Advancements in Digital Image Perception and Recognition Development

1. 第一章 影像感知與辨識發展概況 國立雲林科技大學 資訊工程研究所 張傳育(Chuan-Yu Chang ) 博士 Office: EB 212 TEL: 05-5342601 ext. 4337 E-mail: chuanyu@yuntech.edu.tw Website: http://MIPL.yuntech.edu.tw

2. 第一章 影像感知與辨識發展概況 1.1 何謂「影像感知與辨識」？ 1.2 發展概況 1.3 應用實例

3. What is digital image processing? • What is digital image processing? • An image may be defined as a 2D function f(x,y)wherex, y: spatial coordinatesamplitude of f(x,y) is called the intensity or gray level • Digital image • Digital image processing • Processing digital images by means of a digital computer. • The element of a digital image • Picture elements, image elements, pixel

4. What is digital image processing? (cont.) • Three types computerized processes: • Low level • Basic operations such as noise reduction, contrast enhancement, and image sharpening. • Mid-level • Segmentation, classification • High-level • Understand the meaning of the recognized objects.

5. 應用實例 • 醫學影像處理 • From clinical point of view, • Single-modality-based diagnose • Outlining of boundaries of organs and tumors • Segmenting the suspiciously lesion. • Multi-modality-based diagnose • Image transformation • Image registration • The useful technologies for medical image processing: • 影像對位 (Image registration) • Image fusion • Image Reconstruction • Interpolation • 組織分類 • 腫瘤偵測 • 體積量測 • 良惡性判斷

6. Medical image segmentationAbdominal MR image Otsu’s Cheng’s CCBHNC

7. Medical image edge detection • MR Knee (a) The original MR knee joint based transverse image (b) result by the Laplacian-based method with threshold=5. N=7, d=1, (c) result by Marr-Hildreth’s method (N=9, d=1). (d) result by CHEFNN (p=q=2, A=0.01, B=0.03).

8. Image Registration • The registration of medical scans is the process of finding correct alignment or the proper spatial relation of one medical scan with reference to another.

9. Image Registration • Assessment of NPC with bony involvement CT T1 T2 PD

10. Image Transformation • Image transformation is the process that uses the space overlapping information from two sets of medical images acquired from different acquisition angles, and transforms them into images that are in the same angles and the same cross section.

11. Image Transformation • Truth horizontal axial image (b) Transformed image (c) Difference of (a) and (b)

12. Recurrent Nasal Papilloma Detection Spatiotemporal neural network for diagnosing recurrent nasal papilloma K-means 2nd PCA STNN Eigenimage filtering

13. The Experimental Results Spatiotemporal neural network for diagnosing recurrent nasal papilloma K-means 2nd PCA Eigenimage filtering STNN

14. The Experimental Results Spatiotemporal neural network for diagnosing recurrent nasal papilloma K-means 2nd PCA STNN Eigenimage filtering

15. The Experimental Results Spatiotemporal neural network for diagnosing recurrent nasal papilloma K-means 2nd PCA STNN Eigenimage filtering

16. Spatiotemporal-Hopfield neural network for diagnosing recurrent nasal papilloma K-means 2nd PCA Eigenimage filtering SHNC

17. Spatiotemporal-Hopfield neural network for diagnosing recurrent nasal papilloma K-means 2nd PCA Eigenimage filtering SHNC

18. Spatiotemporal-Hopfield neural network for diagnosing recurrent nasal papilloma

19. Spatiotemporal-Hopfield neural network for diagnosing recurrent nasal papilloma

20. Spatiotemporal-Hopfield neural network for diagnosing recurrent nasal papilloma

21. Spatiotemporal-Hopfield neural network for diagnosing recurrent nasal papilloma

22. Spatiotemporal-Hopfield neural network for diagnosing recurrent nasal papilloma 實驗結果 K-means 2nd PCA Eigenimage filtering Proposed method

23. Thyroid Segmentation • The original US image; • Result of locating suspicious thyroid region; • Result of image enhancement; • Result of segmented thyroid region by RBF neural network classifier; • Result of region growing; • Ground truth image (manually segmented by the physician).

24. Thyroid region segmentation before segmentation after segmentation

25. Thyroid region segmentation before segmentation after segmentation before segmentation after segmentation

26. 應用實例 • 人 • 人臉偵測 • 人臉辨識 • 表情辨識 • 膚質分析 • 手勢辨識 • 行為分析 • 指紋辨識 • 手靜脈辨識

27. What can we do from facial information?

28. Face Recognition Gender/Age Estimation Baby Degree Estimation Expression Estimation

29. Face Detection

30. Face Recognition

31. Face Recognition

32. Face Recognition

33. Face Recognition

34. Gender Recognition

35. Gender Recognition

36. Gender Recognition

38. Age Estimation

42. Baby Degree Estimation Estimates degree of baby-likeness (newborn to under 3 years)

43. Expression Estimation

44. Application Examples Auto focus/exposure for a specific person Auto focus/exposure based on baby detection

45. Application Examples Person search Baby search

46. Application Examples Mobile & Digital Camera Market Image Search

47. OMRON • https://www.omron.com/ecb/products/mobile/okao05.html

48. 應用實例 • 道路交通監控系統 • 車牌辨識系統 • 自動停車系統 • 特定方向偵測 • 逆向行駛的車輛偵測。 • 人車計數 • 行人與車輛統計，提供重要數據作為交警人力配置依據。人車計數智慧功能亦可應用於百貨公司或是參展會場等出入口，可以統計參觀人數。 • 特定物件型態偵測 • 車輛顏色、車型分析，可用來進行證據比對。

49. 應用實例 • 監視系統(Surveillance System) • 入侵偵測 • 遺留物／遺失物偵測 • 應用於捷運車站、機場、車站，偵測疑似爆裂物或不明遺留物。 • 警戒線偵測 • 若有人進入或有不明物體超越警戒線，系統會發出警報。 • 警戒區偵測 • 若有人進入或有不明物體進入警戒區域，系統會發出警報。

50. 應用實例 • 自動光學檢測 • 晶圓缺陷偵測 • LED結構缺陷偵測 • OCR (Optical Character Recognition) • 名片輸入辨識 • 可自動分析名片欄位屬性，並將辨識後的資料儲存手機的通訊錄中，使用者可輕鬆的將名片資料新增至手機聯絡人中