本論文應用稀疏表示分類演算法來對結核菌顯微影像上的組織做分類，主要藉由自動對焦所得的影像來偵測結核菌候選區域，並針對候選區域所擷取的特徵進行稀疏表示運算來加以判讀，再將其分類的結果儲存，方便醫檢師查看和確認。
整個辨識流程可以分為兩個階段：結核菌偵測與結核菌辨識。在結核菌偵測部份，我們進行四項處理：影像類別分類、色彩正規化、結核菌候選區域檢測與特徵擷取。利用影像的亮度標準差和色彩飽和度將影像區分成三大類，針對每張影像所屬類別，分別訓練其所屬參數，再藉由色彩正規化將影像進行各自類別的色彩轉換，以降低色彩上的差異性。這兩項處理的目的是為了讓抹片影像在色彩上能標準化，以利進行後續的處理。在結核菌候選區域的檢測方面，我們使用了線性鑑別分析法(Linear Discriminant Analysis, LDA)來進行色彩的分割，經過標記(Labeling)與型態學上的處理，結核菌的候選區域即可被擷取出來，進而進行特徵擷取。
在結核菌辨識部份，先將候選區域的辨識結果分成三種：結核菌、疑似結核菌和非結核菌。根據稀疏表示分類演算法(Sparse Representation-based Classification, SRC)的規定，需對三種類別的影像各自創建一個字典。因各家醫院的影像類型皆不同，為了解決固定字典只能適應部份類型影像的問題，我們對候選區域的三種結果各自創建子字典，並進行字典學習(K-SVD)，再由這三個子字典組成一個母字典，以利於母字典能適用任何類型的影像。最後輸入特徵資料進行稀疏表示運算，比較輸入的資料跟字典內三種結果的誤差值，根據誤差值來分類輸入的資料屬於哪種結果。
我們的辨識結果是敏感度達到95% 和鑑別度達到 94.26%，與過去的自動結核菌辨識系統的辨識結果相近，而在該系統中，分類器的訓練時間需耗費將近一周，在此論文中，只需花費三天，幾乎節省約一半的時間。

In this thesis, a new image analysis system has been proposed to classify and identify the mycobacterium tuberculosis from the microscopic images. We apply Sparse Representation-based Classification (SRC) to identify the features extracted from image candidate regions, and the images are obtained by an auto-focusing system. After the computation is done, the results of the classifications will be saved, so that the medical technicians can review and check them later.
There are two stages for tuberculosis identification: detection and identification of the mycobacterium tuberculosis. In the detection of mycobacterium tuberculosis, there are four processing steps: image categorization, color normalization, detection of the candidate regions and extracting features. By using the standard deviation of brightness and the color saturation, images can be divided into three groups. The image color parameter training and color normalization are performed individually for each group. The purpose is to make the color distribution of the images become more consistent. It would be beneficial for the subsequent processing. For the detection of the candidate regions, we use Linear Discriminant Analysis (LDA) to do color segmentation. After labeling and applying the morphology processing, candidate regions can be found, from which we can extract the features.
In the identification part, we classify the results of candidate regions into three cases: Tuberculosis (TB), Suspected-Tuberculosis (STB), and Non-Tuberculosis (NTB). According to the rule of SRC, it is needed to construct a dictionary for each image group. The types of the images in each hospital are different. To solve the problem that the fixed dictionaries can only adapt to some types of the images, we construct the parent dictionaries which are built by three sub-dictionaries of TB, STB and NTB. These sub-dictionaries are trained by K-SVD. These parent dictionaries are beneficial to adapt to different types of images. Finally, we use the extracted features of input images and parent dictionaries to compute the sparse coefficients. And then the errors of the representation results are computed with input data based on sub-dictionaries, the identification results can be found according to these errors.
The sensitivity of mycobacterium tuberculosis identification is 95%, and the specificity is 94.26%; these results are similar to the results of previous system. However, in the previous automatic mycobacterium tuberculosis identification system, the time for training the classifiers is about one week, but it takes only three days for our new system, we almost conserve half of the time.

[1] 行政院衛生署, “結核病十年減半全民動員計畫,” 行政院衛生署, 2006。
[2] 行政院衛生署, “結核病十年減半全民動員第二期計畫,” 行政院衛生署, 2012。
[3] 行政院衛生署, “台灣結核病防治年報 2013,” 行政院衛生署, 2014。
[4] 林展頤, “應用自動彩色顯微影像分割之結核菌偵測與評估,” 成功大學資訊工程所碩士論文, 2009。
[5] 陳芷涵, “自動結核菌判讀系統,” 成功大學資訊工程所碩士論文, 2013。
[6] C. F. F. CostaFilho, P. C. Levy, C. M. Xavier, M. G. F. Costa, L. B. M. Fujimoto, and J. Salem, “Mycobacterium Tuberculosis Recognition with Conventional Microscopy,” IEEE Conf. EMBS San Diego, California USA, 2012.
[7] D. Donoho, “For Most Large Underdetermined Systems of Linear Equations the Minimal l1-Norm Solution Is Also the Sparsest Solution,” Comm. Pure and Applied Math., vol. 59, no. 6, pp. 797–829, 2006.
[8] E. Reinhard and M. Ashikhmin et al., “Color transfer between images,” IEEE Computer Graphics and Applications, vol. 21, no. 5, pp. 34–41, 2001.
[9] E. Cande`s, J. Romberg, and T. Tao, “Stable Signal Recovery from Incomplete and Inaccurate Measurements,” Comm. Pure and Applied Math., vol. 59, no. 8, pp. 1207–1223, 2006.
[10] E. Cande`s and T. Tao, “Near-Optimal Signal Recovery from Random Projections: Universal Encoding Strategies?” IEEE Trans. Information Theory, vol. 52, no. 12, pp. 5406–5425, 2006.
[11] G. H. Golub, W. Kahan, “Calculating the singular values and pseudo-inverse of a matrix,” Journal of the Society for Industrial and Applied Mathematics: Series B, Numerical Analysis 2 (2): 205–224, 1965.
[12] J. A. Tropp, “Greed is good: Algorithmic results for sparse approximation,” IEEE Trans. Inf. Theory, vol. 50, pp. 2231–2242, Oct. 2004.
[13] J. Wright, A. Y. Yang, A. Ganesh, S. S. Sastry, and Y. Ma, “Robust face recognition via sparse representation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 31, no. 2, pp. 210–227, Feb. 2009.
[14] K. Veropoulos, C. Campbell, and G. Learnmonth, “Image processing and neural computing used in the diagnosis of tuberculosis,” IEEE Colloquium on Intelligent Methods in Healthcare and Medical Applications, pp. 1–4, 1998.
[15] L. Zhang, W. D. Zhou, P. C. Chang, J. Liu, Z. Yan, T. Wang, and F. Z. Li,“Kernel sparse representation-based classifier,” IEEE Trans. Signal Process., vol. 60, no. 4, pp. 1684–1695, Apr. 2012.
[16] M. E. Wall, A. Rechtsteiner, and L. M. Rocha, “Singular value decomposition and principal component analysis,” In D.P. Berrar, W. Dubitzky, M. Granzow. A Practical Approach to Microarray Data Analysis. Norwell, MA: Kluwer. pp. 91–109, 2003.
[17] M. Forero, F. Sroubek, and G. Cristbal, “Identification of tuberculosis bacteria based on shaped and color,” Real Time Imaging, vol.10. no.4, pp.251–262, august 2004.
[18] M. Aharon, M. Elad, and A. Bruckstein, “K-SVD: An Algorithm for Designing Overcomplete Dictionaries for Sparse Representation,” IEEE Trans. Signal Processing, vol. 54, no. 11, pp. 4311–4322, Nov. 2006.
[19] M.G. Costa, C.F.M. Filho, J.F. Sena, J. Salem, and M.O. de Lima, “Automatic identification of mycobacterium tuberculosis with conventional light microscopy,” IEEE Engineering in Medicine and Biology Society, pp. 382–385, 2008.
[20] R. C. Gonzalez and R. E. Woods, “Digital Image Processing,” Prentice Hall, 2002.
[21] R. Khutlang, S. Krishnan and R. Dendere, A. Whitelaw, K. Veropoulos, G. Learmonth, and T.S. Douglas, “Classification of Mycobacterium Tuberculosis in Images of ZN-Stained Sputum Smears,” IEEE Trans. Information Technology in Biomedicine, vol. 14, no. 4, pp. 949–957, 2010.
[22] R. Rulaningtyas, A. B. Suksmono, and T. L. R. Mengko, “Automatic Classification of Tuberculosis Bacteria Using Neural Network,” IEEE Conf. Electrical Engineering and Informatics., 2011.
[23] S. Mallat and Z. Zhang, “Matching pursuits with time-frequency dictionaries,” IEEE Trans. Signal Process., vol. 41, no. 12, pp. 3397–3415, 1993.
[24] S. Chen, D. Donoho, and M. Saunders, “Atomic Decomposition by Basis Pursuit,” SIAM Rev., vol. 43, no. 1, pp. 129–159, 2001.
[25] World Health Organization (WHO), “GLOBAL TUBERCULOSIS REPORT 2014,” World Health Organization (WHO), 2015.
[26] Y. C. Pati, R. Rezaiifar, and P. S. Krishnaprasad, “Orthogonal matching pursuit: Recursive function approximation with applications to wavelet decomposition,” in Conf. Rec. 27th Asilomar Conf. Signals, Syst. Comput., vol. 1, 1993.
[27] Y. Shi, Y. Gao, Y. Yang, Y. Zhang and D. Wang, “Multimodal Sparse Representation -Based Classification for Lung Needle Biopsy Images,” IEEE Trans. Biomedical Engineering, vol. 60, no. 10, Oct. 2013.
[28] Y. Guo, Y. Wang, D. Kong and X. Shu, “Automatic Classification of Intracardiac Tumor and Thrombi in Echocardiography Base on Sparse Representation,” IEEE Biomedical and Health Informatics. vol. 19, no. 2, March. 2015.