查詢結果分析
來源資料
頁籤選單縮合
題 名 | MFA方法應用於影像中近似橢圓物件之分割=Segmentation of Ellipse-Like Objects in an Image with MFA Approach |
---|---|
作 者 | 王友俊; 周瑞仁; | 書刊名 | 農業機械學刊 |
卷 期 | 15:1 民95.03 |
頁 次 | 頁15-24 |
分類號 | 312.1 |
關鍵詞 | 影像處理; 影像分割; 機器視覺; 橢圓; 主動輪廓模式; Image processing; Segmentation; Machine vision; Ellipse; Mean shift; Fitzgibbon; Active contour model; |
語 文 | 中文(Chinese) |
中文摘要 | 本研究發展一套完整影像處理方法MFA (Mean shift, Fitzgibbon’s fitting ellipses method and Active contour model),使相互接觸之近似圓物件影像得以正確分割。首先利用Canny影像邊綠偵測獲得物件邊界影像,並以所偵測邊界之像點作為動態質點,這些動態質點在影像位能場的聚類,關係著物件影像能否正確分割,除此之外具明顯紋理之物件影像,常無法被前置濾波器濾除而造成雜訊,這些雜訊的存在影響動態質的正確聚類。有鑑於此,運用mean shift演算法決定動態質點位移大小與方向,可有效克服雜訊之影響,使動質點能正確聚類,如此,每個群聚即代表各個物件;接著應用Fitzgibon直接最小平方橢圓擬合法自動決定每個物件之初始形變曲線,最後以active contour model (ACM,主動輪廓模式)重建近似楕圓物件之輪廓。實驗結果顯示,在使用mean shift演算法時,只要設觀察半徑為欲偵測物件平均半徑之0.85-1.2倍,即能有效克服物件內之緩理或雜訊,使動質點得以正確聚類,完成具明顯紋理並相互接觸之近似橢圓物件影之分割。經由此方法所處理之物件影像,可依據其個別封閉輪廓線的取,很容易求得影像中各物件之幾何、紋理或顏色之特徵,方便後續之影像處理。 |
英文摘要 | The study developed a complete image processing approach to segment touching ellipse-like objects in an image, particularly for objects with obvious texture, by interating Mean shift algorithm, Fizgibbon’s fitting ellipses method an Active contour model together, referred to as MFA. First, we employ Canny edge detection to obtain the edges of objects in an image. With the detected edges serving as active points, they would be clustered in an image potential fields. The clustering qualify might greatly affect the correctness of object segmentation. In addition, obvious texture in objects tends to be considered as noise and causes misclassification if the texture could not be pre-filtered out. For this reason, this research proposed a method that employed a mean shift algorithm to estimate the displacement and direction of active points for reducing the interference of noise. Accordingly, the points should be clustered more correctly and then each cluster represents its corresponding object.after having applied mean shift algorithm, we utilize Fizgibbon’s direct least square fitting of ellipses to determine initial deformable contours for contour reconstruction using the active contour model (ACM) approach. The results show that noise or texture within the detected objects could be effectively suppressed a long as the observable radius is set at 0.85-1.2 times the average radius of the object. The complete contours of touching objects reconstructed by the approach proposed in this study would facilitate the subsequent image processing to obtain the geometric, texture, and color characteristics of objects in an image. These features might be used for further clustering, image recognition or understanding. |
本系統中英文摘要資訊取自各篇刊載內容。