利用深度學習辨識貓臉

Abstract

在台灣，每年有3千多隻的狗和貓走失，丟失寵物對於主人來說是非常難過的，這些走失的寵物也可能會對動物收容所造成負擔。儘管植入寵物晶片一直是解決走失寵物問題的方法之一，但它可能會給寵物帶來健康問題（例如發炎反應和癌症）。因此，我們需要一種非侵入式的走失寵物識別方法。本研究利用非侵入式的臉部辨識方法來識別個體貓，拍攝了一個包含150隻不同的個體貓、共900張貓照片的數據集，使用卷積神經網絡偵測臉上的五官（例如眼睛、鼻子和嘴巴）， 將五官的特徵利用特徵臉(Eigenface)量化後，使用支持向量機(Support Vector Machine)來進行辨識。本研究提出的方法達到了94.1％的識別準確度。

In Taiwan, there are more than 3 thousand dogs and cats missing every year. Losing pets could be extremely painful for owners. It also places burden on animal shelters in trying to return the pets to the owners. Although implanting microchips has always been a way to solve the missing pets problem, it may cause health problems (e.g., inflammatory reaction and cancer) to pets. Hence, a noninvasive approach for identifying missing pets is needed. This work proposed to identify cats noninvasively using face recognition. A database that contains 900 images of 150 different cats was developed. Facial parts (e.g., eyes, nose, and mouth) were identified using convolutional neural networks. The features of the facial parts (e.g., eigenface) were then qualified and were used for identifying the cats with support vector machines. The proposed method achieves an identification accuracy of 94.1 %..