Texture Based Dog Nostril Image Segmentation

現今社會中飼養寵物的人越來越多，寵物中狗的比例又是最高的，根據行政院農委會統計，台灣在2015年就約有170萬隻飼養犬。隨著飼養寵物的風氣盛行，流浪狗衍生的問題越來越多，流浪狗議題越來越被人們重視。 如果我們細心觀察犬隻的鼻子，會發現上面有許多紋路，而這些紋路與人類的指紋一樣，具有唯一識別性，因此只要能識別出狗鼻紋就能對犬隻進行完善的管理。狗鼻紋辨識系統主要分成兩大部步驟：狗鼻孔影像切割與狗鼻紋影像辨識，而本論文是針對狗鼻孔影像切割進行研究。 本研究目的為從狗鼻子影像中切割出狗鼻孔，並精準地找出兩鼻孔的中心點，以便未來進行犬隻辨識用。實驗影像是從台中數個公立收容所中以高解析度相機拍攝狗鼻子取得，結果顯示本研究鼻孔平均切割準確率可以達到85.52%，狗鼻子平均中心點差僅有10.4 pixel。In todays’ society, having pets is becoming a norm. The proportion of dogs as pet is the highest. According to statistics of Council of Agriculture, Executive Yuan, Taiwan has about 1.7 million dogs in 2015. With the trend for having pets, problems derived from stray dogs are becoming a serious issue. If we observe the dog''s nose carefully, you will notice there are various patterns, these patterns are similar to human’s fingerprints, which means each of them are unique. Thus we can manage these dogs by recognizing their nose prints using dog nose prints recognition system. This system can be divided into two major steps: dog nostril image segmentation and dog nose prints recognition, and this thesis is for dog nostril image segmentation research. The purpose of this study is to segment the dog''s nostril from the dog''s nose image and find the center of the nostrils for future identification of the dog. The results showed that the average segmentation rate of nostrils in this study was 85.52%, and the average center of the dog''s nose was only 10.4 pixels.致謝 i 摘要 ii Abstract iii 目錄 iv 圖目錄 vi 表目錄 ix 第一章 緒論 1 1.1 研究背景 1 1.2 研究動機與目的 4 1.3 論文架構 7 第二章 相關文獻回顧 8 2.1 Gaussian Blur 8 2.2 Gray Level Co-occurrence Matrix (GLCM) 9 2.3 Otsu 13 2.4 Connected-component Labeling 15 2.5 Morphological Image Processing 17 2.5.1 Dilation and Erosion 17 2.5.2 Opening and Closing 19 2.6 Genetic Algorithm 20 第三章 基於紋理分析狗鼻孔影像切割 23 3.1 整體流程 23 3.2 切割前處理 24 3.3 精密切割判斷 25 3.4 初步切割 27 3.4.1 初步前處理 28 3.4.2 初步特徵擷取 31 3.4.3 初步去噪與切割 32 3.5 精密切割 36 3.5.1 精密前處理 37 3.5.2 精密特徵擷取與降噪 38 3.5.3 精密去噪與切割 40 第四章 參數訓練 43 4.1 切割參數訓練 44 第五章 實驗結果 46 5.1 影像來源 46 5.2 切割前處理準確率評估 48 5.3鼻孔切割準確率評估 49 5.4 鼻孔中心位置準確率評估 52 5.5 實驗階段影像及參數設定 55 5.6 與現有方法比較 57 第六章 結論與未來展望 63 參考文獻 6