1307443 九、發明說明： 【發明所屬之技術領域】 本發明係有關-種影像防振系統與方法，特別是有關—種泛用型數位 影像防振系統與方法。 【先前技術】 數位影像防振技触要區分為三種型式：㈠電子式防振，㈡光學 式防振，u)數位式防振。電子式防振主要是利驗移_器來檢測攝影 機或照相機的簡，並錢财絲補健像·動；光學式防振主要是 利用稜鏡組合的移動來達到補償的效果。由於前二種型式都與硬體的結構 有關，故其應用範圍僅限内建於特定設備的—種處理裝置與程序。數位式 防振是賴㈣像處理程絲姐不必要祕__補償效果， 由於其為數位影像處理程序而非硬體機構如加速計、迴轉儀或液態棱鏡 等’故其不受線上（〇n-line)或離線（Gff_Hne)的限制，以及僅需少量的硬 體即可實現。 有關數位影像防減理的方式可分為社處理單元即移動向量估算與 移動補償兩大部份。移動向量估算主要是決定可靠的整體移動向量 (GMV)’首先透過錢輯財絲搜尋局部飾向量（[爾），整張影 像全區搜尋可得較可靠的局部雜向量，然關其計算量過大，在實際應 用上顯知不合經濟效益，故有另外—佩計算量的比對方式如等 所提出的代表點比對（RPMW, κ·驗等所提㈣邊緣比對（ΕρΜ)或 S· J· Κο等所提出的位元面比對（ΒρΜ)，其主要的目的就是降低計算量同 1307443 時又不犧牲太多可靠性。-般㈣，代表點崎（Rm)科可降低最多 計算量，然献絲常情況卻非t如影像中有軸物體、拍攝時 平移、影像中有重複_或大的低對比區鱗，造献用上的困難。 有鑑於此，本發明係針對上述之問題，提出_種泛用型數位影像防振 系統與方法，可以義於—般或異諸況的防振處理，增加整體移 檢測的可靠性。 【發明内容】 '本發明之主.要目的，係在提供—種數位影像_系統與方法，以有效 率的數位處理方絲職在乏·、重複_献的低氣區域各種 異常情況時整體移動向量檢測的可靠性。 本發明之另-目的’係在提供—魏位影像統與方法，以背景 檢測估异整體鷄向量強化背景鶴補償效果 本發明之再-目的’係在提供—健位影像防振錢財法，增進攝 影機在平移時仍保有防振的功能。 X據本發Θ種數位影像防振系統包括：移動向量估算子系統從一 T置接彳卜補助蝴彡細計_軸向量，再利用倒三 =算可雜指標求得乏條件移動向量；整體移動向量估算子系統連接 部/量估算子祕並接收局部移動向量及乏條件軸向量，並以數個局 估2之移動向量、乏條件移動向量、零移動向量及前—次整體移動向量 接收整Γ_;職向量估算子_接_動_算子系統並 艘移動向量，輯整體軸向量纽_顯向量；歸麵子系統 1307443 連接補償向量轉子紐並接收補償向量，同時根餅聽像使影像經過 補償後產生特_«彡賴出至-輸出裝置。 另外種數位影像防振方法包括··從一影像裝置接收 一序列影像， .將序列影像與前影像比對以計算局部移動向量，再使關三角法計算可靠 性鍊求得乏條件移軸量，«崎個局籠域之移動向量、乏條件移 動向置、零移動向量及前一次整體移動向量估算整體移動向量，接著估算 傳統累積移動向量並以内部積分回路來消除影像位置執跡落後的問題，再 以細限嶋義嫩小術—補_(⑽），最後依補償 向i補侦序列影像並輸出至一輸出裝置。 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政。 【實施方式】 本發明係雜效麵她猶^谢析峨檢測方法， 计异出強健及可靠的影像移動向量，再臂特定的遽波處理程序，解出影 像的補償向量使其產生平滑、;噴暢的影像參列。 請參閲第-圖為本發明之系統方塊圓;【包括兩大模組：移動向 模組1〇與移動補償模組20。移動向量估算模組1〇接收影像並决定 整體移動向量，細餘__ :⑴_____ =讯額，咖物畅靖询12。科，軸補償料 〇主要是減整體機«，料必麵絲姆而軸转移動的部份 使其產生穩定_像’由_子系統所組成，⑴補償向量估算子系㈣ 1307443 及(ii)雜補償子純22。最後輪出穩定後的序像；其中，祕的影 μ峨，顺输顺始序 列影像。 再請參閲第二圖為本發明之局部移動向量與乏條件移動向量枯算子系 =了’原始影像輸八後經上一張影像代表點sm與目前影像比對產生 H«_S113，响物响·私娜可能含有 移動物體、拍攝時的平移、番遴 θ〉、乏特徵影像或是大的低對比區域等 =而造成局部移動向量的可靠性降低，故將共相_比對矩陣取最 ^Y S114 5 " y ^ 與共侧轉所求㈣局部鶴妓挪把可鎌指標低 於門酬心續，物_各局獅向量分量，以中間 值的方法求得乏條件移動向量SU8。 另外’請參嶋三圖為本發明可靠性指標計算示意m y轴的 影曲線中以最小值為三角形的頂點，以一適#的高補償 並 轉峨_形糾恤_機赛的底部， 高二—n作為加重值’故以2d'n作為可靠性指級，值愈小代表可靠性愈 =移動向娜彻12 ^峨晴綱峨據，即四 4的局部移動向量、乏條件移動向量、零軸向量、前— 量:整體鶴向量的估算為避免尋找狀移動物體誤判為整體移動^向 故採用可紐的背綠_絲加雜算如第_解以崎伽的影 1307443 像為例，物㈣，嶋㈣，計_二嶋在此各個背 景區域的絕對值誤差和（膽），採用對等評估方式來來計算每—移動向量 的誤差值，懈纖的从哪—_七個麵峨予不同的 分數’如誤差值大依序給予較大的分數(減柯），最後再將分數總和， 分數最小的移動向量即代表整體移觸量。此種料評估的方法可避免在 影像中某區域高對比的特性主宰了整個估算的結果。同時以高可能性的背 景區域作為估算的依據’使所求得的移動向量與實際背影移動相似。 補償向量估算子系統21是根據整體移體向量來產生-補償向量使影像 經過補償後產生平滑的序聽像。其估算的方侧如第五_示傳統累 積移動向量估算211其計算式為式（1) CMV{t) = k{CMV{t-1)) + (aaGMV(t) + (1 -a)GMV{t-1))...... 〇) 其中鴻影像的數目，CMF(〇為第,個影像的補償移動向量，〇經㈣第嗰 影像的整體移動向量，0小1和。增加綠會減低影像不想要 的振動效S ’但也增加了 CMVm，樣葬償财_影祕域會變小， 尤其影像在平移時情況更為嚴重。故加.部積分回路212來消除影像平 移時所造成的位置触落後的問題，並崎波㈣加來_ CMy值以確保 補償後有效的影像區域，其計算式為式（2) C_)=k，CM^-l)+[aeGM^)+(1_a)··^，c經」㈣ CW_Uf)=CMVJ(t-\)+CW{t) ...... (2) CMV{ή=clippetiCMVQ)) I ^M,p< 1 ,,，， 具甲L」 L1」，”·，’表示陣列相乘，c"pper()定義 為剪波器。 參閱第六圖為内部積分回路5〇2所改善的結果：（a)是傳統方法在影像 1307443 平移時所造成位置執跡嚴重落後的現象，⑹是本發騎提㈣方法，對於 影像平移時所造成位置軌跡嚴重落後有顯著的改善。 .本發明亦可應用於具有處理器之裝置' 數位訊號處理n (DSP)或特殊 .應用碰祕（縦），請參«為本發明之方法流糊示細，包括·· 步驟S1接收-序列影像，從一影像裝置接收一序列影像，步驟泣計算局 部移動向量，將該相影像與前影佩對產生共相__矩陣以計算局 部移動向量’步㈣計算乏條件鶴向量，使_三肖法計算可靠性指尸1307443 IX. Description of the Invention: [Technical Field] The present invention relates to an image anti-vibration system and method, and more particularly to a general-purpose digital image anti-vibration system and method. [Prior Art] Digital image anti-vibration technology is divided into three types: (1) electronic anti-vibration, (2) optical anti-vibration, and u) digital anti-vibration. The electronic anti-vibration is mainly used to detect the camera or the camera, and the money is used to reinforce the image. The optical anti-vibration mainly uses the movement of the cymbal combination to achieve the compensation effect. Since the first two types are related to the structure of the hardware, the application range is limited to the processing devices and programs built into the specific equipment. Digital anti-vibration is based on the treatment of Cheng Sijie's unnecessary secret __ compensation effect, because it is a digital image processing program rather than a hardware mechanism such as accelerometer, gyroscope or liquid prism, so it is not online (〇 N-line or offline (Gff_Hne) restrictions, and only a small amount of hardware can be achieved. The methods for digital image anti-reduction can be divided into two parts: social processing unit, mobile vector estimation and motion compensation. The motion vector estimation mainly determines the reliable overall motion vector (GMV). First, the local decoration vector is searched through the money collection ([Er], the entire image can be searched for a more reliable local miscellaneous vector, and then the calculation amount Too large, in the practical application, it is obvious that it is not economical, so there is another comparison method of the calculation of the amount of the calculation, such as the proposed representative point comparison (RPMW, κ · test, etc. (4) edge comparison (ΕρΜ) or S · J. Κο et al. proposed the bit-plane comparison (ΒρΜ), whose main purpose is to reduce the amount of calculations with the same amount of 1307438 without sacrificing too much reliability. - (4), representing the point Rm (Rm) section can be reduced the most The amount of calculation is not the same as the case where there are axial objects in the image, translation during shooting, repeated _ or large contrasting scales in the image, and difficulties in creation. In view of this, the present invention is directed to In view of the above problems, a general-purpose digital image anti-vibration system and method are proposed, which can be used for anti-vibration processing of general or different conditions, and the reliability of the overall shift detection is increased. [Summary of the Invention] The purpose is to provide a digital image _ System and method, the efficiency of the overall motion vector detection when dealing with various abnormal conditions in the low-gas region of the lacking and repeating in the efficient digital processing. The other-purpose of the present invention is provided Image system and method, using background detection to estimate the overall chicken vector to strengthen the background crane compensation effect. The re-purpose of the present invention is to provide the image-enhanced image anti-vibration method to improve the camera's function of preventing vibration during translation. According to the present invention, the digital image anti-vibration system includes: the motion vector estimation subsystem selects the _axis vector from a T-substance, and then uses the inverse three-counter-difference index to obtain the conditional motion vector; The overall motion vector estimates the subsystem connection/quantity estimation sub-secret and receives the local motion vector and the conditional axis vector, and estimates the motion vector, the conditional motion vector, the zero motion vector, and the pre-to-global motion by a number of estimates. Vector receiving integer _; job vector estimator _ _ _ _ computing subsystem parallel movement vector, edit the overall axis vector _ _ vector; homing subsystem 130743 connect the compensation vector rotor and receive the compensation vector At the same time, the image of the root cake is compensated to produce a special image output device. The other digital image anti-vibration method includes receiving a sequence of images from an image device, and comparing the sequence image with the front image. In order to calculate the local motion vector, and then calculate the reliability chain by the triangulation method, the conditional shift axis amount is obtained, and the motion vector of the cage field, the conditional moving direction, the zero motion vector and the previous overall motion vector estimation are calculated. The overall motion vector, then estimate the traditional cumulative motion vector and use the internal integration loop to eliminate the problem of image location obstruction backwardness, and then use the fine-limit 嶋义嫩小术-complement _((10)), and finally compensate the sequence image by compensation And output to an output device. The purpose, technical content, features, and merits of the present invention are more readily understood by the detailed description of the embodiments and the accompanying drawings. [Embodiment] The present invention is a hybrid effect. She still relies on the detection method, and calculates a robust and reliable image motion vector, and then performs a specific chopping process to solve the image compensation vector to make it smooth; Smooth image participation. Please refer to the first figure for the system circle of the present invention; [including two modules: a moving direction module 1 and a motion compensation module 20. The motion vector estimation module 1 receives the image and determines the overall motion vector, and the __: (1)_____ = signal amount, and the coffee object is greeted by 12. Section, the shaft compensation material is mainly to reduce the overall machine «, the material must be surfaced and the part of the axis is moved to make it stable _ like 'by _ subsystem, (1) compensation vector estimation subsystem (4) 1307443 and (ii The hybrid compensator is purely 22. Finally, the stable sequence is rotated; among them, the secret image is 峨, and the sequential sequence image is transmitted. Referring again to the second figure, the local motion vector and the conditional motion vector of the present invention are the operator system = "the original image is transmitted after the previous image represents the point sm and the current image is compared to produce H«_S113, Objects and privates may contain moving objects, panning during shooting, Panyu θ>, lack of feature images or large low contrast areas, etc., resulting in reduced reliability of local motion vectors, so the common phase _ comparison matrix Take the most ^Y S114 5 " y ^ and ask for the co-rotation (4) local cranes to move the 镰 镰 镰 低于 低于 , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , SU8. In addition, please refer to the three figures for the reliability index calculation of the invention. The peak of the my axis is the triangle with the minimum value, with a high compensation of one suitable and the bottom of the game. -n as the weighting value', so 2d'n is used as the reliability index. The smaller the value is, the more reliable the reliability is. The moving to the Nacher is the local motion vector of the 4th, the conditional motion vector. Zero-axis vector, front-quantity: The estimation of the overall crane vector is to avoid the search for moving objects and misjudge the overall movement. Therefore, the back-green of the can be used as the _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ For example, the object (4), 嶋 (4), _ 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 嶋 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对 绝对Faces are given different scores'. If the error value is large, the score is given in large order (minus), and finally the score is summed. The moving vector with the smallest score represents the overall shift amount. This method of material evaluation avoids the high contrast characteristics of an area in the image that dominates the overall estimated results. At the same time, the high-period background area is used as the basis for estimation' so that the obtained motion vector is similar to the actual back-movement. The compensation vector estimation subsystem 21 generates a compensation vector based on the overall migration vector to compensate for the image to produce a smooth sequenced image. The estimated square side is as shown in the fifth example of the conventional cumulative motion vector estimation 211, and its calculation formula is (1) CMV{t) = k{CMV{t-1)) + (aaGMV(t) + (1 -a) GMV{t-1))...... 〇) The number of images, CMF (〇 is the first, the compensated motion vector of the image, the 移动 ( (4) The overall motion vector of the second image, 0 small 1 and. Increasing the green will reduce the unwanted vibration effect of the image. 'But it also increases the CMVm. The burial payment will be smaller. Especially when the image is shifted, the situation is more serious. Therefore, the integral circuit 212 is added to eliminate the image. The problem of positional backwardness caused by translation, and the wave of _CMy value is added to ensure the effective image area after compensation. The calculation formula is (2) C_)=k, CM^-l)+[aeGM ^)+(1_a)··^,c by (4) CW_Uf)=CMVJ(t-\)+CW{t) ...... (2) CMV{ή=clippetiCMVQ)) I ^M,p< 1 ,,,, with A L L1", "·," means array multiplication, c"pper() is defined as a shear. See Figure 6 for the internal integration loop 5〇2 improved results: (a ) is a phenomenon in which the traditional method is seriously backward in the position caused by the translation of the image 1307443, and (6) is the riding of the hair. (4) The method has a significant improvement on the positional trajectory caused by the image shifting. The invention can also be applied to a device with a processor 'digital signal processing n (DSP) or special application touch (縦), please refer to «The method of the present invention is succinct, including: · Step S1 receiving-sequence image, receiving a sequence of images from an image device, stepping to calculate a local motion vector, and generating a common phase with the front image pair _ _Matrix calculates the local motion vector 'step (four) to calculate the conditional crane vector, so that the reliability of the _ three-dimensional method is
及乏條件軸向#，倾S4估算錄軸向量，讀倾局輕域之移動 向量、該乏條*畅量、零移動向量及前—次整體移動向量估算整體移 動向量’步驟S5估算傳統累積移動向量，估算傳統累積移動向量，步㈣ 消除影像錄鱗驗，雜由輯平料所造狀影像位錄跡落後， 步計算-補償向量，以剪波諫·積鶴向量的大小哺生一補償 向量（㈤，以及步驟S8補償序列影像，倾補償㈣補倾序列影像， 將不必要的振動去除而保留平滑移動的部份使其遽敎的影像，最後再 輸出敎後断冊敍-輸域置；射裝I係為攝職、數 ^相機或數位影賴所齡的影像_，輸錄聽為投影纽、顯示榮 幕或儲存裝置。 本發肿m上述线財法，錄㈣妓服各種異常狀況造成 • ^移動向量可靠性降低與避免因攝影機因平移所造成的位置軌跡落後的 w ’使得影像裝置所接㈣影像_產生平滑的序列影像。 以上所嶋嫩侧纟恢_，_錢峨技術者 1307443 能暸解本發明之内容並據以實施，而非限定本發明之專利範圍，故，凡其 他未脫離本發明所揭示之精神所完成之等效修飾或修改，仍應包含在以下 所述之申請專利範圍中。 1307443 【圖式簡單說明】 第一圖為本發明之系統方塊圖。 第二圖為本發明之移動向量估算子系統方塊圖。 第三圖為本發明可靠性指標計算示意圖。 第四圖為本發明之背景檢測區域。 第五圖為本發明之補償向量估算子系統。 第六圖為本發明之内部積分回路與傳統方法所改善的影像比較圖，圖（a)是 傳統方法，圖（b)是本發明。And the conditional axis#, tilt S4 to estimate the axis vector, read the moving vector of the degraded light domain, the bark * smoothing, zero motion vector and the pre-total global motion vector to estimate the overall motion vector' step S5 to estimate the tradition Accumulate the motion vector, estimate the traditional cumulative motion vector, step (4) eliminate the video recording scale test, the video image recorded by the miscellaneous material is behind, the step calculation-compensation vector, and the size of the clipping wave and the crane vector A compensation vector ((5), and step S8 compensates for the sequence image, and the tilt compensation (4) fills the sequence image, removes the unnecessary vibration and retains the smooth moving part to make the image of the image, and finally outputs the image after the output. The field is set; the shooting I is the image of the job, the number of cameras or the digital image of the age of _, the input is recorded as a projection button, display the screen or storage device. The swelling of the above line of money, recorded (4)妓 各种 各种 各种 各种 ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动 移动,_money The present invention is not limited to the scope of the present invention, and other equivalent modifications or modifications made without departing from the spirit of the present invention should be included in the following. Illustrated in the scope of patent application. 1307443 [Simple description of the diagram] The first diagram is a block diagram of the system of the present invention. The second diagram is a block diagram of the motion vector estimation subsystem of the present invention. The fourth figure is the background detection area of the present invention. The fifth figure is the compensation vector estimation subsystem of the present invention. The sixth figure is an image comparison diagram improved by the internal integration loop of the present invention and the conventional method, (a) It is a conventional method, and Figure (b) is the present invention.
I 第七圖為本發唞之方法流程示意圖。 【主要元件符號說明】 10移動向量估算模組 11局部移動向量與乏條件移動向量估算子系統 12整體移動向量估算子系統 :乂 20移動補償模組 > 21補償向量估算子系統 :二- 22影像補償子系統 211傳統累積移動向量估算 "212内部積分回路 一 213剪波器 12I The seventh figure is a schematic diagram of the method flow of the present invention. [Major component symbol description] 10 motion vector estimation module 11 local motion vector and lack conditional motion vector estimation subsystem 12 overall motion vector estimation subsystem: 乂20 motion compensation module > 21 compensation vector estimation subsystem: two - 22 Image Compensation Subsystem 211 Traditional Cumulative Motion Vector Estimation "212 Internal Integral Loop-213 Clipper 12