可攜式手勢辨識預測X光劑量系統

Abstract

近年來，X光的技術被廣泛的運用在醫學上，當患者內部有創傷或急病時，X光可以有效的檢視病人身體狀況並且對症下藥，然而，過多的X光劑量會對人體造成傷害，並且容易使人罹癌和病變，精準的X光劑量可以有效的降低危害風險，X光劑量預測系統是將患者的身體資訊(厚度、組織密度、骨骼……等等)傳送到雲端資料庫，藉此找尋匹配的X光劑量。 此篇論文中，著重在手勢的配對，因為手擁有很多的關節及變化，而不同的手勢會造成不同的厚度，這些厚度影響著X光劑量的決定；因此，在此系統下模擬著病患的手和資料庫中的X光影像的匹配；為了能夠達到良好的匹配，手的尺寸必須達到一致性，所以在提出的演算法中，藉由手掌的大小去調整手的尺寸來達到正規化。 手勢辨識早已被廣泛的使用和研究，例如：HU moment和principle component analysis (PCA)，傳統的方法只適用於一般的RGB影像匹配，當資料庫中的RGB影像全換成X光影像後，因為骨骼會讓原本手的外框和形狀變形，所以造成辨識精準度下滑；在提出的演算法中，特別設計特徵的擷取來描述影像，包含手指的面積、長度和角度，而這些特徵可以用來分辨一些形狀相似的手勢，並且當資料庫中的影像改變時，也能夠維持手勢的辨識率。

In recent years, the technologies of the X-ray are widely used in the medical treatment. When the patients get hurt in the bodies, the X-ray is helpful to diagnose; however, overdoses of the X-ray may cause unexpected damages to patients. In prediction X-ray system, the information of the patients including the thickness, tissue, skeleton will be sent into the cloud to find out the matching dosage of the X-ray to reduce the risks. In this thesis, the hand postures are focused on, because the postures are varied with the joints. Additionally, the thickness of each hand posture also influences the dosage of the X-ray; hence, the objective of this thesis is to find out the matching X-ray images in the database with the patients’ hands. To match the hands well, the size of the palm is used to adjust the size of the hand in the proposed algorithms to further determine the correct X-ray dosages. When the RGB images are changed into the X-ray images in the database, the accuracy of the traditional methods like HU moment and principle component analysis (PCA) decreases because the skeletons distort the shapes and contours of the hands. Therefore, we proposed a new algorithm, by including more features, such as area, length, and angle to recognize the similar gestures and maintain the accuracy when the database is changed to X-ray.