Abstract: An image is a collection of pixels, which can be acquired from the different types of sources. The heterogeneous image sources have high dense noise, this cause several performance related issues in which the image associated with. So, every application under image processing needs an effective technique to perform noise removal on digital images. Image de-noising is the essential step that should be performed before any image analysis process begins. Image noise reduction involves the manipulation of an image to produce a high quality image. This paper gives the recent techniques of image filters with its merits and demerits. Finally the future objective is identified from the comprehensive study.
INTRODUCTION Image denoising plays an important role in image processing. Selecting appropriate denoising technique is more important one 1. Because, the denoising algorithm selection will be vary for radar images and for biomedical images. The algorithm selection is based on the image and the algorithm.
Reducing noise should have additional concentration, where it can eliminate the features of image such as edges, textures and borders. Before analyzing the techniques, it’s important to know the types of noises.LITERATURE SURVEY 2.1 Noise Types: Digital images are captured from different types of devices, and it creates different types of noises. The noise in an image can hardly affect the interpretation more difficult.
The noise type can be Gaussian noise, impulse noise or sometimes called as salt and pepper noise, speckle noise, Brownian noise, Poisson Noise, shot noise 2, Gaussian Noise: Gaussian noise is a statistical noise, which occurs naturally from probability density function. This is an additive noise. There are several techniques have used different types of filtering to remove Gaussian noise. The most popular filters are bilateral and trilateral filter, which can also help to remove different type of noises in an image.
V.R.Vijaykumar et al 3 proposed a new and fast Gaussian noise removal to restore high Gaussian noise corrupted images.
The algorithm effectively removes noise with edge preserving. Tanzila Rahman et al 4 proposed a modified fuzzy filter for Gaussian noise removal. This is effective on both color and gray scale images. Authors compared the results of the modified filters with different existing algorithms such as Mean Filter, Wiener Filter, Geometric Mean Filter, Harmonic Mean Filter, and Existing Fuzzy Filter. The exisitng fuzzy filter is suitable for impulse noise removal also.
Impulse noise: Impulse noise is also called as salt and pepper noise, which comes under acoustic noise category. This type of noise comprises unwanted, sharp sounds like clicks and pops. Roy, Amarjit, et al 5 proposed support vector machine (SVM) based fuzzy filter for removing impulse noise and restores the corrupted images. This performed channel by channel restoration. This initially classifies the image and performs fuzzy filters on it.Speckle noise: Speckle noise is mostly affected in medical ultrasound images. And this is multiplicative and it is complicated for measuring images with high contrast.
Yang, Jian, Jingfan Fan et al 6 used medical ultrasound images to demonstrate the speckle noise removal. The authors used hybrid noise removal mechanism which combines the local statistics with the non local mean filter.Brownian noise: Brownian noise is also called as brown or red noise and comes under the category of fractal 1/f noises. Brownian noise distribution and Brownian noise corrupted grayscale and color images.Poisson Noise: Poisson noise is signal-dependent and mainly exists in photon images. It occurs when the image is formed by photon particle detection.
The medical images affected by this noise are X-ray imaging, Nuclear Imaging, Positron Emission Tomography (PET), and Single Photon Emission Tomography (SPET). The image is formed when X-ray photons incident on a receptor surface in an exceedingly random pattern. Every individual photon could be a quantum (specific quantity) of energy. It has the quantum structure of the associate X-ray beam that makes the occurrence of quantum noise.
Since the characteristics of this noise are defined by photon counting statistics, it is hard to remove the noise.Shot noise: Shot noise is otherwise called as thermal or Johnson noise. It is a white noise with same power distributed over entire spectrum and is given by Power Spectral Density (PSD) and ?, in which ? value is 0 for shot noise and is proportional to 1 / f? similar to the colors of light differentiated by the frequencies, and the different colors of noise is represented by ?. The beta value zero represents white and the beta value two represents brown.
Table 1.0 shwos the recent approached and techniques used for different filters. Table 1.0 Image Noise and recent filtering techniquesNoise TypeNoisy imageFilters used to removeAdvantagesDisadvantagesGaussianFast Gaussian noise removal filer 3.modified fuzzy filter 4.Preserves edge and details at the time of denoisingHigh computational overhead.Impulse NoiseSVM based fuzzy filter 5.Reduces the removal time and overheadNeed training dataSpeckle NoiseHybrid filter combines local and non-local mean fitlers.Better than the non local mean filter techniqueDifficult for biomedical images.