21-07-2011, 11:20 AM
[attachment=14677]
Image Enhancement
Overview
Are employed to emphasize, sharpen & smooth image features for display and analysis
Image enhancement is the process of applying these techniques to facilitate the development of a solution to a computer imaging problem
Operate in the spatial domain, manipulating the pixel data, or in frequency domain, by modifying the spectral components (figure 8.1.2)
Some used both
Type of techniques :-
Point operations – where each pixel is modified according to a particular equation that is not dependent on other pixel values
Mask operations - where each pixel is modified according to the values in a small neighborhood
Global operations – where all the pixel values in the image are taken into consideration
Spatial domain processing include all three but freq domain use global operations
Gray Scale Modification
Also called gray level scaling or gray level transformation, is the process of taking the original gray level values and changing them to improve image
Relates to improving image contrast and brightness
Image contrast is a measure of the distribution and range of gray levels – the difference between the brightest & darkest pixel values and how the intermediate values are arranged
Image brightness refers to the overall average, or mean, pixel value in the image
Mapping Equations
One method to modify gray levels is a mapping equation
mapping equation changes the pixel’s (gray level) values based on a mathematical function that uses brightness values as input
The outputs of the equation are the enhanced pixel values
Mapping is in the category of point operations
Primary operations applied to gray scale image are to compress or stretch it
Compress that are of little interest to us, and stretch where desire more information
When the slope of the line is 0 – 1, gray level compression
If the slope is greater than 1, gray level stretching
See example 8.2.1
Figure 8.2-2 Gray-level Stretching with Clipping at Both Ends. a) The mapping equation, b) the original image, c) the modified image with the stretch gray levels
Histogram Modification
Use similar function which referred to as histogram modification
Focus on histogram shape and range
Histogram with a small spread has low contrast,
histogram with a wide spread has high contrast
Histogram clustered at the high end corresponds to a bright image
Examination of histogram is useful tools, as it contains information of gray level distribution that easy to see the modifications that may improve
Figure 8.2-10 Histogram Stretching with Clipping. a) Original image, b) histogram of original image, c) image after histogram stretching with out clipping, d) histogram of image ©, e) image after histogram stretching with clipping 1% of the values at the high and low ends
Figure 8.2-11 Histogram Shrinking a) Original image, b) histogram of image (a), c) image after shrinking the histogram to the range [75,175], d) histogram of image ©
Figure 8.2-12 Histogram Slide. The original image for these operations is the image from 8.2-11c that had undergone a histogram shrink process. a) the resultant image from sliding the histogram down up 50, b) the histogram of image (a) , c) the resultant image from sliding the histogram down by 50, b) the histogram of image ©.
Histogram equalization is an effective techn for improving the appearance of a poor image
The function is the same as histogram stretch but often provides more visually pleasing results across a wider range of images
Involves probability theory which treat as the probability distribution of gray levels
Histogram equalization process consists 4 steps:
Find the running sum of the histogram values
Normalize the values from step (1) by dividing by the total number of pixels
Multiply the values from step (2) by the maximum gray level value and round
Map the gray level values to the results
See example 8.2.6
IMAGE SHARPENING
Image sharpening deals with enhancing detail information in an image, typically edges and textures
Detail information is typically in the high spatial frequency information, so these methods include some form of highpass filtering
Many image sharpening algorithms consist 3:-
Extract high frequency information
Combine the high frequency image with original image to emphasize image detail
maximizing image contrast via histogram manipulation
High Frequency Emphasis
Using a high boost spatial filter
This mask is convolved with the image & value x determines the amount of low frequency information retained in the resulting image
Value 8 – highpass filter (output image will contain only the edges)
Larger values will retain more of ori image
Less than 8 – negative of ori
High Boost Spatial Filtering a) Original image
b) results of performing a highboost spatial filter with a 3x3 mask and x = 6
c) histogram stretched version of (b) , note the image is a negative of the original,
d) results of performing a highboost spatial filter with a 3x3 mask and x = 8
e) histogram stretched version of (d), note the image contains edge information only
, f) results of performing a highboost spatial filter with a 3x3 mask and x = 12
g) histogram stretched version of (f)
high boost mask can be extended with -1’s and a corresponding increase in the value x
Larger masks will emphasize the edges more (make them wider), and help to mitigate the effects of any noise in original image
If we create NxN mask, value x is NxN-1,
5x5-1=24
Directional Difference Filters
Similar to high boost filter but emphasize the edges in a specific direction
This filters also called emboss filters, due to the effect they create on the output image
Directional Difference Filters. a) Original image, b) image sharpened by adding the difference filter result to the original image, followed by a histogram stretch,
c) 3x3 filter result with the +1 and -1 in the horizontal direction which emphasizes vertical lines, d) 3x3 filter result with the +1 and -1 in the vertical direction which emphasizes horizontal lines,
e) 7x7 filter result with the +1 and -1 in the horizontal direction which emphasizes vertical lines, d) 7x7 filter result with the +1 and -1 in the vertical direction which emphasizes horizontal lines
Homomorphic Filtering
Digital images are created from optical images
Optical images consist of 3 primary components, lighting & reflectance component
Lighting component results from lighting conditions present when image is captured, & can change as the lighting conditions change
reflectance component results from the way objects in the image reflect light & are determined by properties of object
Many applications it is useful to enhance reflectance component, while reducing the contribution from the lighting component
Homomorphic filtering is a freq domain filtering process that compresses the brightness (from the lighting conditions), while enhancing the contrast (from the reflectance)
Image model is as follows:
I(r,c) = L(r,c) R(r,c)
where L(r,c) represents the contribution of lighting conditions, & R(r,c) represents the contribution of reflectance properties of objects
Assumes that L(r,c) consists of primarily slow spatial changes (low spatial frequencies), & is responsible for overall range of brightness
Assumptions for R(r,c), consists primarily of high spatial frequency information
Consists of 5 steps:
A natural log transform (base e)
the Fourier transform
Filtering
the inverse Fourier transform, and
the inverse log function – the exponential
Decouple the L(r,c) & R(r,c) components
Puts the image into freq domain
Perform filtering
Inverse transforms (step 2)
Inverse step 1
Unsharp Masking
Used by photographers to enhance image
It sharpens image by subtracting a blurred (lowpass) version of original image
This was accomplished during film development by superimposing a blurred negative onto corresponding film to produce a sharper result
The process is similar to adding a detail enhanced (highpass) version to original
To improve image contrast, include histogram modification as part of unsharp masking enhancement algorithm
Original image is lowpass filtered, followed by histogram shrink
Resultant image is subtracted from original image
Histogram stretch to restore image contrast
Different ranges of histogram shrinking
Image Smoothing
Used to give image softer or special effect, or to mitigate noise effects
For spatial domain is by considering a pixel and its neighbors and eliminating any extreme values with median or mean filters
In freq domain, is accomplished by some form of lowpass filtering
Equivalent convolution mask can be approximated with Moore-Penrose
Some form of average (mean) filters
The coefficients are all positive, unlike highpass filters
Some common spatial convolution masks, where first 2 are standard arithmetic mean filters & last 2 are approximations to Gaussian filters
Image Enhancement
Overview
Are employed to emphasize, sharpen & smooth image features for display and analysis
Image enhancement is the process of applying these techniques to facilitate the development of a solution to a computer imaging problem
Operate in the spatial domain, manipulating the pixel data, or in frequency domain, by modifying the spectral components (figure 8.1.2)
Some used both
Type of techniques :-
Point operations – where each pixel is modified according to a particular equation that is not dependent on other pixel values
Mask operations - where each pixel is modified according to the values in a small neighborhood
Global operations – where all the pixel values in the image are taken into consideration
Spatial domain processing include all three but freq domain use global operations
Gray Scale Modification
Also called gray level scaling or gray level transformation, is the process of taking the original gray level values and changing them to improve image
Relates to improving image contrast and brightness
Image contrast is a measure of the distribution and range of gray levels – the difference between the brightest & darkest pixel values and how the intermediate values are arranged
Image brightness refers to the overall average, or mean, pixel value in the image
Mapping Equations
One method to modify gray levels is a mapping equation
mapping equation changes the pixel’s (gray level) values based on a mathematical function that uses brightness values as input
The outputs of the equation are the enhanced pixel values
Mapping is in the category of point operations
Primary operations applied to gray scale image are to compress or stretch it
Compress that are of little interest to us, and stretch where desire more information
When the slope of the line is 0 – 1, gray level compression
If the slope is greater than 1, gray level stretching
See example 8.2.1
Figure 8.2-2 Gray-level Stretching with Clipping at Both Ends. a) The mapping equation, b) the original image, c) the modified image with the stretch gray levels
Histogram Modification
Use similar function which referred to as histogram modification
Focus on histogram shape and range
Histogram with a small spread has low contrast,
histogram with a wide spread has high contrast
Histogram clustered at the high end corresponds to a bright image
Examination of histogram is useful tools, as it contains information of gray level distribution that easy to see the modifications that may improve
Figure 8.2-10 Histogram Stretching with Clipping. a) Original image, b) histogram of original image, c) image after histogram stretching with out clipping, d) histogram of image ©, e) image after histogram stretching with clipping 1% of the values at the high and low ends
Figure 8.2-11 Histogram Shrinking a) Original image, b) histogram of image (a), c) image after shrinking the histogram to the range [75,175], d) histogram of image ©
Figure 8.2-12 Histogram Slide. The original image for these operations is the image from 8.2-11c that had undergone a histogram shrink process. a) the resultant image from sliding the histogram down up 50, b) the histogram of image (a) , c) the resultant image from sliding the histogram down by 50, b) the histogram of image ©.
Histogram equalization is an effective techn for improving the appearance of a poor image
The function is the same as histogram stretch but often provides more visually pleasing results across a wider range of images
Involves probability theory which treat as the probability distribution of gray levels
Histogram equalization process consists 4 steps:
Find the running sum of the histogram values
Normalize the values from step (1) by dividing by the total number of pixels
Multiply the values from step (2) by the maximum gray level value and round
Map the gray level values to the results
See example 8.2.6
IMAGE SHARPENING
Image sharpening deals with enhancing detail information in an image, typically edges and textures
Detail information is typically in the high spatial frequency information, so these methods include some form of highpass filtering
Many image sharpening algorithms consist 3:-
Extract high frequency information
Combine the high frequency image with original image to emphasize image detail
maximizing image contrast via histogram manipulation
High Frequency Emphasis
Using a high boost spatial filter
This mask is convolved with the image & value x determines the amount of low frequency information retained in the resulting image
Value 8 – highpass filter (output image will contain only the edges)
Larger values will retain more of ori image
Less than 8 – negative of ori
High Boost Spatial Filtering a) Original image
b) results of performing a highboost spatial filter with a 3x3 mask and x = 6
c) histogram stretched version of (b) , note the image is a negative of the original,
d) results of performing a highboost spatial filter with a 3x3 mask and x = 8
e) histogram stretched version of (d), note the image contains edge information only
, f) results of performing a highboost spatial filter with a 3x3 mask and x = 12
g) histogram stretched version of (f)
high boost mask can be extended with -1’s and a corresponding increase in the value x
Larger masks will emphasize the edges more (make them wider), and help to mitigate the effects of any noise in original image
If we create NxN mask, value x is NxN-1,
5x5-1=24
Directional Difference Filters
Similar to high boost filter but emphasize the edges in a specific direction
This filters also called emboss filters, due to the effect they create on the output image
Directional Difference Filters. a) Original image, b) image sharpened by adding the difference filter result to the original image, followed by a histogram stretch,
c) 3x3 filter result with the +1 and -1 in the horizontal direction which emphasizes vertical lines, d) 3x3 filter result with the +1 and -1 in the vertical direction which emphasizes horizontal lines,
e) 7x7 filter result with the +1 and -1 in the horizontal direction which emphasizes vertical lines, d) 7x7 filter result with the +1 and -1 in the vertical direction which emphasizes horizontal lines
Homomorphic Filtering
Digital images are created from optical images
Optical images consist of 3 primary components, lighting & reflectance component
Lighting component results from lighting conditions present when image is captured, & can change as the lighting conditions change
reflectance component results from the way objects in the image reflect light & are determined by properties of object
Many applications it is useful to enhance reflectance component, while reducing the contribution from the lighting component
Homomorphic filtering is a freq domain filtering process that compresses the brightness (from the lighting conditions), while enhancing the contrast (from the reflectance)
Image model is as follows:
I(r,c) = L(r,c) R(r,c)
where L(r,c) represents the contribution of lighting conditions, & R(r,c) represents the contribution of reflectance properties of objects
Assumes that L(r,c) consists of primarily slow spatial changes (low spatial frequencies), & is responsible for overall range of brightness
Assumptions for R(r,c), consists primarily of high spatial frequency information
Consists of 5 steps:
A natural log transform (base e)
the Fourier transform
Filtering
the inverse Fourier transform, and
the inverse log function – the exponential
Decouple the L(r,c) & R(r,c) components
Puts the image into freq domain
Perform filtering
Inverse transforms (step 2)
Inverse step 1
Unsharp Masking
Used by photographers to enhance image
It sharpens image by subtracting a blurred (lowpass) version of original image
This was accomplished during film development by superimposing a blurred negative onto corresponding film to produce a sharper result
The process is similar to adding a detail enhanced (highpass) version to original
To improve image contrast, include histogram modification as part of unsharp masking enhancement algorithm
Original image is lowpass filtered, followed by histogram shrink
Resultant image is subtracted from original image
Histogram stretch to restore image contrast
Different ranges of histogram shrinking
Image Smoothing
Used to give image softer or special effect, or to mitigate noise effects
For spatial domain is by considering a pixel and its neighbors and eliminating any extreme values with median or mean filters
In freq domain, is accomplished by some form of lowpass filtering
Equivalent convolution mask can be approximated with Moore-Penrose
Some form of average (mean) filters
The coefficients are all positive, unlike highpass filters
Some common spatial convolution masks, where first 2 are standard arithmetic mean filters & last 2 are approximations to Gaussian filters
