檢視以往在偽卡偵測的研究上,均是以能找出最多偽卡交易的方法為研究目標,再以該方法應用在所有持卡人身上。這些研究中的確找出了許多的好方法，但卻也有著以大多數人的行為模式套用在個人行為模式上的盲點，雖然也有研究是依據個人行為模式來偵測偽卡,但仍然沒有做到完全以個人為研究單位。 為解決上述的問題,本研究採用事件歷史分析法(Event History analysis) ，此方法用於觀察一段時間內某一事件的發生與否，探討事件的發生原因，並據以產生預測模型(hazard function)來預測事件發生的可能性。[王仕茹，民82] [林欣美，民84] [張佩惠，民85] 應用此方法於偽卡交易偵測上，可以得知造成該卡人偽卡交易產生的因素為何?是地點、特店別還是其他因素;更甚而可以建立個人預測偽卡的模型(hazard function)，利用此模型判斷某筆交易是偽卡交易的百分比。 本研究應用此方法找出造成偽卡交易產生的因素，但限於所使用的軟體未提供當時間相關變數存在時，建構危機函數 (hazard function) 的功能，因此本研究採用決策樹來彌補這方面的不足。 / When reviewing the researches about fraud cards detecting in the past, most of them focus on how to enhance the probabilities to detect fraud cards. Then,they applied the methods they found to detect all the cardholders’ credit cards. The shortcoming is it’s easy to treat the fraud transactions as the normal transactions because every cardholder has different consuming behavior. Even some researchers try to detect fraud cards by personal consuming behavior. In fact,they still don’t detect fraud cards by personal consuming behavior completely. To resolve the problems described as above,this research adapt Event History analysis. Event history analysis revolves around two concepts: the risk set and the hazard rate(Allison 1984). The risk set comprises the number of dyads subject to event occurrence at a particular point in time. “The hazard rate is the probability that an event will occur at a particular time to a particular individual, given that the individual is at risk at that time” (Allison 1984, 16). We applied this method on fraud cards detecting to find the factors which cause fraud cards happened. Besides that,we can build the hazard function and use it to know the fraud opportunity of the transaction. This research applied the method to find the factors which cause fraud cards happened successfully. The software this research used doesn’t provide the function to create hazard function when the time dependent variables exist,so this research replaced it with decision tree

No description available.

本研究的主要目的在於評估現有文獻中，用來偵測盈餘管理行為之各種應計模型的績效，以期透過模型間之比較，推論出何種模型較適合國內的會計資料，使研究者在面對模型選擇時，能有所依據。 　　本研究以民國79年9月至84年6月間於台灣證券交易所上市的90家公司，合計1800個公司－季，為研究對象，並將其作為抽取樣本的母體。研究中所使用之樣本單位為公司－季的型態，樣本共分為三類：1.隨機抽取之樣本。2.具有極端財務績效之樣本。3.人為操弄應計項目之樣本。實證研究設計方面主要由顯著水準(level of sigificance)及檢定力(power)兩方面著手，分別評估Healy、DeAngelo及Jones三種估計可裁決性應計項目模型的相對績效。在顯著水準的比較上，主要是觀察各模型在隨機樣本及極端財務績效樣本中犯型一錯誤的比例。檢定力的比較方面，則主要是觀察各模型在人為操弄樣本下犯型二錯誤的比例。 　　實證結果顯示： 　　1. 在隨機樣本中，各模型所對應之真實顯著水準與檢定時所設定的顯著水準間，均無顯著差異。各模型所估計而得之迴歸係數均趨近於0，但迴歸係數之標準差均太大，顯示各模型對可裁決應計項目的估計均存有衡量誤差，尤以DeAngelo模型最為嚴重。 　　2. 在具有極財務績效之樣本中，各模型所對應之真實顯著水準與檢定時所設定的顯著水準間，具有明顯的差異，顯示其無法有效控制財務績效與應計項目間的關聯，以致於造成可裁決性應計項目的衡量誤差。 　　3. 在人為操弄的樣本中，當操弄數額為前期總資產的10%時，各模型拒絕虛無假設的比率均不高，顯示其檢定力偏低。就整體而言，Healy與Jones模型的檢定力較高，DeAngelo模型的檢定力則顯著較低。

在電腦視覺的研究之中，有關物件的偵測與追蹤應用在速度及可靠性上的追求一直是相當具有挑戰性的問題，而現階段發展以視覺為基礎互動式的應用，所使用到技術諸如:類神經網路、SVM及貝氏網路等。 本論文中我們持續深入此領域，並提出及發展一個方向性邊緣特徵集(DEM)與修正後的AdaBoost訓練演算法相互結合，期能有效提高物件偵測與識別的速度及準確性，在實際驗證中，我們將之應用於多種角度之人臉偵測，以及臉部表情識別等兩個主要問題之上；在人臉偵測的應用中，我們使用CMU的臉部資料庫並與Viola & Jones方法進行分析比較，在準確率上，我們的方法擁有79% 的recall及90% 的precision，而Viola & ones的方法則分別為81%及77%；在運算速度上，同樣處理512x384的影像，相較於Viola & Jones需時132ms，我們提出的方法則有較佳的82ms。 此外，於表情識別的應用中，我們結合運用Component-based及Action-unit model 兩種方法。前者的優勢在於提供臉部細節特徵的定位及追蹤變化，後者主要功用則為進行情緒表情的分類。我們對於四種不同情緒表情的辨識準確度如下：高興(83.6%)、傷心(72.7%)、驚訝(80%) 、生氣(78.1%)。在實驗中，可以發現生氣及傷心兩種情緒較難區分，而高興與驚訝則較易識別。 / Rapid and robust detection and tracking of objects is a challenging problem in computer vision research. Techniques such as artificial neural networks, support vector machine and Bayesian networks have been developed to enable interactive vision-based applications. In this thesis, we tackle this issue by devising a novel feature descriptor named directional edge maps (DEM). When combined with a modified AdaBoost training algorithm, the proposed descriptor can produce effective results in many object detection and recognition tasks. We have applied the newly developed method to two important object recognition problems, namely, face detection and facial expression recognition. The DEM-based methodology conceived in this thesis is capable of detecting faces of multiple views. To test the efficacy of our face detection mechanism, we have performed a comparative analysis with the Viola and Jones algorithm using Carnegie Mellon University face database. The recall and precision using our approach is 79% and 90%, respectively, compared to 81% and 77% using Viola and Jones algorithm. Our algorithm is also more efficient, requiring only 82 ms (compared to 132 ms by Viola and Jones) for processing a 512x384 image. To achieve robust facial expression recognition, we have combined component-based methods and action-unit model-based approaches. The component-based method is mainly utilized to locate important facial features and track their deformations. Action-unit model-based approach is then employed to carry out expression recognition. The accuracy of classifying different emotion type is as follows: happiness 83.6%, sadness 72.7%, surprise 80%, and anger 78.1%. It turns out that anger and sadness are more difficult to distinguish, whereas happiness and surprise expression have higher recognition rates.

眼動儀目前的主要用途在分析使用者的觀看行為，藉以改善介面的設計，或幫助身體有缺陷但眼睛還能轉動的使用者與外界溝通。隨著相關技術的發展，眼動儀將可能如同滑鼠、鍵盤一般，成為使用者輸入裝置的選項。本論文的目的在於設計並實作低成本之穿戴式與遠距眼動儀，並將其應用於以視線軌跡為基礎的人機介面，希望能夠增進人與電腦之間的互動方式。由於眼動儀會受到雜訊與角膜反射點等的影響，本研究提出利用瞳孔周圍暗色點比例較高的特性，增加定位之準確性，以改善眼動儀之精確度，此外，頭部的移動亦會造成眼動儀在計算投射位置時之誤差，本研究也針對這個問題提出因應之解決方案。利用前述製作的眼動儀，本論文實作數個以視線軌跡為基礎的人機介面，包括視線軌跡網頁瀏覽器、強化眼睛注視照片區域、井字遊戲、互動式媒體等，並利用眼動儀記錄使用者觀看手機介面的行為。 / Eye tracker, a device for measuring eye position and movements, has traditionally been used for research in human visual system, psychology and interface design. It has also served as an input device for people with disability. With recent progresses in hardware and imaging technology, it has the potential to complement, even replace popular devices such as mouse or keyboard for average users to communicate with the computer. It is the objective of this research to design and implement low-cost head-mounted and remote eye trackers and subsequently develop applications that take advantage of gaze-based interactions. Specifically, we improve the precision of the tracking result by designing a new pupil detection algorithm as well as compensating for head movement. We then present several gaze-based user interfaces, including eye-controlled web browser, attention-based photo browser, interactive game (tic-tac-toe) and media design. We also investigate the feasibility of utilizing the eye trackers to analyze and evaluate the design of mobile user interface.

隨著網路技術的發展，目前網路電話(VoIP)已有逐漸取代傳統電話的趨勢。尤其能夠允許多人同時在線上進行會談是其最大的優勢之一。但在多人參與網路會談時，因為聲音在空間中傳遞或反射等因素，使得由喇叭發出的聲音再次被麥克風收回，造成回音的產生。會談中只要有一位使用者的裝置發生回音時，回音訊號就會在與會者之間擴散，使得所有使用者均會受到影響，進而嚴重影響網路通話的進行。此狀況在參與會談人數越多時，發生機率越高，且對通話品質影響越嚴重。 傳統電話在一對一通話時，通常使用遠端回音消除機制(Near End Echo Canceller)，由接收端在接收聲音後先暫存在記憶體中再播放，再將麥克風擷取的聲音與事先暫存的訊號反向後混合，以抵銷回音。網路會談的環境下，由於沒有標準的聽筒設備，使得回音發生的時間難以預估。且多人參與的網路會談中，由於收聽者所聽到的聲音可能混合多個使用者說話的聲音與回音，使得回音訊號難以偵測。另外，由於網路傳輸的特性，回音訊號到達的時間與順序都難以預估，這使得回音消除機制在多人網路回談中經常失效。 本研究提出藉由語音動態偵測(Voice Activity Detection-VAD)的方式分辨回音訊號，藉由本研究所提出的語音能量VAD判定機制，能夠有效區別正常語音與回音的差異，即可有效的消除回音，同時發揮靜音抑制（Slience Suppression）的效果，阻擋不含語音內容的封包，降低網路頻寬耗用。本研究以自行開發的VoIP軟體進行實地測試實驗，實驗中顯示，我們的方法能消除85%以上的回音。 / With the prosperous development of Internet technology, traditional phone service is being replaced gradually byVoice-over-IP (VoIP) technology. One of the critical problem that is yet to be improved is the echo problem. Due to the difference in working environment, conventional echo cancellation technology may not work well on VoIP system. The echo problem is becoming more critical as the number of participants in a talk session increases. As long as one user fails to depress echos, every other participant in the conference will be infected. The more participant, the higher probability of echo infection. We propose an energy based Voice Activity Detection (VAD) mechnism that effectively differentiate echo from speech signal. Our VAD algrouthm records a user’s speech volume, and based on this information to determine whether the frame is echo or not. By applying this mechnism to network conference, we can filter out echo frames and suppress slience at same time to save bandwidth consumption. We experimented on a self-developed VoIP software platform, the experiment result shows that our method can eliminate more than 85% of the echo.

為有效了解地表資訊，可視研究目的而利用各種遙測方法獲取地面影像，但不論目前常用之飛機、或是以衛星為載具之空中攝影測量等相關研究，在影像獲取方面，常面臨成本費用、機動性、比例尺需求、後續影像處理等問題。因此，本研究選定一小流域面積之河川作為實驗區，並設計以氣球為載具之數值航空攝影方式，裝載CCD攝影機以及數位攝影機，透過CCD攝影機，即時無線傳輸地面影像至監視螢幕，調整氣球至預定之位置後，遙控啟動數位攝影機之快門裝置。 　　研究設計之攝影方式曾實際應用於偵測河道之變遷，由成果影像中檢核點之精度結果，以及套疊正射糾正並鑲嵌後之河道影像與地形圖之成果，證明氣球載具之數值航空攝影方式，確能有效應用於大比例尺製圖之研究。 / Using a flexible and efficient way to obtain aerial images has been the primary purpose of this study. The balloon platform was used to take aerial images. A video camera and a digital camera were fixed together in a durable plastic box, and hung on the balloon. The video camera was used to monitor the ground view, and its image could be telemetered remotely and displayed on a LCD monitor arranged on the ground. Once monitoring the area of interest shown on the LCD, the shutter button of the digital camera was then pushed remotely and the interested image was taken. 　　The resultant images were ortho-rectified for analysis and comparison. The accuracy of aerial images was examined by check points. The results showed that the images achieved sub-pixel accuracy and were well-matched with the 1/1000 digital topographic maps. This expressed that it was really a useful and efficient method of taking large-scale images for a small research area. 　　At last, post-classification comparison method was introduced to detect change of the ortho-rectified images which were taken in three different periods. The classification maps and the from-to change class information clearly indicated the change of river way among various periods.

本論文的目的是利用人臉在彩色影像中所提供的多色彩空間資訊，來達成在變異度較大的光源中即時偵測人臉的任務。彩色影像所擁有的原始RGB色彩資訊，經過轉化到正規RGB以及HSV (色調、飽合、明度)等色彩空間後，擁有對光源變化反應減緩的特性。以此特性為基礎，在４個選定的色彩空間中定義８種不同的類赫爾特徵(Haar-like feature)，再利用推進演算法（Boosting algorithm）選出重要性最高的幾組特徵來進行對人臉的特徵。實驗結果顯示依此方法所產生的辨識器可在２點多秒內處理近百萬個次窗口(sub-window)，並對光源變化有相當程度的抵抗力。 / The main goal of this thesis is to detect human face under varying lighting condition by utilizing multiple color space information in real-time. Images of RGB color space can be converted into normalized RGB and HSV color spaces and thus reduce the interference of lighting condition. Base on this mechanism, we define 8 Haar-like features inside 4 selected color spaces, and then select the important features with boosting algorithm. Experimental results show that detectors constructed with our approach are able to process nearly one million sub-windows within 2.4 seconds, being robust to the changes of lighting conditions.

此篇論文中提出了一個新的環境監管規則衝突偵測方法，其運用於偵測智慧環境中使用者所制定的各種服務（在此稱為「規則」）。不同於傳統的偵測方法，此篇論文考量時間的因素，使具有時間感知的規則也能夠進行偵測。當考慮到時間的關係，這可能會使得原本衝突的規則變成有機會發生衝突或是不發生衝突，而且這也將會影響到整個系統的運作。我們基於特徵交互作用將衝突類型分成矛盾以及循環兩種，並且將時間因素也考慮到衝突類型當中，再針對這兩種衝突分別進行偵測。在矛盾這類型的衝突中，此篇論文利用時間邏輯的歸結原理來找出矛盾的規則；在循環這類型的衝突中，此篇論文將規則轉換成圖，然後再利用深度優先搜索找出圖中的返回邊，若是有返回邊存在也就能說明圖中存在著循環。此方法主要有兩個優點，第一個是在不考慮時間因素的傳統規則中，能夠有效的偵測出衝突。第二個是考慮時間因素的時間感知規則中，也能夠有效的偵測出衝突。在我們的實驗中，除了在模擬的規則資料庫中進行偵測外，我們也針對在實際的農業養殖場域中所使用的規則資料庫進行偵測。我們所提出的方法成功的偵測出智慧環境中使用者所制定的服務。 / A novel temporal-aware conflict detection method for environmental regulation which is used to detect the various services (referred to as “rules”) prescribed by users in a smart environment is proposed in this thesis. This method is different from the traditional detection methods. This thesis considers the temporal-aware so that temporal rules can be detected. When we take the temporal relationship into account, it is possible that the conflicting rules are turned into opportunistic conflicting rules or non-conflicting rules, and it also affects the operation of the entire system. We divide the conflict type into contradiction and cycle based on the feature interaction and take the temporal relationship into account, and then detect these two kinds of conflict types. In the conflict type of contradiction, this thesis uses the temporal resolution to detection conflicting rules. In the conflict type of cycle, this thesis transforms the rules into graphs, and then detect the cycles through the depth-first search (DFS) to find back edges. This method has two main advantages, the first one is that the conflict can be detected among the traditional rules which do not consider the temporal. The second is that the conflict can be effectively detected among the temporal rules. In our experiments, in addition to detecting in simulated rule databases, we also detect the rule databases used in the real greenhouse. The method we proposed successfully detect the services prescribed by the users in the intelligent environment.

智慧家庭環境是一個典型的分散式系統，在此類環境中的智慧服務大都由一至多個節點組成，例如一個智慧空調服務需要冷氣機、溫度感測器和邏輯判斷節點。然而，只要服務其中一個節點故障，整個服務就無法正常運作。由於居住在家庭中的大都是不具技術能力的使用者，故理想的智慧家庭服務，即使在有節點故障的狀況下，也應能在短時間內盡可能自動偵測與排除錯誤，使服務的運作不被中斷。本研究主要目的在於提出一個智慧家庭的強健服務管理系統，基於創新的複合式架構，結合集中式與非集中式錯誤偵測機制的特色，能在短時間內偵測到節點失效，進而恢復由於軟體所造成的節點故障或尋找待用節點，使得服務能繼續運行。 / Smart home systems are different from traditional computing systems. In a smart home system, a service is composed of several service nodes. For example, a smart air conditioning service needs a temperature sensor, an application logic, and an air conditioner. A service fails if one of its affiliating nodes fails. However, unexpected failures are undesirable for mission critical services such as healthcare or surveillance. Moreover, a smart home lacks professional system administrators. Users are generally unable to repair a service when it fails. Consequently, in a smart home system, the failed services have to be diagnosed and recovered automatically. In this paper describes a hybrid failure detection and recovery method for smart home environments. Experiments show that the proposed architecture is able to enhance overall availability of a smart home system in a short time.