- 學位論文
基於人臉五官特徵之唐氏症辨識
Down Syndrome Recognition Based on 2D Facial Features
摘要
摘要: 目前醫生對於罕見疾病患者的診斷，最準確的方法是採用染色體檢驗，但是，這需要昂貴的金錢，以及冗長的時間去做染色體的DNA分析• 在這篇論文的工作裡，我們希望建立一個僅僅需要人臉資訊的疾病辨識系統，來幫助醫生做最簡易、快速的疾病診斷• 某些染色體上的基因異常，常會造成五官或是身體上的畸變，我們可藉由某些特徵來做疾病的辨識•因為這是一個新的研究方向，為了驗證這個想法是否可行，初步的工作，我們一開始先將問題縮小在幼童的唐氏症辨識• 在新生兒出生時，大約會有1/500 的機會，因為染色體異常，而導致新生兒生長畸變，這些變異有時可直接從五官看出，例如 : 五官比例特殊、鼻子太扁、耳朵外觀怪異、臉頰過於肥厚扁平•我們利用人臉五官特徵而發展出一套醫療檢測系統，幫助醫生對於新生兒做疾病的臨床診斷•唐氏症又稱Trisomy 21，此種疾病是由於存在了三條第21號染色體• 在這個疾病辨識系統裡，我們針對正面及側面的兩張人臉，設計了15個重要臉部特徵，其中8個特徵是從正面擷取，剩下7個是從側面，分別針對兩眼的相對位置、兩眼大小、眼睛的傾斜程度、鼻子大小、鼻子的凹陷程度、耳朵與鼻子的相對位置、耳朵的形狀比例，並且提出了快速搜尋這些特徵的方式• 實驗結果的顯示，在做唐氏症患者的辨識上，我們具有9成以上的成功辨識率•
並列摘要
Abstract In recent years, doctors adopt the most accurate method which is called “The examination of the chromosome” for diagnosis of rare diseases. But it costs much and long time to analyze the DNA on the chromosome. In the work of this thesis, we propose the disease recognition system that uses the information of the facial features to help doctors diagnose simply and rapidly. Since the specific abnormality of the chromosome often affects the face or body as to be deformed, we can recognize the disease by the specific facial deformation features. Since this study is a novel research topic, in order to prove the feasibility of the idea, first of all, we begin to focus on the recognition of the Down syndrome in the infant faces. When a baby is born, there is 1/500 possibility to be deformed because of the abnormal chromosome. Sometimes these differences can discern from the facial features, for instance, strange ratio of the facial feature, flat nose, strange contour of the ear, plump chin and so on. We develop the system of the medical diagnosis by the facial features to help doctors make the clinic diagnosis of the disease of infant babies. Down-Syndrome, also known as Trisomy-21 due to the presence of a third twenty-first chromosome, is one of the most common and well known birth anomalies. In our disease recognition system, we consider fifteen important facial features from the front face and lateral face. Eight of which are extracted from the front face, and the after seven features are extracted from the lateral face. They stand for the relative position of the both eyes, size of the both eyes, the slope degree of the eye, the size of the nose, the concave degree of the nose, the relative position between ear and nose, and the shape ratio of the ear. We design a fast algorithm for feature recognition effectively. Experimental results show that the accuracy of our system to recognize Down-Syndrome case correctly is about 90%.