由於輪胎瑕疵大都是裂痕或缺損，且輪胎顏色與瑕疵都為黑色，使得利用影像強化與辨識技術來檢測輪胎瑕疵的工作較不容易，又輪胎上有許多細微的紋路與花紋，更使得輪胎辨識極具挑戰性。因此，運用多類型的感測器作為輪胎瑕疵檢查是必要的，而使用多感測器的辨識技術，最重要的步驟為資訊融合，將多種資料型態匯入同一坐標系或投影至同一影像。

本論文提出一個結合深度影像與視覺影像之對位方法來找出兩種影像的相對幾何位置提供融合的資訊，所提出的方法是將雷射位移計截取深度資訊與彩色攝影機拍攝的彩色影像找到對應幾何座標資訊，讓辨識系統可同時擁有影像上的紋理特性以及深度影像資訊之物理特性，藉此提升影像資訊量。

本論文所提出的對位方法是使用局部對位方法來進行對位，由於輪胎上有許多紋路與花紋在全域對位上往往都會對位失敗，因此使用模板匹配和以尺度不變特徵轉換為基礎之比對來進行對位，其對位步驟可分為5步驟： (1) 使用模板匹配來找出深度影像和視覺影像模板相對位置，(2) 在把模板以外的資訊當作背景遮避掉，此時就只剩下視覺影像和深度影像的模板影像，(3) 接著在以視覺影像模板為基礎進行縮放並覆蓋至深度影像模板相對位置上，(4) 並以視覺模板影像和被覆蓋的深度影像進行以尺度不變特徵轉換為基礎之對位得到其對位特徵點，(5) 並把得到的對位特徵點對應到原圖上以完成局部對位。

關鍵詞：輪胎瑕疵檢測、資料融合、影像對位、深度影像、模板匹配、尺度不變特徵轉換

Since the color of most defects on a tire and the tire itself are black, the inspection of defects is a big challenge, especially only using visual images with image enhancement and recognition technology. A more reasonable approach is using multi-sensor system for inspection, e.g., laser, shearography,..etc. An important step of such a multi-sensor inspection system is the fusion of the multiple types of data acquired from those sensors. The fusion is transforming one data from its original coordinator system to the coordinator system of another data set or projecting one image onto another image. In this paper, an alignment approach is proposed for a depth image and a visual image of a tire captured by a laser displacement sensor and a color camera respectively.

While the global match usually misaligns the regular textures on a tire, the local match scheme proposed in this paper can exactly locate the logo textures on a tire. The proposed method consists of five steps: (1) using template matching to find the relative position of the depth image and visual image template, (2) highlighting the templates of the visual image and depth image, (3) based on the visual image template to scale and cover the relative position of the depth image template, (4) using SIFT to matching the visual template image and the covered depth image and (5) obtaining the corresponding pixel pairs of two images and finding the transformation.

Keywords: tire defect inspection, data fusion, image alignment, depth image, template matching, SIFT

目錄
摘要 i
Abstract ii
第一章 緒論 1
1.1 研究動機與目的 1
1.2 輪胎瑕疵 2
1.3 論文結構 5
第二章 深度影像轉換相關方法 6
2.1影像對比技術相關探討 6
2.1.1 Histogram Equalization (HE) 6
2.1.2 Binary Occupied Histogram Projection (BOHP) 7
2.1.3 HE和BOHP比較 8
2.1.4 基於BOHP/HE 的對比強化改良方法[1] 10
2.2 Sobel 邊緣偵測[6] 11
第三章 深度影像與視覺影像對位 13
3.1深度影像與視覺影像對位方法 13
3.2深度資訊轉深度影像 14
3.2.1 Scharr 邊緣偵測 18
3.3 視覺影像處理 19
3.4 比對演算法 20
3.4.1全域比對法 20
3.4.2以模板為基礎之局部比對法 21
3.4.3 模板匹配步驟 21
3.4.4 模板匹配比較方法 21
3.5 尺度特徵不變轉換(SIFT) 26
3.5.1 SIFT步驟 26
3.6 以特定區域來執行SIFT 31
3.7 影像融合 33
第四章 實驗結果與討論 36
4.1 系統開發環境 36
4.2 模板匹配結果 37
4.3 SIFT 對位結果 38
第五章 結論 48

[1] 王俊昌，一個實用的紅外線影像對比強化方法，國立交通大學電機資訊學院 資訊學程 碩士論文，2004
[2] Du Ming Tsai, Cheng Huei Chiang, Rotation invariant pattern matching using wavelet decomposition, Pattern Rcognition Letters , 2002, pp.191-201
[3] David G. Lowe, Distinctive Image Features from Scale Invariant Keypoints, Computer Science Department University of British Columbia , January 5, 2004,
[4] Histogram Equalization Wikipedia: https://en.wikipediz.org/wiki/Histogram_equalization
[5] Silverman, Jerry, Mooney, Jonathan, Ewing, William, Sato, Darryl United States Patent 5249241 Real time automated scene display for infrared cameras, September 28,1993
[6] M. Mikolajczyk, C. Schmid, Indeximg Based on Scale Invariant Interest Points, Proc. Conf. Computer Vision, 2001, pp.523-531
[7] Scharr edge detection opencv: https://docs.opencv.org/2.4/modules/imgproc/doc/filtering.html?highlight=scharr#cv2.Scharr
[8] SIFT blog: http://blog.csdn.net/dcrmg/article/details/52577555
[9] Jans Glagolevs, Karlis Freivalds, Logo Detection in Images Using HOG and SIFT, Advances in Information, Electronic and Electrical Engineering , November 24, 2017 ,pp.1-5
[10] Yuji Nakashima, Yoshimitsu Kuroki, SIFT Feature Point Selection by Using Image Segmentation, Intelligient Signal Processing and Communication Systems, November 6, 2017, pp.275-280
[11] Soo-Chang pei, Yu-Zhe Hsiao, Color Image Edge Detection Using Quaternion Quantized Localized Phase, Signal Processing Conference, 23 August, 2010, pp.2219-5491
[12] Yi Zhang, Xiaoyuan Han, Han Zhang, Liming Zhao, Edge Detection Algorithm of Image Fusion Based on Improved Sobel Operator, Information Technology and Mechatronics Engineering Conference, 3 October, 2017, pp.457-461
[13] You Lin, Xiu Chunbo, Template Matching Algorithm Based on Edge Detection, Computer Science and Society, 16 July, 2011, pp.7-9
[14] Xinghua Li, Haiyang Chen, Qinglei Chen, A Head Pose Detection Algorithm Based on Template Match, Advanced ComptuationalIntelligence, 18 October, 2012, pp.673-677
[15] Qiao Zhang, Huijie Gao, Zhen Kang, Fast Template Matching with Partial Skipping Using Sub-template, Audio, Language and Image Processing, 11 December, 2012, pp.889-892
[16] Yakun Chang, Cheolkon Jung, Automatic Contrast Limited Adaptive Histogram Equalization with Dual Gamma Correction, IEEE Access, 25 Junuary, 2018, pp.1-1
[17] Niloletta Bassiou, Constantine Kotropoulos, Color Histogram Equalization Using Probability Smoothing, Signal Processing Conference, 4 Septempber, 2006, pp.1-5
[18] Pritee Singh Rajpoot, Amit Chouksey, A Systematic Study of Well Known Histogram Equalization Based Image Contrast Enhancement Methods, Computational Intelligence and Communication Network, 12 December, 2015, pp.242-245
[19] Hu Cao, Li Tian, Jun Liu, Hui Wang, Songlin Feng, Color Image Enhancement Using Power-constraint Histogram Equalization for AMOLED, Computational Intelligence and Communication Network, 12 December, 2015, pp.1-5
[20] Uday Kumar Urimi, Mastan Rao Kongara, C. R. Patil, Real-time Implementation of Modified Adaptive Histogram Equalization for High Dynamic Range Infrared Images in FPGA, Computer Vision, Pattern Recognition, Image Processing and Graphic, 16 December, 2015, 1-4