物件自動分解與重建技術

10.29428/9789860544169.201801.0072

曾易聰

格式塔心理學 ； 封閉性 ； 物件偵測 ； 分解 ； 重建 ； 路徑追蹤 ； Gestalt psychology ； law of closure ； object detection ； decomposition ； reconstruction ； path tracking

NCS 2017 全國計算機會議

2017（2018 / 01 / 01）

377 - 382

繁體中文

在本論文中我們提出一種物件偵測技術，該技術是基於格式塔心理學中封閉性法則來偵測物件。現有技術難以偵測缺乏完整輪廓物件，而本方法藉由分析不完整的輪廓、不連續的邊界資訊，對物件進行分解、重建與偵測工作。本方法由三個主要部分組成，分別為：關鍵點偵測、近似曲線估測、和物件偵測。首先，關鍵點偵測是透過計算邊界和輪廓上每個點的曲率，決定該點是否成為關鍵點，然後將原先的邊界和輪廓分解成數個小邊界。緊接著，估測近似曲線建立任何兩個邊界之間的關聯性。最後，利用路徑追蹤方式找尋封閉迴路，同時偵測物件。實驗結果證明本方法可以重塑物件之外，對於物件亦可分解成多個具有完整輪廓且有意義的子物件，並且決定每個物件的內容屬性，以便實現物件偵測工作。

In this paper we proposed an object detection scheme, which is implemented based on law of closure in Gestalt psychology. Current approaches difficulty detect object without referring to complete contour of object. However, our scheme analyzes incomplete contour and discontinuous edges to detect object, and decomposes/reconstructs object. The proposed scheme consists of three major phases, including key vertex detection, approximate curve estimation, and object detection. First, computing curvatures of pixels on edges and contours is to detect key vertices, and then edges and contours are divided into small edges according to key vertices. Subsequently, the proposed scheme estimates approximate curve in order to establish relationship between any two edges. Finally, path tracking is applied to all edges to find closed loops as well as to detect objects. The experiment results demonstrate that our scheme is capable of forming object, separating a single object into multiple sub-objects, and determining attribute of object content for object detection.