200815909 九、發明說明： 【發明所屬之技術領域】 本發明係有關一種影像對位方法，特別是一種利用相位關 連法（phase correlation)處理之影像對位方法。 【先前技術】BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image registration method, and more particularly to an image alignment method using phase correlation processing. [Prior Art]
按，顯微技術(microscopy)自發明以來經歷多年演變，在 人類科技文明的發展上已做出極大的貢獻，近十餘年來由於高 性能個人電腦的快速發展，更促成了顯微技術的成熟與應用。 而利用光學切片（斷層掃描技術)和數位影像重組的技術，達成 立體的顯微斷層掃描在各領域也已造成極大的影響。 以牙透式電子顯微鏡（transmission electron microscope, 或X光顯微鏡(x_ray micr〇sc〇pe)而言，拍攝待重建物體 ^須針對物體進行不同投射角度的影像組合，進而結合產生立 =的資料以及影像，但是立體的資料和影像需要正確的影像對 的始影像的位移乃是由於從不同的角度拍攝產生機械上 用^二缺陷’或因為溫度差造成的位置移動所造成。習知利 失對位:不僅是時間上的浪費，也容易產生人為疏 解決人工對二：方:係利：互相關法(cross c_lat·)來 角度所拍攝的物體鸟像二::重要的議題在於’針對不同 法即無法完全克服王相同的，而習知的影像對位方 【發明内容】 鑒於上述問題， 决，利用相位關連演 ，本發明目的之—係提供_種影像對位方 算法能快❹鮮销算㈣像之位移。 5 3 200815909 —本發明目的之-係提供—種影像對位方法，㈣相位 々异法能在#貞料完全相同影㈣情況下，達絲像對位的 的 —^發=目的之-係提供—種影像對位方法，利用相位關連 次异法可處理掉大部分的雜訊區域，藉由雜訊 的對位更為精確。 双便如像Microscopy has undergone many years of evolution since its invention, and has made great contributions to the development of human science and technology civilization. Over the past decade, the rapid development of high-performance personal computers has contributed to the maturity of microscopy. And application. The use of optical sectioning (tomographic scanning) and digital image recombination technology to achieve stereoscopic microscopic tomography has also had a great impact in various fields. In the case of a transmission electron microscope (X-ray micr〇sc〇pe), the image to be reconstructed must be combined with an image of different projection angles for the object, and then combined with the data generated and Image, but stereoscopic data and images require the correct image pair. The displacement of the initial image is caused by the mechanical use of the two defects or the positional movement caused by the temperature difference. Bit: Not only is it a waste of time, but it is also prone to artificially solve the problem of artificial pairing. 2: Fangli: cross-correlation method (cross c_lat·). The angle of the object taken by the bird 2:: The important issue is 'different The method can not completely overcome the same king, and the conventional image alignment side [invention] In view of the above problems, the use of phase correlation, the purpose of the present invention is to provide an image alignment algorithm that can quickly The calculation of (4) the displacement of the image. 5 3 200815909 - the purpose of the present invention is to provide a method of image alignment, (4) the phase difference method can be completely identical in the #贞 material (4) Under the circumstance, Dass's image-alignment method is provided by the alignment of the image, and the phase correlation method can be used to dispose most of the noise region, and the alignment of the noise is further improved. For the sake of precision.
ιί發!目的之—係提供—種影像重建方法，利用相位關連 异之影像對财法，可解決錢層掃 旋轉震動所造成的對位問題。 機口本身 —本發Γ目的之一係提供一種影像重建方法，利用相位關連 演异法運算之影像對位方法，可解決在數位取像裝 ， 因手震影響相片品質的問題。 攝守 、為了達到上述目的，本發明一實施例之一種影像對位方 法，包括：操取至少二影像；計算相鄰兩影像之—相對位移， 其中相對位移係利用_相位關連演算法運算得之；計算每一影 ，與影像中之第—張影像的—絕對位移；以及利用相對位移與 絶對位移计异出影像之—共同區域，並移除共同區域之外的影 像。 、 為了達到上述目的，本發明另一實施例之一種影像重建方 法包括·擷取至少二影像；計算相鄰兩影像之一相對位移， 其中相對位移係利用一相位關連演算法運算得之；計算每一影 像與影像中之第一張影像的一絕對位移；利用相對位移與絕對 位私。十#出衫像之一共同區域，並移除共同區域之外的影像； 決定影像之旋轉中心；以及重建影像之立體資料。 為了達到上述目的，本發明又一實施例之一種影像重建方 套匕括·操取至少二影像；計算相鄰兩影像之一相對位移， 6 200815909 其中相對位移係利用一相位關連演算法運算得之；計算每一影 像與影像中之第一張影像的一絕對位移;利用相對位移與絕對 位移計算出影像之一共同區域，並移除共同區域之外的影像； 以及疊加每一影像計算出之共同區域。 底下藉由具體實施例配合所附的圖式詳加說明，當更容易 瞭解本發明之目的、技術内容、特點及其所達成之功效。 【實施方式】 其詳細說明如下，所述較佳實施例僅做一說明非用以限定 本發明。 第1圖所示為根據本發明影像對位方法一實施例之步驟 流程圖。如圖所示，首先，由拍攝好的影像中擷取至少二影像 S10，其中，依據不同應用，這些影像可以是針對相同物體同 一角度拍攝，亦可是針對不同角度拍攝而成；接著，計算相鄰 兩影像之一相對位移S20，其中相'對位移係利用一相位關連演 算法(phase correlation algorithm)運算得之；再來，計算每一影 像與影像中之第一張影像的一絕對位移S30，其中絕對位移之 計算亦是利用相位關連演算法運算得之，或是，利用任兩張影 像所求得之相對位移進而計算出每一張影像與第一影像的絕 對位移；最後，利用相對位移與絕對位移計算出所有影像之一 共同區域，並移除共同區域之外的影像S40。於一實施例中， 共同區域之判定係基於相對位移之位移量，若影像有過大位移 量，例如：超過兩倍均方根值，則去除。找出共同區域後，將 所有共同區域之外影像去除，以完成影像對位的步驟。 接續上述說明，在計算影像之相對位移或絕對位移前，更 包括針對影像進行一影像前處理步驟。於一實施例中，影像前 處理步驟係包括對影像做銳化處理、平滑化處理與去雜訊處理 200815909 之至少其中之任一。其中影像前處理步驟有助於所拍攝的影像 處理一些不必要的雜訊，亦或是，幫助影像的訊號強化，以增 強後續影像對位甚至是影像重建的正確性。 於一實施例中，計算出上述相對位移或是絕對位移的方 法’可以疋利用傅立葉函數轉換(J7〇urier transf〇nil)或快速傅立 葉函數轉換(Fast Fourier transform，FFT)及其運算所得之。其 運算方法如下方程式(I)所示：首先，須將欲運算之兩影像，例 如pi及p2，分別作傅立葉轉換而得兩數值F[pi]及F[p2];接 著’計算兩張影像之相關（correlation)，亦即，取其中一張影 0 像與另一張影像的共輛複數（complex conjugate)相乘，如 F[pl](F[p2])1 ;再來，將上述之值除以兩張影像之絕對值，如 (F[Pl](F[P2])1H(|F[pl]||F[P2]|);接著，乘上一個空間的濾波函 數，如G，於本實施例中，濾波函數係為一低通濾波函數 (low-pass filter);接下來，再將其做逆傅立葉轉換（inverted Fourier transform)運算後，求得空間上的最大值，此最大值即 為預處理兩影像之位移。 1 ..…方程式① 其中： p 1 : —影像； p2 :另一影像； G :濾波函數； X，y : p 1及p2位移量的X軸位置與y軸位置。 200815909 下列即將此影像對位方法應用於 方法上做一說明。 、同實施例的影像重建ι 发 ! 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的 目的The port itself—one of the purposes of this report is to provide an image reconstruction method that utilizes the image correlation method of phase correlation and dissimilar operation to solve the problem of image quality in digital image capture and hand shake. In order to achieve the above object, an image alignment method according to an embodiment of the present invention includes: acquiring at least two images; calculating a relative displacement of two adjacent images, wherein the relative displacement is calculated by using a _phase correlation algorithm. Calculate each image, the absolute displacement of the first image in the image, and the common region of the image using the relative displacement and absolute displacement meter, and remove the image outside the common region. In order to achieve the above object, an image reconstruction method according to another embodiment of the present invention includes: capturing at least two images; calculating a relative displacement of one of the adjacent two images, wherein the relative displacement is calculated by using a phase correlation algorithm; An absolute displacement of each image and the first image in the image; using relative displacement and absolute position. Ten #出衫 like a common area, and remove the image outside the common area; determine the rotation center of the image; and reconstruct the stereoscopic data of the image. In order to achieve the above object, an image reconstruction method according to another embodiment of the present invention includes: performing at least two images; calculating a relative displacement of one of the adjacent two images, 6 200815909 wherein the relative displacement is calculated by using a phase correlation algorithm Calculating an absolute displacement of each image and the first image in the image; calculating a common region of the image using the relative displacement and the absolute displacement, and removing the image outside the common region; and superimposing each image to calculate Common area. The purpose, technical contents, features and effects achieved by the present invention will be more readily understood by the detailed description of the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The detailed description is not intended to limit the invention. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 is a flow chart showing the steps of an embodiment of an image registration method in accordance with the present invention. As shown in the figure, first, at least two images S10 are captured from the captured images, wherein, according to different applications, the images may be shot at the same angle for the same object, or may be shot at different angles; The relative displacement S20 of one of the two adjacent images, wherein the phase 'displacement system is calculated by a phase correlation algorithm; and then, an absolute displacement S30 of each image and the first image in the image is calculated. The absolute displacement is calculated by using a phase-dependent algorithm, or the relative displacement obtained by using any two images is used to calculate the absolute displacement of each image and the first image. Finally, the relative displacement is utilized. The displacement and the absolute displacement calculate a common area of all the images, and the image S40 outside the common area is removed. In one embodiment, the determination of the common region is based on the amount of displacement of the relative displacement, and if the image has an excessive displacement, for example, more than twice the root mean square value, it is removed. After finding the common area, remove all images outside the common area to complete the image alignment step. Following the above description, before calculating the relative displacement or absolute displacement of the image, an image pre-processing step is performed on the image. In one embodiment, the image pre-processing step includes at least one of sharpening, smoothing, and de-noising processing of the image 200815909. The image pre-processing steps help the captured image to process some unnecessary noise, or help the signal enhancement of the image to enhance the correctness of subsequent image alignment and even image reconstruction. In an embodiment, the method of calculating the relative displacement or the absolute displacement may be obtained by using a Fourier transform (F7) or a Fast Fourier transform (FFT) and its operation. The calculation method is as shown in the following equation (I): First, the two images to be calculated, such as pi and p2, are respectively subjected to Fourier transform to obtain two values F[pi] and F[p2]; then 'calculate two images. Correlation, that is, multiplying one of the shadow images by a complex conjugate of another image, such as F[pl](F[p2])1; The value is divided by the absolute value of the two images, such as (F[Pl](F[P2])1H(|F[pl]||F[P2]|); then, multiply the filter function of a space, such as G, in this embodiment, the filter function is a low-pass filter; then, after performing an inverse Fourier transform operation, the maximum value in space is obtained. This maximum value is the displacement of the preprocessed two images. 1 ..... Equation 1 where: p 1 : — image; p2 : another image; G : filter function; X, y : p 1 and p2 displacement X axis Position and y-axis position. 200815909 The following is an explanation of how to apply this image registration method to the method.
第2圖所示為根據本發明影像重 驟流程圖。於此實施例中，係將上述影建方法第一實施例之步 層掃描上。微斷層掃描所需影像係針對=對位方法應用於微斷 行拍攝，其中先將待投影物體固定，^同:物體隨不同角度進 角度之影像拍攝，由於機台本身旋旋轉機台以進行不同 題，故必須將影像對正為同一區域才利展動會造成影像對位問 工作。如圖所示，首先，由拍攝好的亍後續影像重建的 其中此處之影像是針對相同物體^中操取至少二影像 叶算相鄰兩影像之一相對位移S2〇，意δ角度拍攝；接著， ?關連演算法運算得之，例如藉由傅^相對位移係利用-相 茱函數轉換及其運算而得；再來，古十曾數轉換或快速傅立 一張影像的一絕對位移S30,亦即°，一影像與影像中之第 =每—張影像與第-張影像的位移（即弟=影像為基準，計 絕對位移之計算亦可是利用相位關於—實施 二者、:利用相對位移與絕對位移計算出所有 經完外的影像S4°’此時影像對位之 包括對位移或絕對位移前，更 處理步驟係包括對^ 實施例中，影像前 之至少其中、、衫像做銳化處理、平滑化處理與去雜訊處理 法係為^if—°於一實施例中’決定影像之旋轉中心的方 由判断旋轉執跡以求得旋轉中心，但不限於此。 重建麥一ΐ施例中，為方便微斷層掃描的FFT資料重建，在 像4 ’更包括内插（interpolate)或外插（extrap〇iate)各影 200815909 像以利斷層掃描的資料轉換，其中影像可延展至晝素數目為 21^個，且k為一正整數。接下來，便可進行如濾波反投影（Filter Back -Projection, FBP)方法來重建影像的立體資料。 接下來，請參考第3圖，第3圖所示為根據本發明影像重 建方法第二實施例之步驟流程圖。於此實施例中，係將上述影 像對位方法應用於如數位相機的數位取像裝置連續拍攝時之 防手震功能上，首先須先改變數位相機之曝光時間t為曝光時 間t/N後進行曝光，在預定的曝光時間内，擷取N張或少於N 張影像，其中N為一正整數。擷取完影像後之處理，於下列 做一描述。如圖所示，在計算出所有影像之共同區域前之步驟 (如S10、S20、S30及S40)皆與前述影像重建方法相同，此處 便不再贅述。在找出影像中之共同區域後，疊加每一影像計算 出之共同區域S70以增強信噪比。這個方法利用縮短曝光時間 以擷取到雜訊較大但是清晰的影像，但這些影像可經由影像前 處理後，例如銳化處理、平滑化處理與去雜訊處理之至少其中 之任一，再經過相位關連演算法對位，將共同區域的影像疊加 起來以增強信噪比，達到縮短曝光時間但是不模糊影像之數位 取像裝置的防手震功能。 接續上述說明，由於目前數位取像裝置本身資料、中央處 理器以及記憶體的限制，畫面擷取的方式也可能先擷取兩影像 後先進行影像的對位，取得共同區域之後，再將共同區域之影 像直接加到其中一影像中，而另一影像即可刪除以節省記憶體 空間。這個程序完成之後，再取下一影像進行相同的步驟。利 用這種方式，相鄰兩影像之共同區域會一直被累加至最後一張 影像，以完成影像修復。 依據上述，本發明的特徵之一係利用相位關連法計算兩影 像之位移以有效完成影像對位，此種影像對位方法可應用於其 200815909 他影像重建方法的前置處理，如利用於電子顯微鏡、χ光顯微 鏡tof層掃為或微斷層掃描。更甚者，由於數位取像裝置和被 攝物之間的相對移動，如手震，或是被攝物體本身高速的移 動，此影像對位方法亦可應用於如數位相機'照相手機等數位 取像裝置之影像重建，但其制並不限於此，亦於 需要影像對位處理之系統。 之用方、/、他 合上述，本發明係提供一種影像對位方法， 連凟异法能快速且較準確運算兩影像 員 位方法在擷取到不完全相同影像的 ，夕 ' 此種影像對 的目的。再者，利用相位關連演算法可’、可達到影像對位 域，藉由雜訊的減少致使影像的對饮w處理掉大部分的雜訊區 的之一係提供一種影像重建方法，更為精確。另，本發明目 影像對位方法，可解決在微斷層掃^相位關連演算法運算之 造成的對位問題。更甚者，利用相因機台本身旋轉震動所 位方法，可解決在數位取像裝置拍1關連演算法運算之影像對 的問題。 時，因手震影響相片品質 上所述之實施例僅係為說 3的在使熟習此項技藝之人士本發明之技術思想及特 以 點，其目 據以實施，當不能以之限定本發明〜能夠瞭解本發明之内容並 明所揭示之精神所作之均等變化t專利範圍，即大凡依本發 專利範圍内。 2修飾，仍應涵蓋在本發明之 11 200815909 【圖式簡單說明】 第1圖所示為根據本發明影像對位方法一實施例之步驟流程 圖。 第2圖所示為根據本發明影像重建方法第一實施例之步驟流 程圖。 第3圖所示為根據本發明影像重建方法第二實施例之步驟流 程圖。 【主要元件符號說明】 S10 擷取至少二影像 S20 計算相鄰兩影像之一相對位移 S30 計算每一影像與影像中之第一張影像的一絕對位移 S40 利用相對位移與絕對位移計算出所有影像之共同區域， 並移除共同區域之外的影像 S50 決定影像之旋轉中心 S60 重建影像之立體資料 S70 疊加每一影像計算出之共同區域 12Figure 2 is a flow chart showing the image re-sequence according to the present invention. In this embodiment, the step of the first embodiment of the above-described image building method is scanned. The image required for micro-tomography is applied to the micro-breaking method for the =-alignment method, in which the object to be projected is fixed first, and the same: the object is imaged with different angles of the angle, because the machine itself rotates the rotating machine to perform Different questions, it is necessary to correct the image for the same area, which will result in image alignment. As shown in the figure, firstly, the image reconstructed from the subsequent image captured by the captured image is obtained by taking at least two images of the same object and calculating the relative displacement S2〇 of the adjacent two images. Then, the correlation algorithm is calculated, for example, by using the relative phase shift system to perform the -phase 茱 function transformation and its operation; and then, the ancient ten-digit conversion or the fast Fourier image of an absolute displacement S30 , that is, °, the image of each image and the displacement of the first image and the image of the first image (ie, the image = the image as the reference, the calculation of the absolute displacement can also be based on the use of the phase) - the implementation of the relative The displacement and the absolute displacement are calculated for all the external images S4°' before the image alignment includes the displacement or the absolute displacement, and the more processing steps include: in the embodiment, at least the image before the image is made The sharpening process, the smoothing process, and the de-noise processing method are as follows: in one embodiment, the direction of the rotation center of the image is determined by determining the rotation of the image to obtain the center of rotation, but is not limited thereto. In one case In order to facilitate the reconstruction of the FFT data of the micro-tomographic scan, in the image of 4', the interpolate or extrapolation (extrap〇iate) images of the image 1515909 are used to facilitate the tomographic data conversion, wherein the image can be extended to the number of pixels. It is 21^, and k is a positive integer. Next, you can perform the Filter Back-Projection (FBP) method to reconstruct the stereo data of the image. Next, please refer to Figure 3, Figure 3. A flowchart of the steps of the second embodiment of the image reconstruction method according to the present invention is shown. In this embodiment, the image alignment method is applied to the anti-shake function of a digital image capturing device such as a digital camera. First, you must first change the exposure time t of the digital camera to the exposure time t/N and then perform exposure. In the predetermined exposure time, capture N or less than N images, where N is a positive integer. The following processing is described in the following. As shown in the figure, the steps before calculating the common area of all images (such as S10, S20, S30 and S40) are the same as the image reconstruction method described above, and will not be described here. .finding After the common area in the image is superimposed, the common area S70 calculated by each image is superimposed to enhance the signal-to-noise ratio. This method uses the shortened exposure time to capture a large but clear image of the noise, but these images can be transmitted through the image. After processing, for example, at least one of sharpening processing, smoothing processing, and de-noising processing, and then phase-aligned algorithm alignment, superimposing images of the common area to enhance the signal-to-noise ratio, thereby shortening the exposure time but The anti-shake function of the digital image capture device is not blurred. Continuing with the above description, due to the limitations of the current digital capture device itself, the central processing unit and the memory, the image capture method may first capture the two images first. After the image is aligned, the common area is obtained, and the image of the common area is directly added to one of the images, and the other image can be deleted to save the memory space. After this program is completed, take the next image and perform the same steps. In this way, the common area of the adjacent two images is always added to the last image to complete the image restoration. According to the above, one of the features of the present invention is to calculate the displacement of the two images by using the phase correlation method to effectively complete the image alignment. The image alignment method can be applied to the preprocessing of the image reconstruction method of 200815909, for example, for electronic use. Microscope, fluorescence microscope tof layer sweep or micro-tomography scan. Moreover, due to the relative movement between the digital image capturing device and the object, such as a hand shake, or the high speed movement of the object itself, the image registration method can also be applied to digital cameras such as digital cameras. Image reconstruction of the image capture device, but the system is not limited thereto, and is also a system that requires image registration processing. In combination with the above, the present invention provides an image alignment method, which can quickly and accurately calculate the two images of the position method to capture an image that is not exactly the same. The purpose of the right. Furthermore, the phase correlation algorithm can be used to achieve the image alignment domain. By reducing the noise, one of the image processing regions can provide an image reconstruction method. accurate. In addition, the image alignment method of the present invention can solve the alignment problem caused by the operation of the micro-fault sweep phase correlation algorithm. What's more, the problem of the image pair in the digital image capture device is solved by the method of rotating the vibration of the camera itself. The embodiment described above for affecting the quality of the photo due to the hand shake is only for the technical idea of the present invention and the specific point of the person who is familiar with the art, and the object of the invention is implemented, and the present invention cannot be limited thereto. The invention is capable of understanding the contents of the present invention and the scope of the invention is not limited to the scope of the present invention. 2 Modifications should still be covered by the present invention. 11 200815909 [Simplified Schematic Description] Fig. 1 is a flow chart showing the steps of an embodiment of the image registration method according to the present invention. Fig. 2 is a flow chart showing the steps of the first embodiment of the image reconstruction method according to the present invention. Fig. 3 is a flow chart showing the steps of the second embodiment of the image reconstruction method according to the present invention. [Main component symbol description] S10 Capture at least two images S20 Calculate the relative displacement of one of the adjacent two images S30 Calculate an absolute displacement of each image and the first image in the image S40 Calculate all images with relative displacement and absolute displacement Common area, and remove the image outside the common area S50 Determine the rotation center of the image S60 Reconstruct the image of the stereoscopic material S70 Superimpose the common area calculated by each image 12