狗鼻偵測與基於鼻紋的寵物狗身份辨識;Dog′s Nose Segmentation and Nose-Print Pet Identity Recognition

NCUIR > College of Electrical Engineering & Computer Science > Executive Master of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/83922

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/83922

Title:	狗鼻偵測與基於鼻紋的寵物狗身份辨識;Dog′s Nose Segmentation and Nose-Print Pet Identity Recognition
Authors:	陸威穎;Lu, Wei-Ying
Contributors:	資訊工程學系在職專班
Keywords:	狗鼻紋偵測;狗鼻紋辨識;生物辨識;決策融合;U-Net;LBP;GLCM;GGCM;PNN
Date:	2020-06-29
Issue Date:	2020-09-02 17:42:08 (UTC+8)
Publisher:	國立中央大學
Abstract:	現代人飼養寵物的情況與日俱增，卻也因此出現更多如走失、棄養等問題，造成社會的負擔。現存的身分識別技術大多採RFID晶片植入，然而植晶的行為除了對寵物產生健康上的疑慮導致飼主意願降低以外，還必須搭配儀器的掃描，並且可能面臨植入晶片的失效，以上種種皆會造成實施上的困難。本研究中提出一套基於影像處理的寵物犬身份辨識系統，透過犬隻的鼻紋紋理特徵進行身份識別。影像中的狗鼻區域藉由U-Net深度神經網路進行偵測及分割，分割出的狗鼻區域影像結合了LBP、GLCM、GGCM等三種紋理分析演算法進行特徵擷取，並將取得的特徵以PNN進行身份辨識。我們改善了U-Net的batch normalization策略，大幅提升了鼻紋區域的影像分割性能，結合後續的決策融合PNN分類器，使得身份辨識正確性達到優異的表現。在我們自建的10隻寵物狗資料庫，我們的方法可獲得98%辨識率。本研究規劃了一個完整的寵物身份辨識流程，並使用U-Net做為影像分割的基礎，在少量學習樣本的情況下，仍具有良好的切割性能，提供正確且完整的狗鼻區域進行身份辨識。;Nowadays, more and more people keep pets, as a result, more problems have happened, such as loss and abandonment, etc.…, causing a burden on society as well. Most of the existing identity recognition technologies are implanting the RFID chips. However, by this way, it will not only have the health concerns about pets, but also reduce the willingness of the owners to do so; besides, the chips implanted must be scanned by the instrument, and may face the failure of the implanted chips. The above reasons cause the difficulties in implementation of identity recognition. In this study, a pet dog identity recognition system based on image processing is proposed, which uses dog nose texture features for identity recognition. The dog-nose area in the image is detected and segmented by the U-Net deep neural network. The segmented dog-nose area image combines three texture analysis algorithms such as LBP, GLCM, and GGCM for feature extraction, and the obtained Features are identified by PNN. We improved U-Net′s batch normalization strategy, which greatly improved the image segmentation performance of the nose pattern area, combined with the subsequent decision fusion PNN classifier, so that the accuracy of identity recognition reached excellent performance. In our self-built 10 pet dog database, our method can obtain 98% recognition rate. This study planned a complete pet identity recognition process, and used U-Net as the basis for image segmentation. With a small number of learning samples, it still has good cutting performance and provides correct and complete dog nose area for identity recognition.
Appears in Collections:	[Executive Master of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	165	View/Open

社群 sharing

Loading...