›
›

Bài giảng Nhập môn thị giác máy tính - Chapter 5: Binary image analysis - Lê Thanh Hải

Download
Chapter 5 – Binary Image  
5-1  
Chapter 5  
Binary Image Analysis  
Bastian Leibe, RWTH Aachen University  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-2  
Binary Images  
• Just two pixel values  
• Foreground and background  
• Regions of interest (ROI)  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-3  
Uses: Industrial Inspection  
R. Nagarajan et al. “A real time marking inspection scheme  
for semiconductor industries“, 2006  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-4  
Uses: Document Analysis, Text Recognition  
Handwritten digits  
Natural text (after detection)  
Scanned documents  
4
Department of Mechatronics  
Source: Till Quack, Martin Renold  
Chapter 5 – Binary Image  
5-5  
Uses: Medical/Bio Data  
Source: D. Kim et al., Cytometry 35(1), 1999  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-6  
Uses: Blob Tracking & Motion Analysis  
Frame Differencing  
-
=
Source: Kristen Grauman  
Background Subtraction  
=
-
Source: Tobias Jäggli  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-7  
Uses: Shape Analysis, Free-Viewpoint  
Video  
h  
Medial axis  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-8  
Uses: Intensity Based Detection  
• Looking for dark pixels…  
fg_pix = find(im < 65);  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-9  
Uses: Color Based Detection  
• Looking for pixels within a certain color range…  
fg_pix = find(hue > t1 & hue < t2);  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-10  
Issues  
• How to demarcate multiple  
regions of interest?  
Count objects  
Compute further features per object  
• What to do with “noisy”  
binary outputs?  
Holes  
Extra small fragments  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-11  
Outline of Today’s Lecture  
• Convert the image into binary form  
Thresholding  
• Clean up the thresholded image  
Morphological operators  
• Extract individual objects  
Connected Components Labeling  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-12  
Thresholding  
• Grayscale image ⇒ Binary mask  
• Different variants  
One-sided  
ꢄ, ꢂꢅ ꢀ ꢂ, ꢃ ≥ ꢁ  
ꢆ, ꢇꢈꢉꢊꢋꢌꢂꢍꢊ  
ꢀꢁ ꢂ, ꢃ = �  
Two-sided  
ꢄ, ꢂꢅ ꢁꢄ ≤ ꢀ ꢂ, ꢃ ≤ ꢁꢎ  
ꢆ, ꢇꢈꢉꢊꢋꢌꢂꢍꢊ  
ꢀꢁ ꢂ, ꢃ = �  
Set membership  
ꢄ, ꢂꢅ ꢀ ꢂ, ꢃ ∈ ꢏ  
ꢆ, ꢇꢈꢉꢊꢋꢌꢂꢍꢊ  
ꢀꢁ ꢂ, ꢃ = �  
Image Source: http://homepages.inf.ed.ac.uk/rbf/HIPR2/  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-13  
Thresholding  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-14  
Thresholding  
• Load trees  
• Binary_Image = Gray_Image.ThresholdBinary(new Gray(120), new  
Gray(255));  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-15  
Selecting Thresholds  
• Typical scenario  
Separate an object from a distinct background  
• Try to separate the different grayvalue distributions  
Partition a bimodal histogram  
Fit a parametric distribution (e.g. Mixture of Gaussians)  
Dynamic or local thresholds  
• In the following, I will present some simple methods.  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-16  
A Nice Case: Bimodal Intensity Histograms  
Ideal histogram, light object  
on dark background  
Actual observed  
histogram with noise  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-17  
Not so Nice Cases…  
• How to separate those?  
Two distinct modes Overlapping modes  
Multiple modes  
• Threshold selection is difficult in the general case  
Domain knowledge often helps  
E.g. Fraction of text on a document page (histogram quantile)  
E.g. Size of objects/structure elements  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-18  
Global Binarization [Otsu’79]  
• Search for the threshold T that minimizes the within- class  
variance σwithin of the two classes separated by T  
• This is the same as maximizing the between-class variance  
σbetween  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-19  
Algorithm  
• Precompute a cumulative grayvalue histogram h.  
For each potential threshold T  
1.) Separate the pixels into two clusters according to T  
2.) Look up n1, n2 in h and compute both cluster means  
2
3.) Compute σbetween  
Source code  
Department of Mechatronics  
Chapter 5 – Binary Image  
5-20  
Effects  
graythresh - Global image threshold using Otsu's method  
Department of Mechatronics  
Tải về để xem bản đầy đủ
pdf 79 trang myanh 21300 Free
Download
Bạn đang xem 20 trang mẫu của tài liệu "Bài giảng Nhập môn thị giác máy tính - Chapter 5: Binary image analysis - Lê Thanh Hải", để tải tài liệu gốc về máy hãy click vào nút Download ở trên

File đính kèm:

  • pdfbai_giang_nhap_mon_thi_giac_may_tinh_chapter_5_binary_image.pdf