Performance Analysis of Face Detection Systems
2 likes774 views
The document analyzes the performance of three face detection algorithms: one using local SMQT features and a split up SNoW classifier, one using an efficient and rank deficient approach, and a simple and accurate color face detection algorithm for complex backgrounds. It tests the algorithms on several face databases, recording the success rate and processing time. It finds the local SMQT approach has the highest accuracy at 98% but takes the longest time, while the efficient rank deficient approach has a lower accuracy of 87.5% but processes images much faster at around 26 seconds.
![ex01.m
Facefind.m
Convertion to x,y,w,h
Facefind.dll
Image as double
x1,y1,x2,y2 X1,y1,x2,y2
rect=[x,y,w,h]
imcrop(image,rect)
image
Cropped image](https://image.slidesharecdn.com/pp-140421160325-phpapp02/85/Performance-Analysis-of-Face-Detection-Systems-32-320.jpg)
Performance Analysis of Face Detection Systems
- 1. Performance Analysis of Face Detection Systems Melike Sennur Kocal (028354) Onur enol (990406)
- 2. Face Detection Biometric id,Surveillance,Security. Too many variations of human face. Extracting the face.
- 3. Project Aim Analyse Performances of 3 Algorithms
- 4. Project Aim Analyse Performances of 3 Algorithms Face Detection Using Local SMQT Features and Split Up SNoW Classifier Face Detection Using Efficient & Rank Deficient Simple & Accurate Color Face Detection Algorithm in Complex Background
- 5. Project Aim For-Loop for many images Cropping the detected faces Place the cropped faces into an output folder Timer Function for total elapsed time.
- 6. Project Aim For-Loop for many images Cropping the detected faces Place the cropped faces into an output folder Timer Function for total elapsed time. COMPARISON
- 7. Face Databases HumanScan JAFFE ORL YALE Cohn Kanade FG-NET CVL
- 8. Face Databases HumanScan 1521 Grayscale JAFFE 213 Grayscale ORL 400 Grayscale YALE 165 Grayscale Cohn Kanade 180 Grayscale FG-NET 126 Color CVL 797 Color
- 9. Face Databases HumanScan 1521 Grayscale JAFFE 213 Grayscale ORL 400 Grayscale YALE 165 Grayscale Cohn Kanade 180 Grayscale FG-NET 126 Color CVL 797 Color Efficient & Rank Deficient SMQT Features & Split up Snow Classifier Simple & Accurate Face Detection Algorithm in Complex Background
- 10. Face Databases HumanScan 1521 Grayscale JAFFE 213 Grayscale ORL 400 Grayscale YALE 165 Grayscale Cohn Kanade 180 Grayscale FG-NET 126 Color CVL 797 Color Efficient & Rank Deficient SMQT Features & Split up Snow Classifier Simple & Accurate Face Detection Algorithm in Complex Background (2x6) + (1x2) = 14 test runs
- 11. Name & Color Conversions Name & Color Conversions Freeware Software IrfanView
- 12. Local SMQT & Split Up Snow Classifier Main.m Facefind.m Facefind.dll Plotbox.m Plotsize.m
- 13. Input Image
- 14. Convert From RGB to GrayScale
- 16. Sending image to facefind.dll
- 17. Draw Frames
- 19. main.m
- 23. main.m Facefind.m Plotbox.m Plotsize.m Facefind.dll Image as double X1,y1,x2,y2 X1,y1,x2,y2 Min&Max face sizes
- 24. main.m Facefind.m Plotbox.m Plotsize.m Facefind.dll Image as double X1,y1,x2,y2 X1,y1,x2,y2 Min&Max face sizes
- 26. Modification Extension Definition for input images Converting int to str Path i.e:150.pgm 150 + pgm
- 27. Modification Same as the original code
- 29. Modification Converted to integers Conversion of coordinates for imcrop() function
- 30. Modification Image input (again) Rect array Getting cropped image into matrix tem
- 31. Modification Concatenation i.e; 150+jpg Cropped image as temp matrix Full output path (Directory) JPEG COMPRESSION
- 32. ex01.m Facefind.m Convertion to x,y,w,h Facefind.dll Image as double x1,y1,x2,y2 X1,y1,x2,y2 rect=[x,y,w,h] imcrop(image,rect) image Cropped image
- 33. Run DEMO Demonstrating Modified Code on MATLAB
- 34. Efficient & Rank Deficient
- 36. Image
- 37. Image Image & Threshold Fdmex()
- 38. Image Image & Threshold Fdmex() Image & Threshold
- 39. Image Image & Threshold Fdmex() x,y,width,height Image & Threshold S-array
- 40. Image Image & Threshold Fdmex() x,y,width,height Image & Threshold S-arrayRectangle()
- 41. Image Image & Threshold Fdmex() x,y,width,height Image & Threshold S-arrayRectangle()
- 42. Efficient & Rank Deficient Modification Behaves like a sort of sensitivity Get the image Extension Definition
- 43. Efficient & Rank Deficient Modification Get coordinates for each face X1,Y1,W,H
- 44. Efficient & Rank Deficient Modification
- 45. Image Image & Threshold Fdmex() x,y,width,height Image & Threshold
- 46. Run DEMO Demonstrating Modified Code on MATLAB
- 47. Simple & Accurate Lighting Compensation Extracting Skin Removing Noise Finding Skin Color Blocks Checking Face Criterions
- 48. Simple & Accurate Modification
- 49. Simple & Accurate Modification
- 50. Run DEMO Demonstrating Modified Code on MATLAB
- 51. Test Results Database # of photos Success (%) Seconds ORL 400 339 84.75 4,83 YALE 165 142 86.06 2,06 JAFFE 213 213 100 8,67 COHN KANADE 180 176 97.7 32,7 HUMAN SCAN 1521 918 60.355 105,05 FG-NET 126 121 96.031 5,92 AVERAGE 87.50 26,54 Efficient & Rank Deficient Approach
- 52. Test Results Database # of photos Success (%) Seconds ORL 400 400 100 135,9 YALE 165 165 100 39,462 JAFFE 213 211 99.06 1095,77 COHN KANADE 180 164 91.11 3095,838 HUMAN SCAN 1521 1490 97.96 11159,83 FG-NET 126 126 100 452,06 AVERAGE 98.02 % 2663,14 Using Local SMQT Features & Split Up SNoW Classifier
- 53. Test Results Database # of photos # of cropped photos # of correct photos (%) Seconds FG-NET 126 119 75 59.5 102.83 CVL 797 752 575 72.1 4283.835 A Simple & Accurate Face Detection Algorithm in Complex Background
- 59. Conclusion Local SMQT & Split Up Snow Classifier Efficient & Rank Deficient % 98 SUCCESS % 87.5 SUCCESS 2663,14 seconds 26,54 seconds
- 60. Conclusion Local SMQT & Split Up Snow Classifier Efficient & Rank Deficient % 98 SUCCESS % 87.5 SUCCESS 2663,14 seconds 26,54 seconds CVL FG-NET % 98 SUCCESS % 87.5 SUCCESS 4238,835 seconds 102,83 seconds 5.3 secondsimage 0.8 secondsimage