[attachment=15124]
A Novel Two-Stage Impulse Noise Removal Technique Based on Neural Networks and Fuzzy Decision
Introduction
Images corrupted by impulse noise.
Objectives of noise removal:-
- Suppress the noise.
- To preserve the sharpness of edge.
Two-Stage impulse noise removal techniques used:-
- Neural Networks
- Fuzzy Decision
Neural Network
Neural network: information processing paradigm inspired by biological nervous systems, such as our brain
Structure: large number of highly interconnected processing elements working together.
Like people, they learn from experience (by example)
Neural network with back-propagation training algorithm is applied to remove the noise cleanly and keep uncorrupted information well.
Has many inputs and one output.
FUZZY DECISION
Decision making system
Zhang proposed the fuzzy techniques to detect the impulse noise and to remove the noise based on long-range correlation within different parts of the image.
Fuzzy decision system inspired by the HVS is proposed to classify the image into human perception sensitive and non-sensitive regions.
System Architecture
Optimal impulse noise removal should delete the visible noise and maintain detail information of the image.
For removal of impulse noise from input images without blurring the edges-2 stages.
Impulse Noise Removal.
Image Quality Enhancement.
Impulse Noise Removal
Neural Network for Noise Detection
Precise Noise Detection
Noise is more annoying in smooth and edge areas.
Algorithms work well on low noise density images but fail in edge region.
3 types of decision based algorithms are :-
Normalized deviation to detect the noise by threshold.
Differences of adjacent pixel values in the rank-ordered median filter sequence.
Switching Schemes
3 layer NN with one hidden layer.
Input layer consists of three nodes:-
Gray-level difference(GD)
Average Background Difference(ABD)
Accumulation Complexity Difference(ACD)
Second layer
6 nodes
Activation function-Bipolar sigmoid function.
Output layer
One node that denotes the attribution of the
pixel: ’noise’ or ‘non-noise’.
Activation function-Bipolar sigmoid function.
Gray-Level Difference(GD):
GD represents the accumulated variations between the central pixel for identification and each surrounding local pixel.
Average Background Difference(ABD):
Averaging the surrounding pixels of the sliding block and comparing it with the central pixel.
Accumulation Complexity Difference(ACD):
Accumulating the difference between each pixel in the 3x3 sliding block and its four neighboring pixels.
Noise Removal Algorithm
First level
Estimation of image density.
Second Level
Noise rate>10
Detect and remove the residual noises.
Noise Rate>30
Highly corrupted region.
5 x 5 median filter applied.
Adaptive two-level noise Removal Algorithm
Very effective to suppress impulse noise.
Preserve the sharpness of edges and details.
HVS-Directed Image Analysis
To classify the image pixels into human
perception sensitive and non-sensitive
regions.
To compensate the blur of the image.
Destruction caused by median filter.
Causes of HVS
Magnitude difference between the object
and background.
Different structures of images.
3 Input variables of HVS are:-
Visibility Degree:-Magnitude difference between the object and its background.
Structural Degree(SD):-High contrast region and pixel block can be separated into clusters.
Complexity Degree(CD):-Summation of gradient value.
Angle Evaluation:-To determine the orientation of sliding block.
NN Based Compensation:-To enhance the sensitive regions to perform better visual quality.
Experimental results
Peak Signal-To-Noise Ratio(PSNR):-
The maximum possible power of a signal and the power of corrupting noise that affects the fidelity of its representation.
Figure Of Merit(FOM):-
-Compare edge preservation.
-Ranges between 0 & 1.
Conclusion
A novel two-stage noise removal algorithm was proposed to deal with impulse noise.
First StageRemoval procedure with NN-based noise detection applied.
Second StageFuzzy decision rules inspired by HVS proposed.
According to experimental results, proposed methods is superior to other conventional methods.
Provide a stable performance.