201117134 六、發明說明： 【發明所屬之技術領域】 影像處理系統， 及系統。 本發明係關於一種影像處理方法以及 特別是關於複數影像之間的影像處理方法 【先前技術】 在各種多媒體影像的相關應用上，存在著許多影像内容 的辨識輯需求’如賴物體麵或其差異性、印刷電路板 (簡稱PCB)的電路檢測、影像監視防盜或立體⑼㈣)影像的 建立與處理等。 /影像關連性(image correlation)比對是一種廣泛應用於 衫像處理領域令的技術，用以分辨兩張影像之間的相似程 度面低。許多新穎的視訊編碼電路（例如與Η.26χ或 mpeg協定相容的系統)通常會採用影像關連性比對或移 動向置估測等方法來協助於不同影像晝面中找出圖像間的 相似性’進而達到各種影像應用功能。 請參閱圖一 A，圖一 A繪示習知的影像關連性比對 方法下第一影像1〇與第二影像12之間的關連性示意圖。 此第一與第二影像在實際應用中可能為影像串流中的時序 先、後影像，或雙視角立體攝影系統同時產生的兩幅影像 等。在習知的影像關連性比對方法中，主要透過特徵區塊 比對的方法來找出兩影像間的關連性。 如圖一所示，在第一影像10中掏取複數個特徵點 201117134 100 ’在第二影像12亦擷取複數個特徵點12 所神 的=徵點’大多對應晝面中的色彩或亮度變化較大的^ 1 ’或是晝面中各個物件的邊緣。特徵點的梅取可採用 轉換(scale invariant featoe tran〜 / & 氏角偵測(Harris c〇mer detecti〇n)演算法。 〇接者’第—影像1G中的每—個特徵點1QG及其鄰近的 塊102，第二影像12中的每一個= =嶋大小可為㈣、㈣、胸二素兄單= 第-影像ίο中的特徵區塊1〇2與第二影像12中的特徵 =塊122以區塊為單位進行輯，以全面搜尋或部份搜尋 :方iff其最相似的一組具對應關係的特徵區塊配 、藉由第一影像10中的特徵區塊102與第二影像12中201117134 VI. Description of the invention: [Technical field to which the invention pertains] Image processing system, and system. The present invention relates to an image processing method and, in particular, to an image processing method between complex images. [Prior Art] In various related applications of multimedia images, there are many image content recognition requirements, such as the object surface or its difference. Circuitry, printed circuit board (PCB) circuit detection, image surveillance security or stereo (9) (four)) image creation and processing. /image correlation is a technique widely used in the field of shirt image processing to distinguish the low degree of similarity between two images. Many novel video encoding circuits (such as those compatible with the χ.26χ or mpeg protocol) typically use image correlation or mobile orientation estimation to assist in finding images between different image planes. The similarity' further achieves various image application functions. Please refer to FIG. 1A. FIG. 1A is a schematic diagram showing the relationship between the first image 1〇 and the second image 12 in the conventional image correlation comparison method. The first and second images may be sequential first and last images in the video stream or two images simultaneously generated by the dual-view stereo camera system. In the conventional image correlation comparison method, the correlation between the two images is mainly found by the method of comparing the feature blocks. As shown in FIG. 1, a plurality of feature points 201117134 100 'in the first image 10 are captured. In the second image 12, a plurality of feature points 12 are also captured. The sign of the god is mostly corresponding to the color or brightness in the facet. The larger change ^ 1 ' or the edge of each object in the face. The feature point of the feature can be converted (scale invariant featoe tran ~ / && Angle detection (Harris c〇mer detecti〇n) algorithm. 〇接者's - image 1G each feature point 1QG and In the adjacent block 102, each of the second images 12 ==嶋 can be (4), (4), the chest is two singular = the characteristic block in the first image 〇2 and the features in the second image 12 The block 122 performs a series search in units of blocks to perform a full search or a partial search: the most similar set of feature blocks of the square iff, with the feature block 102 and the first image in the first image 10 Two images in 12
^特徵區塊122之間的相對關係，判斷兩個影像之間 連性。 ，如:習知的全面搜尋演算法(full记虹吐a丨卯汾hm) :’第#像10中的每一個特徵區塊1〇2會被拿來和第 f影J 12中所有可能的特徵區塊122逐—比對，以便由 ，一’5V像12j逐一比對找出最相似的特徵區塊1M。也 就因為’目前f見的比對機制通常採用這種區塊比對 (block_based)方式，一般而言，當第一影像10與第二影像 12之間存在程度的水平誤差與垂直誤差等晝面差距 % ’現階㈣f彡像騎性轉方法健能賊到最適切的 201117134 對應關係。 然而’當第—影像與第二影像之間若存在旋轉 Mat10nal dlscrepancy)時，請參閱圖一 b 、 知的影像關連性比對方法下第一影像1〇，旦f、、會不習 間的關連性示意圖。如圖一 b所示，第 m的像素之間存在旋轉角度上的相對偏差而不、匕 水=或垂直方向的娜，如此—來，—般的區塊 便…法找到可形成對應的特徵區塊配對 生时番 _]情形。在實_中，若偏_度在 過習知的比對方法仍村能轉較高的正 :，若取兩個完全相同的影像，將其卜張影像:：’巧 知比對方法的比對正確率已下降至 = :而當鋪角度達3〇度時，正確率更大幅下降至 =疋’本㈣提種影像處理方法以 統，其可校正影像之_旋轉偏差 間之影像關連性，以解決上述問題。 確也_衫像 【發明内容】 本u之範笮在於提供一種影理 數影像之_料_。輯^ 別包含減讎素，顿雜纽方純钉辭驟等私像刀 該第二影像之該等像素中選取複數個特徵點。且由 2〇Πΐ7ΐ34 鄰、斤根據該第一影像輿該第二影像之每牲 #近_物應每1徵點上的-梯度變Γ向量徵點及其 3) 分難據該第一影像鱼 里 等梯度向量，對顧第」影触轉徵點的該 统計分佈。 / ~心第一影像各自產生— 4) 根據該等統計分佈 間的一偏轉角度。 〜弟衫像與該第二影像之 本發明之另一範疇在於提 具體實峰影像處餘統。根據一 像比對模組。 轉偏差修正模組以及影 之間二=差：二=斷第-影像與第二影像 素，修正模 :====擷= —影像之該等像素中選取複數個特 Μ第 別根據該第-影像梯料异单元分 像素機應上變=及㈡的 =====，=像之該等特徵點的該等梯 度支化^ 影像與該第二影像各自產生 計分佈。該角度計算單元根據該等統計分佈，用曾兮 第-影像與該第二影像之間的該偏轉角度。如此一來了= 轉偏差修正模組便可據以消除影像之間的旋轉偏差。 201117134 ί後該影像輯模組便可就轉偏差修正模組調整後 之該第#像與该第二影像進行關連性比對，藉此判斷 -影像與該第二影像之間的影像關連性。 相較於1知技術_直接對兩影像進行區塊比對的作法， 於本發明可預先計算兩影像之_偏轉角度，並對影像進行 旋轉補償’方才進行影像之間的關連性比對。藉此，提高影 像關連性轉的效輪準確性，並降個影像間具有角度 誤差而產生誤判或根本無法對應的情形。 關於本發明之優點與精神可以藉由以下的發明詳述及 所附圖式得到進一步的瞭解。 【實施方式】 «月參閱圖一，圖二繪示根據本發明之一具體實施例中 衫像處理糸統2的示意圖。如圖二所示，影像處理系統2 包含旋轉偏差修正模組20以及影像比對模組22。影像處 理系統2可用以判斷及比對不同的影像檔案之間的影像關 連性。此處影像處理系統2處理的影像可為影像串流中的 時序先、後影像，或雙視角立體攝影系統同時產生的兩幅 影像等。 請參閱圖三，圖三緣示根據本發明之一具體實施例中 第一影像30與第二影像32的示意圖。於此實施例中，影 像處理系統2適用於處理第一影像30與第二影像32。在 實際應用中’在影係檔案的形成過程中（例如攝影或拍照） 可能因為產生影像的拍攝裝置發生晃動、雙視角立體攝影 201117134 系統的左右攝影鏡頭角度不一致等各種原因，第一与 30與第二影像32所呈現的晝面經常可能具有_定的= 偏差’如®三所示’第—影像3G與第二影像幻 二 存在相對的偏轉角度Θ。 双 於此實施例中，本發明中的旋轉偏差修正模組 用^計算第—影像3G與第二影像32之間的偏轉角度二 接著，影像比對模、组22可根據旋轉偏差修正模組2〇又^ The relative relationship between the feature blocks 122 determines the connectivity between the two images. Such as: the comprehensive search algorithm of the known (full record rainbow spit a丨卯汾hm): '## every feature block in 10 like 1 will be taken and all the possible in the fth shadow J 12 The feature blocks 122 are aligned to each other so that a '5V image 12j is aligned one by one to find the most similar feature block 1M. That is to say, because the comparison mechanism currently seen is usually adopted in this block-based manner, in general, when there is a degree of horizontal error and vertical error between the first image 10 and the second image 12, Face difference % 'current order (four) f彡 like riding performance method fitness thief to the most appropriate 201117134 correspondence. However, if there is a rotation of Mat10nal dlscrepancy between the first image and the second image, please refer to Figure 1b. The first image under the image correlation method is 1〇, 旦f, and will not be used. Diagram of relevance. As shown in Fig. 1b, there is a relative deviation between the pixels of the mth, but not the water, or the vertical direction, so that the block can be found to form a corresponding feature. The block pairing time _] situation. In the real _, if the partial _ degree is in the conventional comparison method, the village can still turn higher: if two identical images are taken, the image will be imaged: The accuracy of the comparison has dropped to = : and when the angle of the pawn reaches 3 degrees, the accuracy rate drops sharply to = 疋 'this (4) to improve the image processing method, which can correct the image correlation between the rotation deviation of the image To solve the above problem. Indeed also _ shirt image [invention] This u is to provide a kind of image data. The series ^ includes the sputum, the singularity of the singularity, and the like. The second image of the second image selects a plurality of feature points. And according to the first image, the second image of each of the second images of the second image should be a gradient of the vector sign and the 3rd points of the first sign. Gradient vector in fish, this statistical distribution of Gu Di's signing points. / ~ The first image of the heart is generated - 4) according to a deflection angle between the statistical distributions. ~ Another aspect of the invention of the shirt image and the second image is to mention the details of the real peak image. According to the image comparison module. The deviation correction module and the shadow between the two = difference: two = off the first image and the second image, the correction mode: ====撷 = - the image of the pixels selected a plurality of features according to the The first-image ladder different-unit pixel-splitting machine should be up= and (2) =====, = the gradient branching image of the feature points and the second image respectively generate a gauge distribution. The angle calculation unit uses the deflection angle between the first image and the second image according to the statistical distribution. In this way, the deviation correction module can eliminate the rotation deviation between images. After the 201117134 ί, the image editing module can perform the correlation comparison between the first image and the second image adjusted by the deviation correction module, thereby determining the image correlation between the image and the second image. . Compared with the prior art, the block alignment is directly performed on the two images. In the present invention, the _deflection angle of the two images can be calculated in advance, and the image is rotated and compensated for the correlation between the images. Thereby, the accuracy of the effect wheel of the image related rotation is improved, and the angle error between the images is reduced to cause misjudgment or no correspondence at all. The advantages and spirit of the present invention will be further understood from the following detailed description of the invention. [Embodiment] «Month Referring to FIG. 1, FIG. 2 is a schematic view showing a shirt image processing system 2 according to an embodiment of the present invention. As shown in FIG. 2, the image processing system 2 includes a rotation deviation correction module 20 and an image comparison module 22. The image processing system 2 can be used to determine and compare image relationships between different image files. The image processed by the image processing system 2 herein may be a sequence of first and last images in the video stream, or two images simultaneously generated by the dual-view stereo camera system. Referring to FIG. 3, FIG. 3 is a schematic diagram showing a first image 30 and a second image 32 according to an embodiment of the present invention. In this embodiment, the image processing system 2 is adapted to process the first image 30 and the second image 32. In practical applications, during the formation of the film archives (such as photography or photographing), the shooting device that generates the image may be shaken, the angle of the left and right camera lenses of the dual-view stereo camera 201117134 system is inconsistent, and so on. The facet presented by the second image 32 may often have a _set = deviation 'as shown in the 'three', the first image 3G and the second image phantom have a relative deflection angle Θ. In this embodiment, the rotation deviation correction module of the present invention calculates the deflection angle between the first image 3G and the second image 32. Then, the image comparison mode and the group 22 can be modified according to the rotation deviation module. 2〇又
而得的該偏轉角度θ，先對其中一個影像進行旋轉補十^ fotatumal 〇ffset)後，接著對第一影像2〇與第二影像u 盯關連性比對，藉此判斷該第—影像與該第二影像之間 像關連性。相較於習知技術巾直接對兩影像進行比ς 法，於本發明中，在預先經過旋轉補償之後，第一影像2〇 * 第=影像22之__性比對準確性便能大幅提高= 提高影像Ifl連性比對的效率，鱗低關 ^ 關係的情形。 關於此實施例中，旋轉偏差修正模組20計算第一影像 30與第二影像32之間的偏轉角度θ的作動方式與方法步 驟，將詳述如下。於此實施例中，旋轉偏差修正模組2〇 ^ 含特徵點擷取單元200、梯度計算單元2〇2、統計單元 2〇4、角度計算單元寫以及影像旋轉單元观。請一併象 閱圖四’圖赠示根據本發明之—具體實施例中影像處理 方法的方綠侧’影像歧方法可配合在影像處理系統2 上執行。 於此實施例中’第-影像3Q與第二影像32分別包含 201117134 複數個像素’例如640x48〇或102知768個像素等等，但 不以此為限，視實際影像大小而定。 ^首先，執行步驟S100，利用特徵點擷取單元200自第一 y像3〇之該等像素_選取複數個特徵點3〇〇(如圖三戶斤示）， 並自該第一衫像32之該等像素中選取複數個特徵，點32〇(如圖 ，所不）’於此實施例中，特徵點擷取單元2〇〇可基於尺度不 I特徵轉換(seale limnant featufe transform，SIFT)演算法或 =氏角债測(Harris comer detecti〇n)演算法擷取該等特徵點， 逆些特徵點(第-影像30之特徵點·、第二影像32之特徵 =320)大致上主要分佈影像中色彩彩度、亮度或灰階值變化 較大的地方，或是分佈在晝面巾物件的邊緣處。 / ---- ’丨'<叫仰反叮异早兀21^針對第一 像30根據母特徵點3⑻及其鄰近的像素，計算對應每一 特徵3 3^0上的梯度變化向量。於此同時，梯度計算單元 ^2針對第一影像32根據每一特徵點汹及其鄰近的像 算對應每-特徵點32〇上的梯度變化向量。 … 斤。需特別進一步說明的是，在上述步驟sl〇2當中，梯度計 算单元施係大致以每-特徵點，或特徵點32()為^ 點由第-影像30與第二影像32中各自選取鄰近特徵點 3—〇〇或特徵點320的複數個像素形成特徵區塊。舉例來說‘、、， 母個特徵區塊可為以其中—個特徵點為中心且由i7*i7個 ^斤共_成的區塊’但本發明並不以此為限’特徵區塊的 ^可視計算上的枝性、處理騎計算能力、使 求的處理時限等因素而定。接著，梯度計算單元搬可根據 201117134 徵點 每一 塊内的所有像素，計算產生該特定特徵點(特 或特徵點32G)上的梯度變化向量。於此實 =度變化向量包含了強度變化量以及變化肖度。Μ ^化量係指該特徵區塊内像素的色彩彩度變化量、色衫 ::否特=:::= =j，於此同時，利用統計單元綱根據第二 =等特徵點320的梯度變化向量，產生第二影像32之統計分 第一二俊彳=五，圖五％示根縣發明之—具體實施例 之統計八欲齡^，知驟S1。4 #中，所謂第—影像30 中具有1目_生’主要是_第—影像3Q的梯度變化向量 不’同變化的特徵點之強度變化量累加，形成依照 生，係將該第二聲像32祕/—衫像32之統計分佈的產 度的特徵上;梯度變化向量中’具有相同變化角 .，'強度、交化罝累加所形成的(如圖五所示）。 ，像H執行步驟S1G6 ’利用角度計算單元观根據第一 ^像32 摊32的崎分佈，計算第—影像3G與第二 衫像32之間的偏轉角度Θ大小。 需特別進—步說明的是，在上述步驟S106當中’角 11 201117134 該第—影像3G之統計分佈中選取對應最 最大變化量角度度，糾實施例中」 算單元206亦由該第二$像 不）’另一方面，角度计 大的強度變化累積值十分佈中選取對應最 爭女變π旦&庙&的取大變化I角度，於此實施例中， 算上比第二^ \!4 (如圖五所示)。角度計算單元206計 表大變化量角度(53。)與該第二影像之最 例如於值’作為兩影像間的偏轉角度θ， 例如於此貝施例中，偏轉角度θ即為11〇〇 仏當得知第一影像3〇與第二影像％之間的偏轉角度θ之 像2便可對其中—張影像進行旋轉補償。舉 像處理方法，接著便可執行步驟麗， 根據計算而得的偏轉角度θ，先對第 二隨後影像比對模組22便可對旋轉 判斷筮、^7 ’、第—影像3〇進行關連性比對，藉此 判斷弟-4 30與第二影像32之間的影像關連性。 需進-步說日㈣是，在上述步驟㈣中，旋轉偏差 彡像旋轉單元期建立一個二維旋轉矩陣 (於此實％例中此二維旋轉矩__ ί 弟二影像％的像素簡内容乘上此二維旋轉矩 I t紅影像32逆時針旋轉U。之後，隨後影像比對 表t ^即可對第一影像3〇與旋轉後的第二影像32進行關 藉此，便可判斷第一影像％與第二影像32在 去除㈣偏差情況下的影像_性。也就是說， 201117134 偏差修正模組2〇所計算而得的偏轉 模組22可以較高的精確度進行後續 ^小’影像比對 影像塵縮或其他影像處理上的判斷步驟。理、物件辨視或 可預Γ明之影像處㈣統與影像處理方法， 預先计异兩衫像之間的偏轉角度，並對 、 方才進行影像之間的關連性比對。藉此，提== 而Ϊ = 確性，並降低因影像間具^誤差 向產生决判或根本無法對應的情形。 …藉由〖X上較佳频實關之弱，鱗雜更加清 描述本發日狀特徵與精神，而並非以上述所揭露的較^具 體實施例來對本發明之範疇加以限制。相反地，其目的& 希望能涵1各種改變及具相等性的安排於本發明所欲申= 之專利範圍的範疇内。 μThe obtained deflection angle θ is first rotated to compensate for one of the images, and then the first image 2〇 is aligned with the second image u, thereby determining the first image and The image is related to the second image. Compared with the conventional technology towel, the two images are directly compared, and in the present invention, the accuracy of the first image 2〇* the second image 22 can be greatly improved after the rotation compensation is performed in advance. = Improve the efficiency of the image Ifl connection, the situation of the scale is low. In this embodiment, the rotation deviation correction module 20 calculates the operation mode and method steps of the deflection angle θ between the first image 30 and the second image 32, which will be described in detail below. In this embodiment, the rotation deviation correction module 2 includes a feature point extraction unit 200, a gradient calculation unit 2〇2, a statistical unit 2〇4, an angle calculation unit write, and an image rotation unit view. The square green side image discrimination method of the image processing method according to the present invention may be performed in conjunction with the image processing system 2 as shown in FIG. In this embodiment, the first image 3Q and the second image 32 respectively include 201117134 plural pixels, such as 640x48〇 or 102 768 pixels, etc., but not limited thereto, depending on the actual image size. ^ First, step S100 is executed, and the feature point extracting unit 200 selects a plurality of feature points 3 from the pixels of the first y image 3 (as shown in Figure 3), and from the first shirt image. A plurality of features are selected from the pixels of 32, and the point 32 〇 (as shown in the figure). In this embodiment, the feature point extraction unit 2 〇〇 can be based on a scale non-I feature transform (SIFT). The algorithm or the Harris comer detecti〇n algorithm captures the feature points, and the feature points (the feature points of the first image 30 and the features of the second image 32 = 320) are substantially reversed. The color distribution, brightness, or grayscale value of the main distribution image changes greatly, or is distributed at the edge of the face towel object. / ---- 丨 & & 叫 ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ At the same time, the gradient calculating unit ^2 calculates a gradient change vector corresponding to each of the feature points 32A according to each feature point 汹 and its adjacent image for the first image 32. ... jin. It should be further noted that, in the above step sl2, the gradient calculation unit applies the proximity of each of the first image 30 and the second image 32 by using a per-feature point or a feature point 32(). A plurality of pixels of feature point 3 - 〇〇 or feature point 320 form a feature block. For example, ',,, the parent feature block may be a block that is centered on one of the feature points and is composed of i7*i7, but the invention is not limited thereto. The visibility of the visual calculation, the ability to handle riding, and the processing time limit required. Then, the gradient calculation unit can calculate the gradient change vector on the specific feature point (special feature point 32G) according to all the pixels in each block of the 201117134 sign. The actual degree change vector contains the intensity change and the change profile. Μ ^ Quantization refers to the color chroma change of the pixels in the feature block, the color shirt:: No special =::: = = j, at the same time, using the statistical unit according to the second = equal feature point 320 Gradient change vector, the statistical result of the second image 32 is generated as the first two 彳 彳 = five, Figure 5 is the invention of the invention of the Shige County - the statistics of the specific embodiment of the eight ages, the knowledge of S1. 4 #, the so-called - In the image 30, the gradient change vector of the 1st image is mainly _first image 3Q, and the intensity change amount of the feature point that does not change is accumulated, and the second sound image 32 is formed according to the life. The characteristic of the statistical distribution of 32 is statistically significant; in the gradient change vector, 'the same angle of change.' is formed by the accumulation of intensity and cross-linking (as shown in Figure 5). The H-execution step S1G6' uses the angle calculation unit to calculate the deflection angle 之间 between the first image 3G and the second shirt image 32 based on the sagittal distribution of the first image 32. In particular, in the above step S106, the statistical distribution of the first image 3G of the angle 11 201117134 is selected to correspond to the maximum maximum amount of change angle, and in the embodiment, the calculation unit 206 is also determined by the second $ Like no) on the other hand, the cumulative value of the intensity variation of the angle meter is very large, and the angle of the largest change I is selected for the most popular female π dan & temple & in this embodiment, it is calculated as the second ^ \!4 (shown in Figure 5). The angle calculation unit 206 counts the large change angle (53.) and the second image, for example, the value 'as the deflection angle θ between the two images. For example, in this example, the deflection angle θ is 11〇〇. When the image 2 of the deflection angle θ between the first image 3〇 and the second image % is known, the image can be rotated and compensated. After the image processing method, the step 丽 can be executed, and according to the calculated deflection angle θ, the second subsequent image comparison module 22 can be related to the rotation determination 筮, ^7 ', and the first image 3 〇. Sexual comparison, thereby determining the image correlation between the brother-4 30 and the second image 32. The step-by-step (4) is that, in the above step (4), the rotation deviation image is rotated to obtain a two-dimensional rotation matrix (in this case, the two-dimensional rotation moment __ ί The content is multiplied by the two-dimensional rotation moment I t red image 32 is rotated counterclockwise U. Then, the image comparison table t ^ can be used to close the first image 3 〇 and the rotated second image 32. Determining the image_sexity of the first image % and the second image 32 in the case of removing the (four) deviation. That is, the deflection module 22 calculated by the deviation correction module 2〇 can perform higher accuracy subsequently. Small 'image comparison step on image dust reduction or other image processing. Reason, object recognition or predefinable image (4) system and image processing method, pre-difference between the two shirt images, and Only then can the correlation between images be compared. By doing this, == and Ϊ = positivity, and reduce the situation that the decision is made or the image cannot be matched due to the error between the images. The frequency is weak, and the scales are more clear. The scope of the present invention is not limited by the specific embodiments disclosed above. Conversely, the purpose & hopes to include various changes and equal arrangements in the present invention. Within the scope of the patent scope. μ
13 201117134 【圖式簡單說明】 圖一 A緣示習知的影像關連性比對方法下第一影像 與第二影像之間的關連性示意圖。 圖一 B繪示習知的影像關連性比對方法下第一影像 與第二影像之間的關連性示意圖。 圖二繪示根據本發明之一具體實施例中影像處理系統 的示意圖。 圖三繪示根據本發明之一具體實施例中第一影像與第 二影像的不意圖。 圖四繪示根據本發明之一具體實施例中影像處理方法 的方法流程圖。 圖五繪示根據本發明之一具體實施例第一影像與第二影 像分別產生的統計分佈不意圖。 【主要元件符號說明】 10、1(V、30 :第一影像 12、12’、32 :第二影像 100、120、300、320 :特徵點 102、122 :特徵區塊 2 :影像處理系統 20 :旋轉偏差修正模組 200 :特徵點擷取單元 204 :統計單元 202 :梯度計算單元 201117134 206 :角度計算單元 208 :影像旋轉單元 22 :影像比對模組 Θ :偏轉角度 S100〜S108 :步驟13 201117134 [Simple description of the diagram] Figure 1 shows the relationship between the first image and the second image under the image correlation comparison method. FIG. 1B is a schematic diagram showing the relationship between the first image and the second image under the conventional image correlation comparison method. 2 is a schematic diagram of an image processing system in accordance with an embodiment of the present invention. Figure 3 is a schematic illustration of a first image and a second image in accordance with an embodiment of the present invention. 4 is a flow chart of a method of an image processing method in accordance with an embodiment of the present invention. Figure 5 illustrates a statistical distribution of the first image and the second image, respectively, in accordance with an embodiment of the present invention. [Description of main component symbols] 10, 1 (V, 30: first image 12, 12', 32: second image 100, 120, 300, 320: feature points 102, 122: feature block 2: image processing system 20 : Rotational deviation correction module 200: feature point extraction unit 204: statistical unit 202: gradient calculation unit 201117134 206: angle calculation unit 208: image rotation unit 22: image comparison module Θ: deflection angle S100~S108: steps
1515