基於人臉正面和側面特徵之唐氏症辨識

Down Syndrome Recognition Based on Frontal and Lateral Facial Features

劉書宏(Shu-Hong Liou)

指導教授 ： 陳永昌 ； 鐘太郎

國立清華大學/電機資訊學院/電機工程學系/碩士(2013年)

摘要

唐氏症(Down Syndrome)是目前新生兒群體中，較為常見的先天性遺傳疾病。其患者可能因為人體中第21條染色體上的基因異常，導致發展遲緩，甚至五官和身體上的部份畸變。而目前醫學上，對於新生兒的唐氏症疾病診斷，最準確的方法乃是採用染色體檢驗。然而，儘管染色體檢驗方式能夠達成較高的準確率，但此方法仍需花費昂貴的金錢和冗長的時間去做DNA分析。因此，我們希望能提出一個辨識方法，以用來進行簡單、快速的疾病診斷。其中使用者不需要擁有醫學上的專業知識，就可自動化的將判別作業交由系統處理，進而有效節省其人力資源。

正如之前所提，唐氏症患者由於基因的異常，因此五官部分相較於正常人會有部分明顯的特徵。例如 : 鼻子太扁、耳朵位置較低和兩眼距離過寬等等。因此，此篇論文主要想法就是希望藉由擷取這些明顯的人臉特徵，以區分出孩童中患有唐氏症的患者。我們的目的是希望能夠建立一個唐氏症疾病辨識系統，其能夠藉由輸入的人臉正面和側面影像，並擷取其中的主要特徵，再經過分類器的篩選，以確定患者是否患有唐氏症疾病。

在這個辨識系統中，我們針對提供的正面及側面的兩張人臉，設計了8個重要且常見的唐氏症臉部特徵來加以擷取：其中5個特徵是從正面擷取，剩下3個是從側面。其分別是針對兩眼的距離、眼睛的傾斜程度和形狀、鼻梁的傾斜度、鼻尖的上仰角度和耳朵在側臉中的相對位置來進行分析。而擷取出來的特徵，會再被輸入之前所訓練好的分類器加以分類，進而得到其識別的結果。

實驗結果顯示，其系統在唐氏症患者的辨識上，對於輸入的測試資料具有將近九成的成功辨識率。

關鍵字

唐氏症 ； 人臉

並列摘要

Down syndrome is one of the genetic disorders that usually happened to the humans. Children born with Down syndrome may have a delay in physical growth and a particular set of facial characteristics because of the presence of an extra 21st chromosome. In order to diagnose the patient with Down syndrome, a chromosomal test can be done to check for the extra chromosome and confirm the diagnosis. However, the testing consumes a large amount of time and money. Therefore, we want to develop a method to recognize the child with Down syndrome in a simple and fast way.

As mentioned before, patients with Down syndrome usually have multiple abnormal facial features. Therefore, we can develop a disease recognition system with the facial characteristics to help doctors examine the patient. By inputting the photo, including lateral and frontal face of the patient, into our proposed system, the important facial features of Down syndrome can be extracted. After the extraction, the features will be inputted into the classification system, with a recognized model trained before. The system can discriminate between the person with Down syndrome and normal one by the trained model. Hence, we can recognize the child with Down syndrome.

We select the eight important facial features that usually happen to the patient with Down syndrome. The characteristics are as follows: hypertelorism, almond eyes, up-slanting palpebral fissures, up-turned nose, flat nasal bridge and low-set ears.

Experimental results show that the accuracy rate of our disease recognition system is nearly 90% to recognize Down syndrome case correctly.

參考文獻

[1] G. W. Pong, “Down Syndrome Recognition Based on 2D Facial Features,” Master’s thesis, National Tsing Hua University, Hsinchu, Taiwan, R.O.C., July 2008.

[2] K. L. Jones, “Smith's recognizable patterns of human malformation,” 6th edition, pp. 7–11, Philadelphia, Elsevier Saunders, 2006.

[4] Y. Sivan, P. Merlob, and S. H. Reisner, “Assessment of ear length and low set ears in newborn infants,” Journal of medical genetics, 20.3, pp. 213–215, 1983.

[6] J. Cai and A. Goshtasby, “Detecting human faces in color images,” Image and Vision Computing, 18.1, pp. 63–75, 1999.

[7] J. Kovac, P. Peer, and F. Solina, “Human skin color clustering for face detection,” Vol. 2, IEEE, 2003.