| DC 欄位 |
值 |
語言 |
| DC.contributor | 資訊工程學系 | zh_TW |
| DC.creator | 吳鼎汶 | zh_TW |
| DC.creator | Hendrik | en_US |
| dc.date.accessioned | 2012-7-17T07:39:07Z | |
| dc.date.available | 2012-7-17T07:39:07Z | |
| dc.date.issued | 2012 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=995202074 | |
| dc.contributor.department | 資訊工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 近年來空照監控影像的需求不斷增加，其應用範圍廣泛，主要用在軍事與交通運輸上。在交通運輸方面，空照監控影像不但能讓警察掌握交通狀況還能協助道路秩序之管理。為了能自動化地處理空照影像，必須先對影像中的物件與區域進行分割與標記的動作，如物件偵測、事件偵測、等。
本論文提出一套自動化空照影像之區域分割與分類系統。任何影像分割的演算法都可以套入本系統中。其中比較了兩種分割演算法，分別是分水嶺(Watershed)與均值偏移(Mean-shift)分割演算法，但由於上述演算法會導致區域被過度切割，故本系統將過小的區域進行合併。其合併方式是根據八連通的方式建構一個無向圖(undirected graph)，將影像中的每個區域視為一個頂點(vertex)，而連接兩區域的邊(edge)會給定一個權重(weight)，該權重代表兩區域的相異程度，再根據該權重對鄰近的小區域進行合併。而對於每個區域，本系統會擷取其顏色與紋理特徵，並使用支持向量機(Support Vector Machine)將每個區域進行分類的動作，最後再將同類的鄰近區域合併，獲得最終的分類結果。經實驗證實，本論文所提出的方法能有效的對各種空照影像進行分割與分類。
| zh_TW |
| dc.description.abstract | The demand for aerial surveillance video keeps growing in recent years. It has been proved to be an effective way to collect information for a wide range of applications, such as intelligence transportation or military applications. In intelligence transportation applications, aerial surveillance not only provides traffic monitoring but also assists traffic management. Manual labeling of objects and regions in aerial videos is a tedious task. Objects and regions in aerial videos need to be segmented and labeled to enable automated video processing, such as object detection, event detection, automated aerial videos understanding, etc.
In this thesis we propose an automatic image segmentation and labeling system for aerial surveillance images. Any kind of segmentation algorithm can be applied in our system. In this work, we compare watershed and mean-shift segmentation algorithms. Because the above mentioned segmentation algorithms might lead to over-segmentation, small regions need to be merged. We then construct an undirected-graph based on 8-connected local neighborhood, where each region is a vertex and the weight of edge connecting two regions is the dissimilarity measure of this two regions. Adjacent small regions are merged according to the weights of the edges. For each region we extract low-level features and use Support Vector Machine (SVM) classifier to label the region. Based on the output of the SVM classifier adjacent regions with the same label will be further merged to obtain the final labeling result. The experimental results have shown that our proposed system can effectively segment and label various aerial images.
| en_US |
| DC.subject | 空照監控 | zh_TW |
| DC.subject | 區域分割 | zh_TW |
| DC.subject | 影像分類 | zh_TW |
| DC.subject | image labeling | en_US |
| DC.subject | region segmentation | en_US |
| DC.subject | aerial surveillance | en_US |
| DC.title | 航空監控影像之區域切割與分類 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Region Segmentation and Labeling in Aerial Surveillance Images | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |