建構於高斯混合模型和區域成長的快速物件擷取工具

Abstract

本篇論文提出一個快速且精確的物件裁剪方法，讓使用者可以簡單經由拖拉一個包圍感興趣物件的矩形將物件精確切割出來。這項技術可讓使用者快速從圖片中擷取目標物件，作為圖片合成或編輯等應用中圖塊素材的來源。所提方法基於高斯混合模型來表示圖片的顏色分布，並設計一個迭代式區域生長演算法執行圖片切割運算。這個方法具下列優點：(1)操作方式十分簡便易學，使用者只需利用滑數拖曳一個包圍住目標物件的矩形即可。(3)物件剪裁具高度精確度。針對具明確邊界的物件或複雜輪廓的物件都能有效的裁剪出。本論文實作所提方法並與文獻中高效率的物件擷取方法The GrapCut做比較，實驗結果顯示所提方法無論在擷取物件的完整性或裁剪時間都優於The GrapCut，顯示所提方法的可行性。

In this thesis a fast and accurate image cutoff method is developed. The method enables the users to clip object of interest out of an image, which is a useful tool for various applications such as image composition and/or editing. The proposed method represents the colors of an input image in Gaussian Mixture Model, and designs an iterative region growing based segmentation algorithm to draw out the target object. The proposed scheme has the following advantages: (1) the level of user interaction is low. The cut out operation is accomplished through simply drawing a rectangle encompassing the target object, and (2) the extracted objects are well-tailored. Both object with explicit contour and object with complicate contour can be extracted accurately. The proposed scheme is implemented and compared with the efficient object extraction method – the GrapCut. Experiment results show the proposed method exhibits higher performance than the GrapCut, both in the completeness of the extracted object and the computation time.