „ , ~ ..............(9) P-Stdavg kxHp_Stdild .. ................................. P_thru =Vp^Stdavg +kxVp_Stdstd .......... ....... ..............(11)
Vp_thn =Vp^Stdmg ................................... 其中ϋ代表該第一瑕疵檢測參 (1: 代表該等第一訓練標準差的一個平 义Vp_thn = Vp^Stdmg ....................................... where ϋ represents the first 瑕疵 detection parameter ( 1: represents a flat meaning of the first training standard deviation
表該等第-訓練標準差的—個標準差 hp~sKA 檢測參數下限值、™第二瑕二 10 201250235 、少代表該等第二訓練標準差的一個平均值' 匕 代表該等第二訓練標準差的一個標準差、印—岣代表該第二 瑕庇檢測參數下限值H個預設常數。在本較佳實施例 中，众=3。 【系統檢測階段】 參閲圖2及圖7’在進行完系統初始化階段後，即可將 預先求得的該最適區域大小（⑽）、該第—㈣檢測參數 上' 下限值（H母—叫），及該第二瑕疲檢測參數上、 下限值（）應用於三維光學膜的瑕癌檢測。 該〜像#貞取單元22持續操取對應於待檢測的三維光學 膜的待檢測影像。 在步驟71巾，該處理單元23接收—個待檢測影像。 在步驟72巾，該處理單元23將該待檢測影像進行二 值化處理’以得到二值化的該待檢測影像。 冰在步驟73巾，該處理單元23根據該最適區域大小（ W)’將二值化的該待檢測影像分割為複數個區域；其中 每區域的大小等於該最適區域大小。 旦,步驟74中，該處理單元23根據二值化的該待檢測 衫像中的該等區域，求得每一區域於該第一方向（水平方 Μ上的複數個第-投影量’及於該第二方向（垂直方向 具：：複數個第二投影量；其計算類似於式子(5)〜(6)，故不 丹賢述。 —江在=驟75中，該處理單元23根據每一區域的該等第 杈影篁，及該等第二投影量，計算對應於每一區域的— 11 201250235 個第-㈣檢測參數（以①表示），及—個H疵檢測參 數（以G表不）；在本較佳實施例中，該處理單S 23係根據 每-區域的該等第―投影量計算—第—標準差作為對應於 該區域的該第-難檢測參數，並根據每—區域的該等第 一投影量计算一第二標準差作為對應於該區域的該第二瑕 疲檢測參數，其計算類似於式子⑺〜⑻，故不再贊述。 在步驟76巾，該處理單元23根據對應於每一區域的 該第一瑕疵檢測參數及該第二瑕疵檢測參數，並配合該第 -瑕庇檢測參數上、下限值（心，及該第二暇 疲檢測參數上、下限值（…H叫），以判斷該待檢測 影像中的每-區域是否具有瑕气。其中，對於每一區域， 若其第-瑕庇檢測參數及第二瑕疵檢測參數兩者其中任一 者不符合關係式（11)〜(12)，則代表該區域具有瑕疵；否則， 代表該區域為無瑕疵。Table - the standard deviation of the standard - training standard hp ~ sKA detection parameter lower limit, TM second 瑕 2 10 201250235, less represents an average of the second training standard deviation ' 匕 represents the second A standard deviation of the standard deviation of the training, 岣-岣 represents the lower limit of the second refuge detection parameter H preset constants. In the preferred embodiment, the population = 3. [System detection phase] Referring to Fig. 2 and Fig. 7', after the system initialization phase is completed, the optimum region size ((10)) and the first (4) detection parameter can be obtained as the lower limit value (H mother). - called), and the second fatigue detection parameter upper and lower limits () applied to the three-dimensional optical film of cancer detection. The image capturing unit 22 continuously takes the image to be detected corresponding to the three-dimensional optical film to be detected. At step 71, the processing unit 23 receives an image to be detected. In step 72, the processing unit 23 binarizes the image to be detected to obtain a binarized image to be detected. In step 73, the processing unit 23 divides the binarized image to be detected into a plurality of regions according to the optimal region size (W)'; wherein the size of each region is equal to the optimal region size. In step 74, the processing unit 23 obtains, according to the binarized regions in the to-be-detected shirt image, each region in the first direction (a plurality of first-projection amounts on the horizontal square) and In the second direction (the vertical direction has: a plurality of second projection amounts; the calculation is similar to the equations (5) to (6), so the Bhutan syllabus. - Jiang in = step 75, the processing unit 23 Calculating the corresponding parameters of each region corresponding to each region, based on the first image of each region, and the second projection amount, and the number of detection parameters (indicated by 1) and the H detection parameters ( In the preferred embodiment, the processing order S 23 is calculated according to the first-projection amount of each region-the first standard deviation as the first-difficulty detecting parameter corresponding to the region. And calculating a second standard deviation according to the first projection amount of each region as the second fatigue detection parameter corresponding to the region, and the calculation is similar to the equations (7) to (8), and therefore is not further described. 76 towel, the processing unit 23 according to the first flaw detection parameter corresponding to each region and the second疵Detecting parameters, and matching the upper and lower limits of the first-neck detection parameter (heart, and the upper and lower limits of the second fatigue detection parameter (...H) to determine each of the images to be detected - Whether the area has helium. For each area, if either of the first-neck detection parameter and the second detection parameter does not conform to the relationship (11) to (12), it means that the area has瑕疵; otherwise, it represents the area as innocent.
Hp_thr, <Hd<Hp thr ...................... 一 ....................⑴）Hp_thr, <Hd<Hp thr ...................... One................... .(1))
Vp _ thr, <Vd <Vp_ ............................... _ ...................(12) 在步驟77中’該處理單元23輸出步驟％的判斷結果 ，即，關於該等區域是否具有瑕疵的檢測結果；然後回 到步驟71繼續接收對應於下—個（next)待檢測的三維光 學膜的下一個待檢測影像。 值得-提的是，由步驟76的判斷結果，可得知待檢測 的三維光學膜中瑕庇區域的比例；再者，由於該三維光學 膜的瑕疵檢測系統2的該光學鏡頭21之架設位置為已知， 故該待檢測影像與該待檢測的三維光學膜之幾何對應關係 12 201250235 亦為已知，因此，由該待檢測影像中該等區域的檢測結果 ’即可對應得知該待檢測的三維光學膜中瑕疲發生的相關 位置。 ^综上所述，藉由收對應於該待檢測的三維光學膜的該 待檢測影像中分別對應於該等區域的該等第一、二瑕庇檢 測參數’可對具有紋理特性的三維光學膜進行瑕疵檢測； 更進一步來$，配合預先求得的該最適區域大小、該第一 瑕'疵檢/則參數上、下限值，及該第二瑕疲檢測參數上、下 限值，可對具有紋理特性的三維光學膜進行全自動的瑕疵 檢測，故確實能達成本發明之目的。 淮上所述者，僅為本發明之較佳實施例而已，當不 μ此限疋本發明實施之範圍’即大凡依本發明申請專利 範圍及發明說明内容所作之簡單的等效變化與修飾，皆仍 屬本發明專利涵蓋之範圍内。 【圖式簡單說明】 圖1疋不意、圖，說明一個無瑕疯的偏光片，及一個 具有㈣的偏光片’兩者分職應的影像； 圖κ丨塊圖，說明本發明三維光學膜的瑕疯檢測 糸統之一個較佳實施例； 圖 3 县一一 I π /、’、圖，說明對應於一個無瑕疵的三維光學 2個第-訓練影像’及對應於—個具有瑕㈣三維光 予膜的一個第二訓練影像； 制古I 4是—流程圖’說明在本發明三維光學膜的瑕疲檢 之個較佳實施例的系統初始化階段中，用於求得 13 201250235 一最適區域大小的步驟； 圖5是一示意圓，說明一個二元範例影像、一個第— 曲線圖及一個第二曲線圖； 圖6是一流程圖，說明在該系統初始化階段中，用於 长付個第一瑕疲檢測參數上限值、一個第一瑕疲檢測參 數下限值、一個第二瑕疵檢測參數上限值，及一個第二瑕 庇檢測參數下限值的步驟；及 圖7是一流程圖’說明本發明三維光學膜的瑕疵檢測 方法之該較佳實施例的系統檢測階段所包括的步驟。 14 201250235 【主要元件符號說明】 2 ....... …三維光學膜的瑕 51…… …二元範例影 疫檢測系統 511 ···· …晝素 21…… …光學鏡頭 512 ···· …晝素 22…… …影像擷取單元 52…… …第 曲線圖 23…… …處理單元 53…… …第二曲線圖 31…… …第一訓練影像 61〜66· …步驟 32…… …第二訓練影像 71〜77. …步驟 41〜44. …步驟 15Vp _ thr, <Vd <Vp_ ......................... _ ........ ..... (12) In step 77, the processing unit 23 outputs the result of the determination of step %, that is, whether or not the areas have detection results of 瑕疵; then returns to step 71 to continue receiving. Corresponding to the next image to be detected of the next three-dimensional optical film to be detected. It is worth mentioning that, from the judgment result of the step 76, the proportion of the shelter area in the three-dimensional optical film to be detected can be known; further, the position of the optical lens 21 of the flaw detection system 2 of the three-dimensional optical film is set. It is known that the geometrical correspondence between the image to be detected and the three-dimensional optical film to be detected 12 201250235 is also known. Therefore, the detection result of the regions in the image to be detected can be correspondingly known to be The relative position of the occurrence of fatigue in the detected three-dimensional optical film. In summary, the three-dimensional optical with texture characteristics can be obtained by receiving the first and second shelter detection parameters corresponding to the regions in the image to be detected corresponding to the three-dimensional optical film to be detected. The membrane is subjected to helium detection; further, $ is matched with the optimum region size obtained in advance, the upper and lower limits of the first flaw detection/then parameter, and the upper and lower limits of the second fatigue detection parameter, The full-automatic flaw detection of the three-dimensional optical film having the texture property can be achieved, and the object of the present invention can be achieved. The above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention, that is, the simple equivalent changes and modifications made by the scope of the invention and the description of the invention. All remain within the scope of the invention patent. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a schematic diagram showing an innocent polarizer and an image having a polarizer of (4), which are divided into two parts; and a block diagram showing the three-dimensional optical film of the present invention. A preferred embodiment of the madness detection system; Fig. 3 County I π /, ', diagram, description corresponding to a flawless three-dimensional optical 2 -th training image 'and corresponding to a 瑕 (four) three-dimensional A second training image of the light film; the formula I 4 is - the flow chart 'illustrated in the system initialization stage of the preferred embodiment of the three-dimensional optical film of the present invention, used to obtain 13 201250235 Figure 5 is a schematic circle illustrating a binary example image, a first-graph and a second graph; Figure 6 is a flow chart illustrating the long-term payment during the initialization phase of the system a first fatigue detection parameter upper limit value, a first fatigue detection parameter lower limit value, a second 瑕疵 detection parameter upper limit value, and a second smear detection parameter lower limit value step; and FIG. 7 is a flowchart 'description The step of detecting phase system of the embodiment comprises a three-dimensional flaw detection method of the invention, the optical film of the preferred embodiment. 14 201250235 [Explanation of main component symbols] 2 ....... 3D optical film 瑕 51...... Binary paradox detection system 511 ·····昼素 21...... ...optical lens 512 ·· ··昼素22...... Image capture unit 52... Figure 23... Processing unit 53... Second curve 31... First training image 61~66·...Step 32... ...the second training image 71~77. ...Steps 41~44. ...Step 15