1336586 年ιΐ~月ίο 替換頁 九、發明說明： --— 【發明所屬之技術領域】 纟發明是—種影像修復方法’尤其是關於-種用於相 . 機感光几件塵染造成影像缺陷的影像修復方法。 【先前技術】 數位相機的技術在最近這幾年來有了突破性的進展， 由於省去傳統相機需要沖洗底片的麻須，m以使用數位相 機的人口正快速的增長中。而隨著數位單眼相機價格下滑 ’以及數位單眼相機強大且精準的測光性、優異且精準的 對：功能、高感光度以及高擴紐，例如使用者可因應自 己而求，更換不同效能或效果的鏡頭，使得使用數位單眼 相機的人數大為增加。不過’雖然數位單眼相機有上述許 多優點，⑯由於數位單眼相機是透過電力操控機械設備， 容易產生輕微的靜電，因此在更換鏡頭的時候，若不注意 就很容易將小塵埃吸人該數位單目M目機内部的感光元件^ 感光輕合元件（CCD)或金氧互補感光搞合元件（cm叫 等元件）纟面’所以用該已染塵的數位單眼相機拍攝時， 將在數位照片上產生小汙點’這就是所料「染塵 一般來說，如果數位單 都得送往廠商手動清除，不 心就有可能傷及該感光元件 出「感光元件自我清理架構 眼相機出現染塵現象，大部分 過通常價格昂貴，而且一不小 。最近也開始有相機廠商發明 」，其原理是使用壓電元件產 1336586 生超音波震動，讓粉塵脫落， 法對付黏度較高的粉塵，甚為 但這樣的機制’可能還是無 可惜。1336586 ιΐ~月ίο Replacement page IX, invention description: --- [Technical field of invention] 纟Invention is a kind of image restoration method' especially related to Image restoration method. [Prior Art] The technology of digital cameras has made a breakthrough in recent years. Due to the elimination of the need for conventional cameras to wash the negatives of the negative film, the population using digital cameras is rapidly growing. And with the price reduction of digital monocular cameras' and the powerful and accurate metering, excellent and precise pairing of digital monocular cameras: function, high sensitivity and high expansion, for example, users can change their performance or effect according to their own needs. The lens has greatly increased the number of people using digital monocular cameras. However, although the digital monocular camera has many of the above advantages, 16 because the digital monocular camera is mechanically controlled by electric power, it is easy to generate a slight static electricity. Therefore, when the lens is changed, it is easy to attract the small dust if not paying attention to the digital single. Photosensitive element inside the M mesh machine ^ Photosensitive light-emitting element (CCD) or gold-oxygen complementary light-sensing component (cm called component) behind the surface 'So when shooting with this dusty digital monocular camera, will be in digital photos There is a small stain on the 'this is what is expected. Dust. Generally speaking, if the digital order has to be sent to the manufacturer for manual removal, it may cause damage to the photosensitive element. "Photosensitive component self-cleaning structure, the camera is dusty. Most of them are usually expensive, and they are not small. Recently, camera manufacturers have invented them. The principle is to use piezoelectric elements to produce 1336586 ultrasonic vibrations, so that the dust can fall off, and the dust is highly resistant to dust. But such a mechanism' may still be a pity.
為了避免前述的相機感光元件於清洗時造成二次傷害 或者無法完全清除的問題，可利用既有之影像處理軟體或 影像修補技術處理或修補數位相片的染塵缺陷，亦可達到 相同的效果。其中，影像處理軟體例如Adobephotoshop的 修復筆刷工具（HeaHng Brush^ Ulead ph〇t〇impact 的修容 工具（Touch-up Tool)可將數位相片的染塵缺陷去除其作法 是先在未染塵的影像區域中選定要填補在染塵缺陷區域之 ，質，接著再將該選定材質貼在染塵缺陷的位置，如此即 ° π成‘補效果，但是這樣的方法必須手動選取要清除的 染塵缺陷並自行選取材質進行填補，戶斤以耗時耗力。除此 之外，無論是Photoshop的修復筆刷工具或ph〇t〇impact 的修容工|，其主要工作原理都是利用材質複製方法將 周圍的影像資訊貼補到有染塵缺陷的地方，如果染塵缺陷 :洛在影像中具有線性圖形結構的地方，則利用修容工具 修補的效果將會不佳。 正$除了影像處理軟體之外，也可以用影像修補技術來修 日^塵造成的圖像缺陷。所謂影像修補⑽哪化—㈣ 1用來修補被破壞影像的一項技術，可進行數位影像 :裂痕修補、移除不想要的物件或移除附加在圖片上的 —=與戮印。既有之影像修補技術主要分為兩種類型，其 :利用材質合成(texture別讣㈣的方法，可以用來修補 乾圍的目標區域，但是卻只適合修補影像中純材質的部 1336586 99年11月10曰修正^換頁j 份；另一則是由Bertalmio等提出之基於偏微分方程（PanialIn order to avoid the above-mentioned problem that the camera photosensitive element causes secondary damage during cleaning or cannot be completely removed, the existing image processing software or image repair technology can be used to process or repair the dust defect of the digital photo, and the same effect can be achieved. Among them, image processing software such as Adobe Photoshop's repair brush tool (HeaHng Brush^ Ulead ph〇t〇impact's Touch-up Tool) can remove the dust defects of digital photos, which is first in the dust-free In the image area, the quality to be filled in the dust-defective area is selected, and then the selected material is attached to the position of the dust-defective defect, so that π is a 'complementary effect, but such a method must manually select the dust to be removed. Defects and self-selecting material to fill, the household is time-consuming and labor-intensive. In addition, whether it is Photoshop repair brush tool or ph〇t〇impact repair workman|, its main working principle is to use material copy The method attaches the surrounding image information to the place where the dust is damaged. If the dust is defective: Luo has a linear graphic structure in the image, the effect of repairing with the repair tool will be poor. In addition, image repair technology can also be used to repair image defects caused by dust. The so-called image repair (10) - (4) 1 a technique used to repair damaged images, Digital imagery can be performed: crack repair, removal of unwanted objects, or removal of -= and stencils attached to the image. The existing image patching techniques are mainly divided into two types, which: use material synthesis (texture (4) The method can be used to repair the target area of the dry circumference, but it is only suitable for repairing the pure material of the image. 1336586 November 10, 2010 corrections ^ page replacement j; the other is based on the partial differential equation proposed by Bertalmio et al. (Panial
Differential Equations)的影像修補演算法（M Bertalmi〇, g. Sapiro, V. Caselles, and C. Ballester，“Image inpainting，，，inImage Correction Algorithm for Differential Equations (M Bertalmi〇, g. Sapiro, V. Caselles, and C. Ballester, “Image inpainting,,, in
Proc. ACM Conf. Comp. Graphics (SIGGRAPH), NewProc. ACM Conf. Comp. Graphics (SIGGRAPH), New
Orleans，LA，July 2000, pp. 417-424 )以及據此演算法推演Orleans, LA, July 2000, pp. 417-424) and derivation based on this algorithm
的一系列演算法修補技術，其中’以Criminisi等提出的「 基於範例影像修補（exemplar_based image inpainting)」演算 法（Criminisi，P. ：Terez and K T〇yama， “ Regi〇n FUUng and Object Removal by Exemplar-Based Image Inpainting," IEEE Trans· lmage Processing，v〇1 13, N〇 9, 2〇〇4)的影像 修補效果最佳，其係先找出要填補區域的輪廓，然後沿著 輪廓的每個像素，以其為中心定義一個9χ9像素區域的補丁 區域（patch)，然後計算每個補丁區域的優先權（ρΗ〇Γί^) 大小，有最大優先權的補丁區域則優先填補。如此，即能 夠同時把數位影像中的大範圍之材質缺陷及線性結構缺陷 完成修補，並且相對其他演算法顯得非常有效率。為進一 步說明之’請配合參考第五圖’首先，先將影像中被染塵 部份標記出來’ ^義為目標區域⑼’其中該目標區域叫 輪廊則標記為沿，之後’重複以下步驟： (1) 初始化：在初始化的時候，先給予每個像素一 個信賴值（confidence value)，以C(p)表示，其規則是假 如—P點是在目標區域裡，則C(p) = 0，否則= i。 (2) 對於每個位在輪廓®上的點P，定義一個以p 為中〜且$&圍為9x9像素大小的一補丁區域（叫&，以 1336586 99 年 11 月 10 日 符號％表示）。之後計算每個位在輪廓上的補丁區域之優 先權（P(p))，其計算公式為：= C〇) X Ζ)(ρ)，其中c(p) 該信賴項目之計算方式如 ’ I % I則是補丁區域（) 為信賴項目（confidence term) 〜、Σγψρη(/-Q)C(g) 下：c(p)=—^~~;其中 的面積大小’ / =整張影像，而ΖΧ/7)為data term，計算方 · np 式如下：乃化）—a ，其中，為等照度線，即垂直 於梯度向量的向量，％為垂直於輪廓的單位向量，而〇為 255 ’係為對D(尸）進行正規化（N〇rmaiize)。 (3 ) 選擇一個有最高優先權的補丁區域（以符號心 表示）先填補，然後在來源（source)區域（φ)，選定一 9Χ9像 數大小的補丁塊（以符號％表示），利用RGB三色的方差 和（Sum of Squared Differences ，SSD)來比對該補丁區 域以及f補丁塊的相似程度，其中方差和的公式如下： (〜(P) iGp(P)~Gq(p)f (Bp(p)~Bq(p)fA series of algorithmic patching techniques, including 'exemplar_based image inpainting' algorithm proposed by Criminisi et al. (Criminisi, P.: Terez and KT〇yama, “Regi〇n FUUng and Object Removal by Exemplar -Based Image Inpainting," IEEE Trans lmage Processing, v〇1 13, N〇9, 2〇〇4) has the best image patching effect, which first finds the contour of the area to be filled, and then follows the contour For each pixel, a patch of 9χ9 pixel area is defined around it, and then the priority (ρΗ〇Γί^) of each patch area is calculated, and the patch area with the highest priority is preferentially filled. That is to be able to repair a large range of material defects and linear structural defects in the digital image at the same time, and it is very efficient compared to other algorithms. For further explanation, please refer to the fifth figure. First, the image is dyed first. The dust part is marked out as '^ is the target area (9)' where the target area is called the edge of the ship, and then the following steps are repeated: (1) Initialization: At the time of initialization, each pixel is given a confidence value, denoted by C(p), and the rule is that if the point P is in the target area, then C(p) = 0 , otherwise = i. (2) For each point P on the contour ®, define a patch area with p as medium ~ and $& around 9x9 pixels (called &, 1336586 99 years 11 The symbol % is displayed on the 10th of the month. Then calculate the priority (P(p)) of the patch area of each bit on the contour, which is calculated as: = C〇) X Ζ)(ρ), where c(p) The calculation method of the trust project is as follows: 'I % I is the patch area () is the confidence term (containment term) ~, Σγψρη(/-Q)C(g): c(p)=-^~~; The area size ' / = the entire image, and ΖΧ / 7) is the data term, the calculation formula · np is as follows: Nahua) - a , where is the iso-illuminance line, that is, the vector perpendicular to the gradient vector, % is perpendicular to The unit vector of the contour, and 〇 255 ' is the normalization of D (the corpse) (N〇rmaiize). (3) Select a patch area with the highest priority (with a symbol heart) Show) fill first, then in the source area (φ), select a 9 Χ 9-image patch block (indicated by the symbol %), using RGB three-color variance and (Sum of Squared Differences, SSD) to compare The similarity between the patch area and the f patch block, where the formula of the variance sum is as follows: (~(P) iGp(P)~Gq(p)f (Bp(p)~Bq(p)f
SSD = ^ ，其中’心、G々、心分別表示 J高優先權之補丁區域（❼）内某-像素P的RGB三色色 度值：而〜、％、&則分表表示補丁塊（6)内像素q的 色度值。其中，方差和值最小的補丁塊（％)即 ，、取先推之補丁區域（^)最為相似，設定為最佳補 丁塊（以4號&表示）。找出該最佳補丁塊（心）之後， 針對最高優先權之補丁區域（❼）未填滿的像素對應於最 ς 8 1336586 99年11月1〇曰修正替換頁 佳補丁塊（^)的梯IΦ , 的像素位置，一個像素對一個像素貼補過 去即可。 (4)更新前一個被修補區域的信賴值（c〇nfidence ) ”係為把步驟（2)計算出來的C(p)值給予剛被修補 的每個像素’作為它們的信賴值（）。 (5 )如果目標區域Ω依舊存在，則重複步驟（2)〜(4) ，直到整各區域都被填滿為止。 然而Criminisi提出的演算法卻存在如下缺點： 1.由於原始决异法是針對整張影像作搜尋以尋找相似 的補丁區域填補目標區域，因此如果修補的區域是落在影 像中重覆性的結構上，而且在這些重複性的結構上又有i 他物體存在的話’則很可能會造成錯誤的修補結果。 2.該方法在影像大小超過_χ6⑽時的執行時間相當 長’而由於目則數位單眼相機所拍攝的數位影像總像素 數超過七百萬像素，所以造成目前實用上的困擾。、 如上所述，目前數位相機感光元件染塵而造成數位昭 片影像形成染塵缺陷的解決方法及其缺點如下： 1. 實際清除數位相機感光元件可能造成該感光元件二 次傷害或無法完全清除數位相機感光元件染塵； 2. 以影像處理軟體必須手動選取染塵缺陷區域，因 時費工，且處理的品質不佳； 3_以Criminisi等提出的「基於範例影像修補 ㈣mp — ed image inpainting)j 演算法之影像 雖可獲得良好的修補效果，但是可能因為修補時的^ = 99年11月10日修正替換頁 圍過廣導致修補許誤 锢錯為及修補時間過長的問題。 【發明内容】 為解决刖述既有改善相機感光元件染塵造成數位相片 本發明係-種相機感光元件塵染造成影 像缺的影像修復方法，其步驟包含： 取付已木塵相機之—染塵區域並根據 立一數位遮罩，其作法包含如下㈣： 域建 ，a中°亥：：：相機對—白色背景拍攝並取得-基礎影像 塵毕所1A Μ 子在°亥已染塵相機之内部感光元件因 歷木所k成的一染塵區域； 以一梯度運算定JI # 成-第-遮罩影像； ^像中染塵區域的邊緣並形 對該第-遮罩影像取— 二遮罩影像；以及 &進仃一值化而形成—第 對該第二遮罩影像 其甲該擴張運算係為_开二擴張運异並形成一數位遮罩， 係為幵n之擴張運算； 以該數位遮罩對該相機所 域加以標記；以及 數位相片之染塵區 以影像修補演算法修 ，該影修補演算法係改良自c疋成L己之染塵區域，其中 其改良在於Criminisi 寺徒出之肩异法， 的補丁區域後，於修補法取得-最高優先權 心擴張延展而定義出—目二=，以該補丁區域為中 心參考區域，以目標參考區域作 山6586 丁區域之參考 為修補該補 如此’在預先取得_ p ，再蔣、4 已木塵之數位相機之數位遮罩後 15亥數位遮罩套 得之數位相片》，制 數位相機所拍攝而 進一& 1 '用改良的Criminisi等提出之演算法 =像修補’即可達到良好的影像修補目的，而使 : 明具有如了優，點： 1定不發 必對該已染塵之數位相機進行其内部感光元件 塵區^以前述數位遮翠的方法可快速選出數位相片之染 3.以改良的Criminisi算接山λα —外 地修補該染塵區域。 ^出的以法可快速且正確 【實施方式】 請參考第一圖，係為本發明 . 只月之較佳實施步驟，其包含 .取得一已染塵相機之染塵區域並t 砜並建立—數位遮罩（10) 、以該數位遮罩對由該相機所拍攝 m ^ 數位相片之毕鹿F 域（2 0)加以標記，以及 χ 、 °° 〜像修補演算法修補已標記 之染塵區域（30) 。 ύ 請參考第二圖，前述取得一 Ρ ,九 侍已染塵相機之染塵區域並 形成一數位遮罩（10)，其包含牛锁· G3步驟.以該已染塵相機對 -白色背景拍攝亚取得一基礎影像（12)、以一梯产運曾 (gradient operation) $義該基礎影像之染塵區域的邊緣: 形成-第-遮罩料（⑷、對該第—遮罩影像取一臨界 1336586 99年11月1〇日修正替 值進行二值化而形成一第二遮罩影像（16)、對該第二遮 罩’?、/像進行擴張運真（dilation operation )並形成一數位遮 罩（1 8 )。 其中，以β亥已染塵相機對一白色背景拍攝並取得—基 礎影像（12 )之步驟中，先將該已染塵相機的光圈開到最 並對張全白的紙面，或對任何全白的物體進行拍攝 ’以取得該基礎影像。 以一梯度運异（gradient operation)定義該基礎影像之 染塵區域的邊緣並形成—第一遮罩影像（14)步驟中，由 於染塵缺陷在基礎影像令之亮度相對較小，因此可以利用 梯度運算將染塵區域的邊緣债測出來，其中，本較佳實施 步驟係採用—索柏遽波器（S*l filter)進行該梯度運算，之 取得基礎影像中之染塵區域之邊緣後的結果即為該 第一遮罩影像。 對該第-遮罩影像取一臨界值進行二值化而形成一第 m彡I (16)之步驟，係將該第—料影像進行二值 化處理’也就是先選定一臨界值（thresh〇id)後，對該第一遮 罩影像内的每一像素之灰階變化進行筛選，使超過該臨界 值的像素為王I，而低於該臨界值的像素為全白，該臨界 值可依據需求選取0〜255像數值之範圍内之一特定數值， 〇以本較佳實施步驟為例，係將該第一遮罩影 像取臨界值為27而形成該第二遮罩影像。 對該第二料f彡料行擴張運算（d— Gperation ) 並形成-數位遮罩（18)之步驟1 了避免臨界值設定過 12 1336586 保寸導T第二遮罩影像中的染塵區域範圍太小，以及 :了避免該'塵區域内可能存在的-雜點，本步驟（18)SSD = ^ , where 'heart, G々, and heart respectively represent the RGB trichromatic chromaticity value of a certain pixel P in the patch area (❼) of J high priority: and ~, %, & 6) The chrominance value of the inner pixel q. Among them, the patch block (%) with the smallest variance and the smallest value, that is, the patch area (^) with the first push is the most similar, and is set as the best patch block (indicated by 4 & After finding the best patch block (heart), the unfilled pixels for the highest priority patch area (❼) correspond to the last 13 8 1336586 99 November 1 〇曰 corrected replacement page good patch block (^) The pixel position of the ladder IΦ, one pixel can be pasted to one pixel. (4) Updating the trust value (c〇nfidence) of the previous repaired area is to give the C(p) value calculated in the step (2) to each of the pixels just patched as their trust value (). (5) If the target area Ω still exists, repeat steps (2) to (4) until the entire area is filled. However, the algorithm proposed by Criminisi has the following disadvantages: 1. Because the original decision method is Searching for the entire image to find a similar patch area to fill the target area, so if the patched area falls on the repetitive structure of the image, and there are other objects in these repetitive structures, then It is likely to cause incorrect patching results. 2. The method takes a long time to execute when the image size exceeds _χ6(10)', and the total number of pixels of the digital image taken by the digital single-head camera exceeds 7 megapixels. Practical troubles. As mentioned above, the current digital camera photosensitive element dusting and causing the digital camera image to form dust defects and its shortcomings are as follows: 1. Actually remove the digital phase The photosensitive element may cause secondary damage to the photosensitive element or may not completely remove the dust from the digital camera's photosensitive element; 2. The image processing software must manually select the dust-defective area, which is time-consuming and labor-intensive; 3_ According to Criminisi et al., "Image-based image patching (4) mp - ed image inpainting) j algorithm image can obtain good repair effect, but may be corrected due to the correction of the replacement page on November 10, 1999 The invention solves the problem that the repairing time is too long and the repairing time is too long. [Summary of the Invention] In order to solve the problem, the image fixing method for improving the image of the camera photosensitive element is caused by the dusting of the camera photosensitive member. The steps include: taking the dust-collecting area of the wood dust camera and according to the vertical digital mask, the method includes the following (4): domain construction, a medium temperature::: camera pair - white background shooting and acquisition - basic image 1A Μ 子 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在Image; ^ the edge of the dust-stained area is shaped to take the image of the first-mask image - the image of the second mask; and the & is formed by the value of the image - the second mask image is expanded by the second mask image The system is _open two to expand the difference and form a digital mask, which is the expansion operation of 幵n; mark the field of the camera with the digital mask; and the dust-removed area of the digital photo is repaired by the image patching algorithm. The shadow patching algorithm is improved from the dust-stained area of C, and its improvement is defined by the patch area of the Criminisi temple, which is defined by the patching method. - 目二 =, with the patch area as the center reference area, with the target reference area as the reference of the mountain 6586 area to repair the supplement so 'pre-acquired _ p, then Jiang, 4 has been the digital image of the digital camera The digital photo of the 15th digital mask after the cover is taken, and the digital camera shoots into a & 1 'the algorithm proposed by the modified Criminisi et al = repairing ' can achieve good image repair purposes, and make: Bright as Excellent, point: 1 will not send the dusty digital camera to the internal photosensitive element dust zone ^ by the above-mentioned digital hiding method can quickly select the digital photo dye 3. With the modified Criminisi calculation mountain λα — Repair the dusted area in the field. The method can be quickly and correctly [implementation] Please refer to the first figure, which is the invention. The preferred implementation step of the month only includes obtaining a dust-dyed area of a dust-collected camera and establishing a sulfone and establishing a digital mask (10) that marks the deer F field (20) of the m^ digital photo taken by the camera, and χ, °°~image repair algorithm to repair the marked dye Dust area (30). ύ Please refer to the second figure, the above for a glimpse, the nine servant has dusted the dust-proof area of the camera and form a digital mask (10), which contains the cattle lock · G3 step. With the dusted camera pair - white background Shooting a sub-image (12), using a gradient operation (edge operation), the edge of the dust-stained area of the base image: forming a -th-mask material ((4), taking the image of the first-mask A critical 1336586 on November 1st, 1999, the correction value is binarized to form a second mask image (16), and the second mask '?, / image is subjected to a dilation operation and formed. a digital mask (1 8 ). Among them, in the step of taking a white background and taking a basic image (12) with a β-ray dusting camera, first open the aperture of the dust-dyed camera to the most and open the white Paper surface, or photographing any white object to obtain the base image. Define the edge of the dust-stained area of the base image with a gradient operation and form a first mask image (14) In the light, the brightness of the basic image makes the relative brightness It is small, so the edge of the dust-dyeing area can be measured by the gradient operation. The preferred embodiment uses the S*l filter to perform the gradient operation, which is obtained in the basic image. The result of the edge of the dust-receiving area is the first mask image. The step of binarizing the threshold value of the first-mask image to form a m彡I (16) is the step of The image is binarized, that is, after selecting a threshold (thresh〇id), the grayscale change of each pixel in the first mask image is filtered, so that the pixel exceeding the threshold is Wang I, and the pixel below the threshold is all white, and the threshold value may be selected according to a specific value within a range of 0 to 255 image values, and the preferred embodiment is taken as an example. The mask image takes a critical value of 27 to form the second mask image. Step 2 of the second material f expansion operation (d-Gperation) and forming a digital mask (18) avoids threshold setting After 12 1336586, the size of the dust-absorbing area in the second mask image Too small, as well as : to avoid the possible miscellaneous points in the 'dust area, this step (18)
〇perati〇n)J :塵區域補滿二而完成本㈣（18)之擴張運算後，即可 冬。亥第一遮罩衫像形成該數位遮罩。 前述之以該數位遮罩標記由該相機拍攝之一數位相片 Γ:=Γ ’利用前述㈣（18)所形成的數位遮 即可將问由該已染塵相機所拍攝的一數位相片中的毕 i區域選取出來，而使該染塵區域内的像素均為該數位迻 罩之染塵區域之像素色度數值。 议遮 前述之以影像修補演算法修補標記之染塵區域（ ’係利用改良的C⑽inisi等所提出的「基於範例影 (eXemPlar_baSed i卿_ —)」演算法對該數位相;的 染塵區域進行修補，其中，為了避免傳統⑽心提片出= 「基於fe例影像修補（exempUr_based ―丨如 算法於修補影像時係以整張數位相片為參考範圍 致該參考範圍過廣而致使修補錯誤及修補時間過長= ’本步驟〇〇) miminisi演算法取得修補區域中 最高優先權的補丁區域之後，再進一步定義以該補丁 為中心所擴張而出之一目標參考區域（RR)以作 丁 區域選取填補材質的參考範圍，如第三圖所示，如此， 可同時避免修補錯誤及修補時間過長的問題。其中，、’即 步驟（30)為例，係將該目標參考區域設定為該=本 的的6至8倍。 品域 1336586 99年11月10日修1替換頁~η °月配σ參考第四圖，其係為本較佳實施步驟以及以傳 統criminisi演算法修補同一影像的修補速度比較由該圖 可知處理影像解析度較夫沾奴， 大的數位照片，本較佳實施步驟比 傳統Criminisi演算法右知a „ #亡有相畲優異的處理速度表現。 【圖式簡單說明】 第-圖為本發明之較佳實施步驟之流程圖。 第一圖為本發明之取得一已染塵相機之染塵區域並形 成一數位遮罩之步驟流程圖。 第三圖為本發明之以Cdminisi提出之演算法所改良的 演算法。 第四圖為本發明之改良的Criminisi演算法與傳統 Criminisi演算法的處理時間比較圖。 第五圖係為傳統CHminisi演算法的示意圖。 【主要元件符號說明】 (1 〇 )數位遮罩 （ (14)第一遮罩影像 （ (1 8 )數位遮罩 12 )基礎影像 16 )第二遮罩影像 (20)染塵區域 (30)染塵區域〇perati〇n)J: After the dust area is filled up and the expansion of this (4) (18) is completed, it can be winter. The first mask of the Hai is like the digital mask. The digital mask is used to capture a digital photo taken by the camera: Γ 'The digital mask formed by the above (4) (18) can be used to ask a digital photo taken by the dusted camera. The area i is selected, and the pixels in the dust-receiving area are the pixel chromaticity values of the dust-removed area of the number displacement cover. It is proposed to cover the dust-removed area with the image patching algorithm repair mark ('Using the modified image (eXemPlar_baSed iqing_)) proposed by the modified C(10)inisi et al. Patching, in order to avoid the traditional (10) card release = "based on the fe image repair (exempUr_based - such as the algorithm used to repair the image with the entire digital photo as the reference range caused by the wide range of reference caused patching errors and repairs The time is too long = 'this step〇〇') After the miminisi algorithm obtains the highest priority patch area in the patched area, it further defines a target reference area (RR) that is expanded with the patch as the center to select the area. Fill in the reference range of the material, as shown in the third figure, so that the problem of repairing errors and repairing too long can be avoided at the same time. Among them, 'step (30) is an example, the target reference area is set to the = 6 to 8 times of this. Product 1336586 November 10, 2010 Revision 1 Replacement page ~ η ° monthly σ reference to the fourth figure, which is the preferred implementation step and the traditional c The riminisi algorithm fixes the repair speed of the same image. It can be seen from the figure that the resolution of the processed image is better than that of the husband and the large digital photo. The preferred implementation step is better than the traditional Criminisi algorithm. The following is a flow chart of a preferred embodiment of the present invention. The first figure is a flow chart of the steps of obtaining a dust-absorbing area of a dust-collected camera and forming a digital mask. The third figure is an improved algorithm of the algorithm proposed by Cdminisi in the present invention. The fourth figure is a comparison of the processing time between the improved Criminisi algorithm and the traditional Criminisi algorithm of the present invention. Schematic diagram of CHminisi algorithm. [Main component symbol description] (1 〇) digital mask (14) first mask image ((1 8 ) digital mask 12) base image 16) second mask image (20) Dust-removed area (30) dust-receiving area