Image Enhancement Introduction
application:
The first is to improve the visual effect of the image
features of the second type of image projection, to facilitate computer processing
The reason the image quality degradation :
1. Contrast Problem: partial or full low contrast, affect the visual image
2. The noise problem: interference and damage suffered image ( additive noise and multiplicative noise, periodic noise, quantization noise, salt particles, black pepper noise, background interference )
3. decreased sharpness problems, make the image blurred, or even seriously distorted
The main objective of image enhancement techniques are:
By processing of the image, the image is more suitable than a pre-processing application specific
Highlight the image of "useful" information, expand the image characteristic differences between different objects, to lay the foundation for the identification and extraction of image information
Airspace processing
1. Enhanced point operations - tone converter
What is gray-scale transformation
The interval is mapped to a gray gradation conversion section referred to another gradation transformation
Gradation conversion action
Gradation transformation can increase the dynamic range of the image, the image contrast stretching, the image is clear, obvious characteristics, image enhancement is an important means
Application of gray level transformation
Brightness adjustment, contrast stretching, gray level slice
Classification gradation conversion
1. The linear transformation: linear transformation, piecewise linear transformation (gradation number of the image drawing details are not interested part relative inhibition.)
2. The non-linear transformation: nonlinear extension than the entire grayscale range image is expanded, but only selectively on certain gray value range expansion, another gray value range is likely to be compressed
And piecewise linear stretching differences: instead of non-linear stretching in different gray value interval by selecting different linear equations to achieve different gray value interval expansion and compression, but uniform over the entire range of the gradation value nonlinear transform function, a function using mathematical properties different gray values to achieve the expansion of the compression section
Gray level transformation is applied:
A brightness adjustment - is highlighted, the image dimming
2. contrast stretching - increase, decrease the contrast
To improve the contrast: the position is usually obtained by two inflection points histograms
Reduction in contrast: decrease the gradation of the input image is smaller than the gray level contrast generally used for the output device, such as a Fourier spectrum when displaying
3. Local increase locally decrease the contrast
4. gray level slice: mainly used for projecting a specific gradation range, thereby enhancing a special feature
Two kinds of nonlinear extension method: logarithmic expansion, exponential expansion
The method of obtaining the transfer function
Fixed functions: a sine function, piecewise linear, exponential function, logarithmic function
Interpolation sample interaction: used cubic spline interpolation curve points obtained transform function
Histogram
Analysis grayscale conversion
Grayscale image conversion have a negative impact on the level
The reason: As the conversion is carried out on a limited number of gray levels, and therefore result in reduced levels of
Improvement: higher levels through the input (e.g.,> ), to ensure the grayscale conversion on the image, which retain sufficient levels of output
2. Histogram Enhancement - Airspace filter
Histogram equalization
An automatic image contrast quality adjustment algorithm
The method used is gradation transformation: s = T (r)
The basic idea is the probability density function p r gradation level (the ) to determine the gradation transformation function T (r)]
From a mathematical point of view, histogram equalization is the cumulative distribution function transformation method based on histogram modification method
Histogram equalization - gradation conversion function of
gradation 1) obtaining a histogram of the original image f, is set to h. h is a 256-dimensional vector.
2) determine the overall number of pixels in the image f, a, = m × n-Nf of
. 3) calculates the percentage of the number of pixels for each gradation based on the whole image:
HS (I) = H (I) / N F (i = 0,1, ..., 255 )
cumulative distribution 4) is calculated for each image gradation hp:
5) Theoretical gradation transformation function: T (r) = hp (i) * 255
Without changing the number of gradation appears, it is changed corresponding to the number of occurrences of the gray level. Whereby an image without changing the information structure
The number of pixels occurring within the interval to try to make nearly equal as long as
Histogram equalization is substantially reduced in exchange for grayscale image contrast increase. During the balancing process, the histogram of the original gray scale number is smaller as merged or a very few gray scale level
Histogram matching
Algorithm summarizes the process:
1) obtaining a gray level transformation T
2) is determined gray level transformation G
. 3) is obtained inverse transformation
4) T and by obtaining the composite conversion H
. 5) H make the image with gray level transformation
3. The color image enhancement
Enhanced on the RGB model - color balance
Cast: sampling process, due to the equipment, the environment will result in three different color components of the image transformation relationship, all the colors in an image of the object deviated from its original true color, a phenomenon known as color cast . As part of the image with a gray color
Gray balance: the color components of RGB mixed color device, color hue and saturation to generate the loss of gray, the color mixing is referred to as gray balance, under normal circumstances, an equal amount of generated RGB gray.
Color balance: color cast correction process is called color balance
To achieve color balance, gray balance by adjusting the area of color cast, to achieve recovery of gray.
When the brightness reaches a certain level of gray, it appears white, it is sometimes also referred to as white balance adjustment
Enhanced on the HSI model
Processed by tone
The basic idea: 1 converts the image into HSI color space 2. adjusting the specified hue value H, H '= H +/- Δh
The main application
changes the image atmosphere (e.g., change the atmosphere of warm and cool colors, sooner or later change the atmosphere)Color change (tone color designated to be replaced), desaturate
Processed luminance
The basic idea:
Converts the image into HSI color space
Designated luminance value I, a multiplied by the adjustment amount [Delta] I
the I '= the I [Delta] I *Main applications:
(1) multiplying each pixel of a luminance component of a constant (e.g., 1.3) is greater than, so that the image becomes brighter, improving brightness of the image
(2) for multiplying each pixel of a luminance component is less than constant (e.g. 0.8) 1, so that the brightness of the image decreases.
(3) adjust the brightness of the image are selectively and to be in a color tone, depending on whether the brightness as the selection process. For example, only red tones to improve the brightness.
(4) luminance component histogram equalizationProcessed by the color saturation
The basic idea:
Converts the image into HSI color space
Designated luminance value S, multiplied by a quantity [Delta] S
S '= S [Delta] S *Main applications:
(1) by a constant greater than 1 (e.g. 1.3) of the saturation component of each pixel, so that the color of the image more distinctive
(2) of the saturation component of each pixel by less than a 1 constant (e.g., 0.8), so that the sharpness of the color image is reduced.
(3) adjust the color saturation of the image are selectively and to be in a color tone, depending on whether the selection as a saturation process. For example, only increase the saturation of red-hued
Pseudo-color image processing
Method: pseudo-color conversion, density slice
4. The spatial filter template
The basic concept of the spatial filtering: spatial processing using an image template, known as spatial filtering.
Linear filter classification
A low pass filter - main purposes: a smoothed image, remove noise
High pass filter - main purposes: edge enhancement, edge extraction
Band-pass filter - main purposes: to delete a specific frequency, with little enhancement
Nonlinear filter classification
Median filter - main purposes: a smoothed image, remove noise
Maximum value filter - the main purposes: to find the most highlights
Minimum value filtering - main purposes: to find the most scotoma
Smoothing (low pass filtering substantially median filtering)
The main purpose of the smoothing filter
Former large image processing, deleting useless tiny details
Interruption of the connection line and curve
Reduce noise
Smoothing processing, image restoration excessive sharpening
Creative image (shadows, soft edges, hazy effect)
Several simple low-pass filter
Mean filter - local average method (to be processed pixel value, equal to the average of all pixels of adjacent pixels therearound)
A weighted average value of the pixel to be processed is equal to all neighboring pixels surrounding a pixel - weighted average filter
The larger the size of the template, the more blurred image, the more image detail is lost
Disadvantages and problems: If the object of image processing is to remove the noise, then the low-pass filter removing noise while also smoothing the sharp edges and details
Median filter
principle:
With pixel values within the template region, as the result value MID = {R & lt | K = 1,2, ...,}. 9
Forced prominent spot (dark spot) is more like a value around it, to remove isolated bright spots (scotoma)
Median filtering advantages:
Noise suppression
While removing noise can better retain details of the edge contour information and images
Sharpening filter (basic high-pass filtering, high gain filtering, differentiation filter)
The main purpose of the sharpening filter
Strengthening the scene image edges and contours
Fine detail printing emphasis. Make up the scanning, the image smoothing hanging
An ultrasound imaging probe, low resolution, edge blur, sharpen be improved by
Image recognition, the segmentation of the front edge
Sharpening excessive smooth recovery, lack of exposure of the image
Creative image (only the special image boundary)
Target Recognition sophisticated weapons, positioning
1. The basic high-pass filtering
High pass filtering the same time reinforcing the edges, the loss of the level and brightness of the image
In some cases, high-pass filtering to enhance the small scale features
2. High gain filter
High pass filtering to compensate for the defects, while enhancing the edges and details, without losing the low frequency component of the original image
High-pass filter can be seen as: a high-pass = picture - a low-pass
In the above formula picture multiplying a scale factor A, has a high filter gain: Gain = A picture of high - low-pass
Effectiveness analysis:
High-gain advantage than the high-pass: only enhanced side, retaining the level.
Noise has an important influence on the visual effect of the resulting image, high gain at the same time enhancing the edges also enhance the noise.
3. The differential filter
Frequency domain processing
1. Frequency domain filtering - low-pass filtering, high pass filtering, homomorphic filtering
Low-pass filtering
The basic idea of the frequency domain low pass filtering
Over the low-pass filter, Butterworth low pass filter, exponential filter, a Gaussian low pass filter (glpF), a ladder filter
High-pass filtering
The basic idea of high-pass filtering in the frequency domain
Over the high-pass filter (ILPF), Butterworth high-pass filter, Gaussian high pass filter
Homomorphic filter
The basic idea of homomorphic filter
Defined homomorphic filter
Analysis of the effect of homomorphic filter
Wavelet transform filter - different from other frequency domain filtering is assumed that the wavelet transform filtering the difference signal and noise in the transform domain is not in position, but at different amplitude coefficient, whereby it is possible to filter out noise and simultaneously reducing loss of detail