face recognition and biometric systems 2005/2006 filters
TRANSCRIPT
Face Recognition and Biometric Systems 2005/2006
Plan of presentation
Review of available filters Filter application in various parts of automatic face recognition system Further research
Face Recognition and Biometric Systems 2005/2006
Filter grouping
One pixel operations Pixel area operations
Image histogram operations Image rotation & scaling Complex techniques
Face Recognition and Biometric Systems 2005/2006
One pixel operations
Linear function Power function Logarithmic function
Application Contrast improvement Image sharpness enhancement
)],([),( yxIfyxI inout
Face Recognition and Biometric Systems 2005/2006
Linear function
Scaling Dynamic range scaling in a chosen
sections
222255
2255)2(
)2,1(112
12)1(
11
1
),(
rIforsr
srI
rrIforsrr
ssrI
rIforr
sI
yxI
inin
inin
inin
out
Face Recognition and Biometric Systems 2005/2006
Power function
Gamma correction Image after translation still looks
naturally
45,0),(),( whereyxIyxI inout
0
50
100
150
200
250
1 51 101 151 201 251
Face Recognition and Biometric Systems 2005/2006
Logarithmic function
Gray level compression Natural image look Partial lost of image information
1)(max255
1
)ln(*
)1),(ln(),(
,
Ic
b
cb
yxIayxI
yx
inout
0
50
100
150
200
250
1 51 101 151 201 251
Face Recognition and Biometric Systems 2005/2006
One pixel filters - example
Input image
Logarithm
Scaling
Gamma
Face Recognition and Biometric Systems 2005/2006
One pixel filters - example
Input image
Logarithm
Scaling
Gamma
Face Recognition and Biometric Systems 2005/2006
One pixel filters - example
Input image
Logarithm
Scaling
Gamma
Face Recognition and Biometric Systems 2005/2006
One pixel filters
Advantages:Improvement of image contrastBetter sharpness
Disadvantages:Too bright pixels Difficulties with optimal parameters selection
Face Recognition and Biometric Systems 2005/2006
Area filters
Lowpass filters Mean filter Gauss Median
Highpass filters Roberts Prewitt Sobel
Laplacian
Face Recognition and Biometric Systems 2005/2006
Lowpass filters
Noise reduction Image smoothing Contour blurring
Face Recognition and Biometric Systems 2005/2006
Mean filter
Linear filter Light image smoothing
111
111
111
9
1
group
inout II9
1
Face Recognition and Biometric Systems 2005/2006
Gauss filter
Filter uses power function Stronger image smoothing in a shorter time
2
22
222
1),(
yx
eyxG
121
242
121
16
1
Face Recognition and Biometric Systems 2005/2006
Median filter
Nonlinear filter Good for noise removal from image without important information elimination
Face Recognition and Biometric Systems 2005/2006
Lowpass filters - example
Input image
Gauss
Mean
Median
Face Recognition and Biometric Systems 2005/2006
Highpass filters
Image sharpness enhancement Contour detection In case of noisy images the errors will multiply
Face Recognition and Biometric Systems 2005/2006
Roberts filter
Gradient method
000
010
100
xR
000
010
001
yR
||||||
|| 22
yx
yx
RRR
RRR
y
I
x
II
,
Face Recognition and Biometric Systems 2005/2006
Prewitt filter
Gradient method
111
000
111
xP
101
101
101
yP
||||||
|| 22
yx
yx
PPP
PPP
y
I
x
II
,
Face Recognition and Biometric Systems 2005/2006
Sobel filter
Gradient method
121
000
121
xS
101
202
101
yS
22||
||||||
yx
yx
ssS
SSS
y
I
x
II
,
Face Recognition and Biometric Systems 2005/2006
Laplacian filter
Method uses second derivative properties
111
181
111
2
2
2
2
,),(y
I
x
IyxL
Face Recognition and Biometric Systems 2005/2006
Highpass filters - example
Input image
Prewitt
Roberts
Sobel
Face Recognition and Biometric Systems 2005/2006
Histogram operations
Stretching Fitting Equalization
Face Recognition and Biometric Systems 2005/2006
Histogram stretching
Image dynamic range enlargement for image contrast & sharpness enhancement
Does not work on images with characteristic histogram
minmax
min),(*)12(),(
yxI
yxI inBout
Face Recognition and Biometric Systems 2005/2006
Histogram equalization
Equal distribution of gray scale levels in input image Contrast enhancement
Face Recognition and Biometric Systems 2005/2006
Histogram equalization
countpixelKwhereKIhIp /)()(
levelsgreynwhereipiDn
i
0
)()(
valueimageorginalzerononfirstD
D
DIDI
in
B
in
ininout
0
0
0 )12(1
)(
Algorithm:
Face Recognition and Biometric Systems 2005/2006
Histogram fitting
Its aim is a transformation of an input histogram so it looks like the given one Image lighting unification
Face Recognition and Biometric Systems 2005/2006
Histogram fitting
Algorithm: Input & output image histogram
calculation (hIn ,hOut ) Histogram normalization
Increment function calculation
countpixelKwhereKIhIp /)()(
levelsgreynwhereipiDn
i
0
)()(
Face Recognition and Biometric Systems 2005/2006
Histogram - exampleInput image
Equalization
Stretching
Fitting
Face Recognition and Biometric Systems 2005/2006
Histogram
Minimization of lighting differences in images from different sources Image sharpness and contrast enhancement
Face Recognition and Biometric Systems 2005/2006
Complex filters - techniques
Kuwahara Canny Unsharp Masking LogAbout GammaAbout
Face Recognition and Biometric Systems 2005/2006
Kuwahra filter
Nonlinear filters Good image smoothing Low contours blurring Algorithm: For each region: Result:
region
insr In
I1
region
srin II 2)(
)()min( rIIr sroutregions
Face Recognition and Biometric Systems 2005/2006
Canny filter
Optimal contour detection Algorithm: Gauss filter Sobel filter Borders direction described as Direction definition Pixel tracking in the direction of borders
and removal of unnecessary pixels Thresholding
)/(tan 1xy SS
Face Recognition and Biometric Systems 2005/2006
Unsharp Masking
Image sharpening Minor noise elimination Algorithm: I(x,y) = Gauss(Iin(x,y)) Ihp(x,y) = Iin(x,y) – I(x,y) Ihp(x,y) = 0 dla Ihp(x,y) < threshold Iout(x,y) = Iin(x,y) + a*Ihp(x,y)
Face Recognition and Biometric Systems 2005/2006
LogAbout method
Contour detection improvement
Highpass filter
Logarithmicfilter
Face Recognition and Biometric Systems 2005/2006
HistAbout method
Contour detection enhancement
Histogram stretching
Gauss
LogAbout
Face Recognition and Biometric Systems 2005/2006
GammaAbout method
Contour detection improvement
Gamma
Gauss
LogAbout
Face Recognition and Biometric Systems 2005/2006
Where use filers?
Input image Detection Normalization
Face Recognition and Biometric Systems 2005/2006
Input image
Problems: Noises
Solution: Gauss filter Median filter
Face Recognition and Biometric Systems 2005/2006
Input image/Detection
Problem: Dark image
Solution: Histogram stretching Gamma correction GammaAbout
Face Recognition and Biometric Systems 2005/2006
Detection
Problem: Contour detection
Solution: Roberts filter Prewitt filter Sobel filter Canny’s method
Face Recognition and Biometric Systems 2005/2006
Shape normalization
Problem: Lack of size unification Solution: Scaling
Problem: Non frontal face Solution: Rotation
Face Recognition and Biometric Systems 2005/2006
Lighting normalization
Problem: Irregular face lightning
Solution: Histogram operations
Face Recognition and Biometric Systems 2005/2006
Filter usage
Image quality enhancement Object detection method efficiency improvement Image normalization Lighting normalization
Face Recognition and Biometric Systems 2005/2006
What further??
Lighting normalization is still an area for research Dark image brightening