隨著自駕車的級數越來越純熟，透過其中的行人偵測系統可以減少意外事故的發生並且增加行車安全。但當自駕車因為使用的時間過長，可能會使影像處理電路因為高溫、電源切換、製程上的缺陷以及電路老化等，有可能使影像產生錯誤，使得行人偵測的判斷出現錯誤的情況，這時將會讓路上行人的安全受到極大的威脅。
其它文獻雖然也有對影像容誤進行分析，但都是針對人類視覺所做的分析結果，因此在本論文中，我們將以行人偵測為例，分析以電腦視覺的角度來評估行人偵測系統的容誤程度，接著再分析人類視覺與電腦視覺對影像容誤能力的標準。本論文分析的結果顯示，電腦視覺相較於人類視覺對影像錯誤品質的容誤能力比較大。因為現今大多數的影像品質評估方法都是只考慮人類視覺的評估結果，這些方法將無法應以評估錯誤影像的錯誤程度是否被電腦視覺接受之影像，所以我們無法應用在電腦視覺上。為了能夠將影像容誤的概念應用在電腦視覺上來增加行人偵測系統電路的準確度與可靠度，我們在行人偵測系統中加入一個無參考的影像容誤評估方法，藉由加入此電路，我們可以提高行人偵測系統的可靠度，並同時提升行人偵測電路的使用時間。
本論文中為了評估此方法的有效性，因此我們也將支援向量機(Support vector machine，SVM)用以實現容誤評估。這個方法主要是透過機器自我學習的方式來處理影像品質評估的問題，藉由自我學習，SVM能夠預測一張影像的品質是否可以被電腦視覺接受。
在過去的文獻中，假性邊緣偵測的準確度最高可達93.39%，但如果應用在電腦視覺上的話，準確度會下降為86.22%。在本論文中，我們會分析並改變此方法的臨界值以應用在電腦視覺上，並探討此方法在這個情況下將無法適用之情況，最後我們會將本論所提出之無參考容誤評估方法與假性邊緣偵測和SVM的結果進行比較。實驗結果顯示，應用在電腦視覺上，我們所提出之方法可達約93.48%的準確度，而且在行人偵測系統可靠度為0.99以上的情況下，可以讓行人偵測系統的使用時間延長1年以上。

With the rapid development of autonomous cars, pedestrian detection systems of autonomous cars are developed to enhance the safety of driving. However, when the system is used for a period of time, high temperature, power switch, defects, and wear-out of the image processing circuits of the system may result in erroneous images. This may validate the system and thus threaten the safety of pedestrians.
In the literature, there are some previous work addressing evaluation of the image error-tolerability, but most of these work are for human vision. In this thesis, we investigate and compare image error-tolerability for the human vision and computer vision. Our results show that the image error-tolerability for computer vision is much larger than human vision. To classify the acceptability of erroneous images for computer vision, we propose a no-reference error-tolerability evaluation method for computer vision, and apply this method to the pedestrian detection system. This method can enhance the reliability of the pedestrian detection system and extend the lifetime of the pedestrian detection system.
To better demonstrate the strength of the proposed method, we also develop a SVM-based method as a basis for comparison, which mainly processes images through machine learning. The error-tolerability classification result is then generated. We compare the classification results of our method, previous work and the SVM based method. The results show that our method has the best cost-effectiveness. About 93.48% is achieved for the proposed method. Also, with the high reliability of 0.99 for the pedestrian detection system, the lifetime can extend about additional one year.

論文審定書 i
致謝 ii
摘要 iv
Abstract vi
目錄 vii
圖次 ix
表次 xi
第一章 概論 1
1.1 研究動機 1
1.2 研究貢獻 2
1.3 論文大綱 3
第二章 研究背景及相關文獻回顧 4
2.1 影像品質評估參數 4
2.1.1 PSNR(Peak Signal to Noise Ratio) 4
2.1.2 SSIM(Structural Similarity) 5
2.1.3 FSIMc(Feature Similarity Index with chrominance) 7
2.2 影像容誤測試方法 8
2.2.1 Reserved Pixel Pairs (PPR) 8
2.2.2 假性邊緣偵測(False edge detection) 10
2.3 聚合通道特徵(Aggregated Channel Features，ACF) 14
2.3.1 簡介 14
2.3.2 行人特徵擷取 15
2.3.3 分類器 17
2.3.4 訓練資料 19
第三章 應用於行人偵測之影像容錯評估與測試 20
3.1 簡介 20
3.2 容誤(Error-Tolerability) 20
3.3 人類視覺與電腦視覺之標準差異 21
3.3.1 人類視覺 21
3.3.2 電腦視覺 21
3.3.3 影像品質評估參數與電腦視覺之分析結果 22
3.4 參考影像 26
3.5 電腦視覺應用之無參考容誤評估方法 27
3.4.1 極端值與邊緣檢測 27
3.4.2 SVM(Support Vector Machine) 31
3.4.3 訓練資料(Training data)與測試資料(Testing data) 34
3.4.4 Confusion matrix 34
3.4.5 訓練集(Training set)與驗證集(Validation set)的比例 35
3.4.6 核函數(Kernel)與訓練模型 37
3.4.7 訓練模型與測試資料 39
第四章 錯誤影像產生方法與實驗結果分析 43
4.1 錯誤影像產生方法 43
4.1.1 JPEG2000影像壓縮標準 43
4.1.2 單一固接錯誤(Single stuck-at fault) 43
4.1.3 反離散小波轉換電路(IDWT)模擬錯誤方法 44
4.1.4 編碼器(Encoder)模擬錯誤方法 45
4.1.5 錯誤影像分析 46
4.2 實驗結果 48
4.2.1 準確度分析 48
4.2.2 可靠度分析(Reliability) 50
4.2.3 軟體執行效能比較 59
4.2.4 硬體成本效能比較 61
第五章 結論與未來展望 63
第六章 參考文獻 64

[1] B. Javidi, Image Recognition and Classification: Algorithms, Systems, and Applications, CRC Press, 2002.
[2] M. A. Breuer, S. K. Gupta and T. M. Mak, “Defect and error-tolerance in the presence of massive numbers of defects,” IEEE Design & Test of Computers, 21(3): pp. 216-227, 2004.
[3] L. Zhang, D. Zhang, X. Mou and D. Zhang, “FSIMc: A feature similarity index for image quality assessment,” IEEE Trans. on Image Processing, 20(8): pp. 2378-2386, 2011.
[4] Z. Wang, A. C. Bovik, H. R. Sheikh and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,” IEEE Trans. on Image Processing, 13(4), pp. 600-612, 2004.
[5] N. Ponomarenko, L. Jin, O. Ieremeiev, V. Lukin, K. Egiazarian, J. Astola, B. Vozel, K. Chehdi, M. Carli, F. Battisiti and C.-C. Jay Kuo, “Image database TID2013: peculiarities, results and perspectives,” Signal Processing: Image Communication, 30: pp. 57-77, 2015.
[6] N. Ponomarenko, V. Lukin, J. Astola and K. Egiazarian, “Analysis of HVS-Metrics’ properties using color image database TID2013,” Advanced Concepts for Intelligent Vision Systems, 9386: pp. 613-624, 2015.
[7] T.-Y. Hsieh and Y.-H. Peng, “Filtering-based error-tolerability evaluation of image processing circuits,” Proc. IEEE Int’l. On-Line Testing Symp., pp. 132-137, 2015.
[8] Y.-H. Peng, Development and Implementation of Efficient Test Methods for Image Processing Circuits, Master’s Thesis, Department of Electrical Engineering, National Sun Yat-sen University, Kaohsiung, Taiwan, 2015.
[9] T.-Y. Hsieh, Y.-H. Peng and K.-C. Cheng, “Structural variance based error-tolerability test method for image processing applications,” IEEE Trans. on Computer Aided Design of Integrated Circuits and Systems, 37(2): pp. 485-498, 2018.
[10] K.-C. Cheng, A Low-Cost Dependability Evaluation and Grading Method and Its Hardware Implementation for Image Processing Circuits, Master’s Thesis, Department of Electrical Engineering, National Sun Yat-sen University, Kaohsiung, Taiwan, 2015.
[11] T.-Y. Hsieh and C.-R. Chen, “A no-reference error-tolerability test methodology for image processing applications,” Int’l. Test Conf. in Asia, 2018.
[12] C.-R. Chen, A No-Reference Error-Tolerability Test Methodology and Its Hardware Implementation for Image Processing Circuits, Master’s Thesis, Department of Electrical Engineering, National Sun Yat-sen University, Kaohsiung, Taiwan, 2018.
[13] W. K. Pratt, Digital Image Processing. 3rd ed., New York: Wiley-Interescience, 1976.
[14] J. Canny, “A computational approach to edge detection,” IEEE Trans. Pattern Analysis and Machine Intelligence, 8: pp. 639-643, 1986.
[15] D. Marr and E.C. Hildreth, “Theory of edge detection,” Royal Soc. London B, 207: pp. 187-217, 1980.
[16] N. Dalal and B. Triggs, “Histograms of oriented gradients for human detection,” in CVPR, 2005.
[17] P. Dollár, R. Appel, S. Belongie and P. Perona, “Fast feature pyramids for object Detection,” IEEE Trans. on Pattern Analysis and Machine Intelligence, 36(8): pp. 1532-1545, 2014.
[18] J. Friedman, T. Hastie, and R. Tibshirani, “Additive logistic regression: a statistical view of boosting,” The Annals of Statistics, 38(2): pp. 337–374, 2000.
[19] C.-C. Chang, C.-J. Lin, "LIBSVM: A library for support vector machines," ACM Trans. on Intelligent Systems and Technology, vol.2, 2011.
[20] G. K. Wallace, “The JPEG still picture compression standards,” IEEE Trans. on Consumer Electronics, 38(1): pp. xviii-xxxiv, 1991.
[21] OpenJPEG library: an open source JPEG2000 codec, http://www.openjpeg.org/
[22] J.-R. Chen, Development of A Behavior-Level Error Simulation and Analysis Platform and Its Applications to Image Processing Circuits, Master’s Thesis, , Department of Electrical Engineering, National Sun Yat-sen University, Kaohsiung, Taiwan, 2016.
[23] D. Mahmoodi, A. Soleimani, H. Khosravi and M. Taghizadeh, “FPGA simulation of linear and nonlinear support vector machine,” Journal of Software Engineering and Applications, 4(5): pp. 320-328, 2001.
[24] T.-Y. Hsieh, T.-A.Cheng and C.-R. Chen, "Error-tolerability evaluation and test for images in face detection applications," IEEE Asian Test Symp., pp. 201-206, 2017.