音材施教--簡易音高辨識程式
我們製作了一個音準練習程式：使用者輸入聲音後，經由音頻辨識方法求出其頻譜中最高振幅之頻率，以之為音高，再將其與目標音高相較，得到其誤差率及走音程度。此外還可發出對應的鋼琴及正弦波的聲音，方便使用者校音。文中說明音頻辨識的方法，一些關於音樂的基本知識，微軟公司的wave 檔格式，及此系統之應用。我們使用FFT 辨識頻率，且將針對此部分演算法做簡單的說明，並探討如何達到所需之頻率準確度，及如何以較高效率辨識。目前誤差率已可達到1Hz 以下，判斷時間也在秒之內。儘管國內外也有一些具備相似功能之音樂編曲軟體，但其功能十分繁複，使用者常需花費數週時間學習，且價格常高至數千元，非一般使用者所能負荷。而這個程式不但使用方便，功能簡單，容易上手，且不需任何費用。We have developed a singing-practicing program: after input the sound, we judge its pitch from the corresponding spectrum; then we compare it with the selected one, and output the deviation, so that the users can see if they had been out of tune. Also, we provide the sound of the piano and the sine wave sound of the chosen pitch, which can help users get the right pitch. This report will briefly introduce the method of pitch recognition, some basics of music, and Microsoft's wave file format. In addition, we explain the application of this program. We use FFT to transform a sound wave into the spectrum, which are briefly explained in the article, too. Also we’ll discuss how to improve the accuracy and efficiency of the transformation. So far the deviation is less than one Hertz, and the recognition takes less than one second. Though there have been some commercial software with similar functions, they are often complicated to use, and cost a lot. This is not affordable to most users. On the contrary, our program is not only convenient and easy to use, but also has a simple user interface. What’s more, it costs no money!
Colour Blindness
Purpose of Research\r Colour blind individuals make up a large part of the population but are not usually considered. I aim to design an interactive guide that can be used when designing websites, social networking pages and various digital presentations to ensure that they are legible to colour blind individuals, by determining which colours generally confuse both red-green and blue-yellow colour blind people.\r Procedures\r I e-mailed my test, which consisted of 240 slides depicting every possible combination of the 16 basic web colours in the form of a frame, a heading-appropriate sized text and a paragraph-appropriate sized text, to a group of red-green and blue-yellow colour blind individuals, as well as an equally sized control group. I asked them to flip through the slides at a comfortable pace and note the slides that were difficult to read offhand, thus getting an idea of which colour combinations are not suitable for colour blind individuals, as well as the combinations that are not visible to those with normal colour vision.\r Data\r I wrote up my results using tables and stacked column graphs to determine which colour combinations were not visible to an unusually great number of test subjects. I then represented the information I collected in the form of a flash guide. This interactive guide allows one to choose a background colour from a wheel of 16 colours, and then be given a choice of appropriate text colours. The user is also presented with an example of the combination that was chosen, showing how the text appears on the background. The guide is easy to navigate and understand, can be posted on the internet or e-mailed, and is not overly technical so it can be used by designers as well as ordinary internet users.\r Conclusion\r Colour blind people are easily confused by colour combinations which might seem clear to us. My hypothesis was proved as I was able to set up a guide that would make it possible for both colour blind individuals and those with normal colour vision to read the text on websites, social networking pages and digital presentations.
AI 演化技術
此研究之重點在於如何建構一套人工智慧方法，人工智慧含多種進行方式，例如以類神經網路訓練近似於人腦之結構，而專家系統係於不同的科學領域內以自己之所知判斷。我們先以 John Holland 的 Genetic Algorithm （暫譯基因演算法，以下暫稱GA）的理論來實作出一套人工智慧系統之方法。Genetic Programming 係以ＧＡ為基礎之實作方法，主要的內容不出基本的演化定義；在這次的試驗中，首要為先定出程式欲演化成何種類型，在此我們先定義為排序型的演算法，經過分析實際程式之結構後定出適合基因元件的資料結構，分化為兩個部份進行，一為產生器，亦為突變，交配器，一為評審程式，亦為執行器，兩者使用相同之基本元件，再以不斷的交配和突變以達到全域最佳化。我們將兩種部份完成後，加上現有的PVM 分散運算函式庫來增加演算的速度。
Abstract The research mainly discusses AI. AI, involves in several types, for example, neural network(NN),which adopts human-like training method; Expert systems determine and make decision by what it has known. We will use John Koza's Genetic Programming theory to implement an AI system.Genetic Programming theory is based on GA. In our experiments, we have to define what kind of applications we want : a sorting application, which is divided into two parts - a producer (along with crossover and mutation operations) and an executor (along with judging operations), is an easyimplemented algorithm. Our program, also with PVM, will approaches global optimal after evolutions.
植物葉片自動辨識系統
在我們週遭環境中常可見到許多種類的植物，然而可以叫出名字的卻少之又少，或許我們可以查閱植物百之類的書籍，但是這類書籍通常多不在手邊，就算有了植物百科，也不易翻到顯示該種植物的正確章節。假如我們可以將想要認識的植物葉片影像取得後，透過網路將該影像傳送至植物葉片資料庫查詢，經過電腦的自動分析辨識後，再將結果傳送回來，這樣不是比查閱植物百科方便多了嗎？本研究提出一種利用輸入葉片的影像來進行植物資料庫辨識查詢的方法，藉著兩階段處理的策略及最佳權重組合式的特徵值來調校系統，以達到較佳的整體辨識效能，從實驗測試的結果得知，我們的策略與方法確實有效，有82%的查詢葉片可以被精確的辨識出來，而每次查詢的平均反應時間只要17.22 秒。In our living environment, there are many kinds of plants, but we can only name a few. We may consult an encyclopedia about plants, we always can’t find any encyclopedia with us. Besides, even if we have one, it won’t be easy to find out the proper section or the exact page immediately. How should we solve this problem? One significant improvement can be expected if the plant recognition can be carried out by a computer. First, we take a picture of the unknown plant’s leaf. Then, we transmit this image into a leaf database to recognize. After the recognition we will get the answer easily. By using a computer-aided leaf recognition system, non-professionals can also identify many plant species. Isn’t it much more convenient than checking the encyclopedia? In this study, we present an efficient method for leaf database retrieval by inputting leaf images. We use a two-stage approach and combined features with optimized weight to adjust the system to get the best system performance. The result of the experiment shows that our approach is workable and efficient. 82% leaves of the query images can be recognized accurately. And in general, the average response time only takes 17.22 sec per query.
國民身分證相片規格驗證暨浮水印防偽系統
政府全面換發國民身分證，並訂定新式身分證之規格，以防範遭不法偽造之情事，確保民眾權益。然而其中的照片規格，有十多條規格的限定，若用傳統的辨別方式，近1876 萬張照片是否合乎規定，那將耗費多少的人力呢？於是本次研究主題「新式國民身分證相片規格驗證暨浮水印防偽系統」即產生，設計一套程式，提高換發國民身分證的工作效率及確保換證使用相片的正確度。並且延伸研究出使用內崁式的數位浮水印〈Digital Watermarks〉，將全國民眾的身分證照片統一建立資料庫，並自動加入個人資料浮水印。日後，照片只需透過本程式分析，即可知道其姓名、身分證字號、有無犯罪前科等個人資料。希望藉此達到降低偽造身分證之犯罪率，以保護民眾之權利。 The government is launching to renew national identification cards with new norms, to avoid fake ones. However, there are more than ten limits on photos, it could be wasting time to discern by people. Thus, I launched a research on "The xamination on new national ID card photos and watermark forgery-proof system". The program will help both to enhance efficient renewal process and to use correct photos. Also we developed the embeded Digital Watermark technology, which would create a database for ID cards of the nation and could add personal infomation automatically. With the help of the program, simply run the photo analysis, we could find out the names, ID number, criminal background, etc.We hope to decrease crimes via fake ID cards, and protect the national right.
繪身繪影-正三角形磁磚設計方法與碎形密舖之研究
本研究主要以正三角形作為基本單元，透過窮舉討論得到正三角形邊的作用方式只有五種，再經由排列組合歸納出11 種正三角形密鋪磁磚設計方法。進一步，運用我們的研究結果，配合數學簡報系統製圖，創作新圖樣，也彌補了Escher 在手繪時所造成的誤差，達到完全密鋪的效果。碎形磁磚的部份，我們也依據其背後的數學理論創作幾套結構圖，利用結構解析，碎形密鋪磁磚將變得十分容易，學習者將可輕鬆製作富有創意的新圖樣。 ;This research mainly takes the regular triangle as the basic unit. Through the enumeration, we obtain that there are only five operations for edge of the regular triangle, and then 11 kinds of regular triangle design methods are induced. Even more, utilizing our findings and Mathematical Presentation System (Math PS), we created the new pattern which makes up Escher’s errors and achieves the tiling. As to Fractal Tiling, we create several sets of structure drawings according to its mathematics theory. Using structure analysis, the Fractal Tiling will become extremely simple, and the learner can make the rich creative new pattern easily.
倒車攝影機影像之變形校正
與傳統的鏡頭比較起來，廣角鏡頭可取得較大範圍的影像，所以廣泛地運用在很多地方，例如倒車攝影機、內視鏡、防盜警眼等，但是其影像會有很大的變形失真，容易造成人類眼睛的錯覺與誤判。本篇研究以倒車攝影機為例，使用了多項式函數來校正其影像的變形與失真，從程式實作的校正結果看來，我們的方法之效果還不錯，相信同樣的模式可用來校正其他使用廣角鏡頭的儀器設備之變形影像。 Compared with the traditional lens, wide-angle lens can make images of larger range, so they are widely used in a lot of places, such as move backward cine-cameras, laparoscopes, and burglarproof eyes. The drawback is that the images distort a lot and may cause the illusion to human eyes and lead to erroneous judgments. This reserch takes the backward cine-camera as example and utilizes multinomial function to correct the deformation and distortion of the image. The results show that the method we use are quite effective and the model can be applied to other equipment with wide-angle lens.
網路監控與家庭自動化系統之研究
The paper presents a new technique for the Internet monition and control system. The user not only can “watch” the current operation of remote home appliances from the client computer, but also directly control the behavior of remote home appliances through the Internet network. In the mean while, concerning the interface between users and appliances, the virtual reality is combined into the system so that users operate the system easily. First, virtual reality, CGI , HEML . I/O control code, 8255 I/O card and CCD are applied in the server computer to control and “watch” home appliances. Next, the client computer is connected to the sever computer through the Internet WWW(World Wide Web) network. Then, the user can control remote home appliances directly through the homepage. Finally, the experiment results are presented and discussed.本研究提出一種高效率且易維護的網路監控技術，從遠端確實“看”到監控系統現場動作，並下達控制指令。同時顧慮到使用者操作介面，結合虛擬實境，讓使用者可更輕鬆操作。首先，伺服端結合VR(虛擬實境)、CGI技術、基本的HTML、I/O控制指令、8255 I/O介面卡及CCD(網路攝影機)，應用於家用電氣，接著用戶端透過WWW全球資訊網連上伺服端首頁，然後依首頁之畫面，來達到遠端監控與家庭自動化之目的。