158 Academic Journal of Nawroz University (AJNU) Digital Image Denoising Techniques in Wavelet Domain with another Filter: A review 1Barwar Mela Ferzo, 2Firas Mahmood Mustafa of IT, Duhok Polytechnic University, Duhok, Kurdistan of Iraq 2Department of Communication, Nawroz University and (Duhok Polytechnic University), Kurdistan Region of Iraq 1Department ABSTRACT Image denoising is a challenging issue found in diverse image processing and computer vision problems. There are various existing methods investigated to denoising image. The essential characteristic of a successful model that denoising image is that it should eliminate noise as far as possible and edges preserving and necessary image information by improving visual quality. This paper presents a review of some significant work in the field of image denoising based on that the denoising methods can be roughly classified as spatial domain methods, transform domain methods, or can mix both to get the advantages of them. This work tried to focus on this mixing between using wavelet transform and the filters in spatial domain to show spatial domain. There have been numerous published algorithms, and each approach has its assumptions, advantages, and limitations depending on the various merits and noise. An analyzing study has been performed comparative in their methods to achieve the denoising algorithms, filtering approach and wavelet-based approach. Standard measurement parameters have been used to compute results in some studies to evaluate techniques while other methods applied new measurement parameters to evaluate the denoising techniques. Keywords: Image Denoising, Discrete Wavelet Transform (DWT), Complex Wavelet Transform, Wiener Filter (WF), Median Filter (MF). 1. Introduction 1 In an increasingly digital world, Digital Images play an affected by different types of noise, such as salt and essential role in the day to day applications such as pepper noise, additive noise, speckle noise, etc. [1]. Digital Cameras, Magnetic Resonance Imaging, and The noisy image leads to confusion of the image and Satellite TV as well as in fields of research and loss of its features so, the noise considered as a technology Information significant factor in degrading the image quality; thus System. Overall, datasets obtained by noise polluted noise reduction is an essential technology in image image sensors. Images often corrupted with noise analysis and the first step to be taken before images are throughout the acquisition, transmission, the retrieval analyzed [2, 3]. Therefore, Image Denoising techniques from storage media, and interfering with natural are required to prevent image contamination of this phenomena, and this noise can damage the data of form of noise [4]. Image denoising re-processing is a interest. The noise is considered to be the most critical significant function before images further processed problem because it corrupts the image due to blurring, like texture analysis, segmentation, feature extraction, movement, camera misfocus, etc. The images are etc. [5]. Over the years, several denoising approaches including Geographical have been proposed, such as wavelet transform, linear Academic Journal of Nawroz University (AJNU) Volume 9, No 1 (2020). Regular research paper : Published 4 March 2020 Corresponding author’s e-mail : barwar.mela@gmail.com Copyright ©2018 1Barwar Mela Ferzo, 2Dr. Firas Mahmood Mustafa. This is an open access article distributed under the Creative Commons Attribution License. doi : 10.25007/ajnu.v9n1a587 filters, non- linear filters (mean, median, bilateral, and wiener filter) Have been used to eliminate noise from images, but these traditionally filters may result in some problems, such as blurring sharp edges, 159 Academic Journal of Nawroz University (AJNU) destroying lines and other fine image details, and fail to information of the image during denoising process [11- effectively remove heavy-tailed noise data [6], while it 14]. is well known that wavelet transform is a signal This review paper provides different methodologies for processing technique which can display the signals on noise reduction depending on wavelet transform and in both time and frequency domain. various filters. It also gives us insights into the Wavelet transform is a superior approach to improve approaches by which the reliable method is to be the quality of images due to its multi-resolution and concluded and estimated an approximation of the subbands property and provides signals to localize in original image from the noisy image. The rest of the frequency and time domain [7]. Donoho's suggested paper organized as follows: Wavelet-based threshold is wavelet thresholding as a signal estimation technique mentioned briefly in section one. A broad study of the that literature review shown in section 2. Conclusion and takes advantage of wavelet transformation capabilities to denoising signal. This technique reduces Discussion presented in section 3. noise by destroying insignificant coefficients relative to 2. Image Denoising Problem Statement a certain threshold and turns out to be efficient and The problem of image denoising can be modeled straightforward. These coefficients subbands processed mathematically as follows: Y=X+z via hard or soft thresholding. The hard thresholding …… (1) eliminates (sets to zero) coefficients that are smaller Where: than a threshold; the soft thresholding shrinks the Y : is the noisy image. coefficients that are larger than the threshold as well. X : is the unknown clean image (denoised image). The effectiveness of thresholding for denoising image z depends on the selection of a suitable threshold such as (AWGN) with standard deviation σn. VisuShrink, BayesShrink, SureShrink, etc. [8]. The noise z can be estimated in practical applications by Several researchers have shown that apply the only various methods, such as robust median estimator and wavelet transform on the image is insufficient to obtain other methods. an integrated noise-free image and image stay blurring. The aim of the denoising process is to rebate the noise Hybrid algorithm of WT with liner filter and non-linear in the images while minimizing the loss of original filter applied on noisy image to filtering image noise features and increasing the signal-to-noise ratio (SNR). out while the edges of the image are well preserved. The significant challenges for image denoising are as WT is integrating with a filter to solve the problem of follows [15]: : is represents additive white Gaussian noise unsharpened edges and poor quality of background • Flat areas should be smooth. during the removal of blurriness of image [9]. Wavelet • Edges should be protected without blurring. denoising technique eliminates image details and • Textures should be preserved. produces smooth image sharpness. So, there is a need • New artifacts should not be generated. of such filter in image denoising to filtering noise Owing to solve the clean image X from the Eq. (1) is an without affecting essential image characteristics [10]. ill-posed problem, we cannot get the unique solution Also to remove mixed noise and produce a right quality from the image model with noise. To obtain a good image with loss of as small as a possible value of estimation image, image denoising has been well- doi : 10.25007/ajnu.v9n1a587 160 Academic Journal of Nawroz University (AJNU) studied in the field of image processing over the past in the signal while preserving the characteristics of the several years. Generally, image denoising methods can signal regardless of its frequency signification. There be roughly classified as spatial domain methods, are three steps involved as a linear forward wavelet transform domain methods, or can mix both to get the transform, a nonlinear thresholding step, and a linear advantages of them. This work tried to focus on this inverse wavelet transform. The process, commonly mixing [15]. called denoising by using 2D-Wavelet Transform, 3. Wavelet Based Thresholding consists of the following main stages, which shown in Wavelet thresholding is a signal estimation strategy figure 1. The first stage is to apply the 2D discrete that takes advantage of Wavelet transform capabilities wavelet transformation to the input image (noisy to denoising signal. This strategy reduces noise by image), while the second stage is to calculate the destroying insignificant coefficients relative to a threshold values and how to apply each value with its certain threshold and turns out to be efficient and related sub band to generate a new and clear sub straightforward, depends heavily determines the value band, whereas the third stage is about to reconstruct of the threshold parameter and the choice of this the denoised image (The output) by using the 2D threshold, to a great extent the efficiency of denoising. inverse wavelet transformation with the new clear sub There are several studies on thresholding the Wavelet bands. coefficients [16]. Wavelet denoising removes the noise Denoised Image Noisy Image (The input) (The output) Threshold Estimation Applying (WT) Wavelet Transformation Applying Threshold Applying (IWT) Inverse Wavelet Fig (1): The Block diagram of Image denoising using 2D-Wavelet Transform. 4. Wavelet Domain and Filtering Denoising Methods implemented in this field by mixing the wavelet In contrast with spatial domain filtering methods, transform domain filtering besides the spatial domain Wavelet transforms domain filtering methods first filtering methods, some of these works will be transform the given noisy image to another domain. illustrated and evaluated as follows: Then they apply a denoising procedure on the [17] proposed an adaptive wavelet filter for image reconstructed different denoising. In their suggested method, a real-world characteristics of the image and its noise (larger remote sensing image dataset built from AGRI on coefficients denote the high-frequency part, i.e., the FengYun-4An and a learning adaptive wavelet filter details or edges of the image, smaller coefficients (LAWF) was suggested to reduce the variable stripe denote the noise). To increase the efficiency of the noise caused by different SRF. The proposed algorithm denoising suggested a new measurement parameter to determine image methods, according many doi : 10.25007/ajnu.v9n1a587 to works the have been 159 Academic Journal of Nawroz University (AJNU) the appropriate adaptive filter coefficients termed using Speckle Reducing Anisotropic Diffusion (SRAD) Weight Sum Variance of Digital Number Probability filter combination with the discrete wavelet transform (WSVODP). Also, new measurement parameters are by selection Bayesian threshold technique. In their used to measure the denoising image, which is proposed method, the uncompressed noisy image has harshness processed information with separated direction by SRAD filter, then logarithmic (HISD) to demonstrates details of the image instead of transformation (log) is performed on the filtered image traditional measurement coefficients such as (PSNR), to convert speckle noise to Gaussian noise. Finally, the Normalized mean square error (NMSE), average DWT using Bayesian thresholding has been done on gradient value of the image (AGVI). Experimental the log compressed image to obtain a denoising image. results of the presented approach illustrate that the The performance of the proposed algorithm compared algorithm could effectively reduce the variable stripe with the noise from different observation targets, and it was measurement parameters such as PSNR, RMSE, SSIM, prevented ringing artifacts to edge compensation and computational time (s) metrics. Experimental approach, which was equipped by the AGRI scanning results show that the proposed method somewhat model. Therefore, the proposed algorithm is useful in demonstrates reducing variable stripe noise where applied on cloud reconstructed image by making a proper equalization detection image. in speckle-noise reduction and preservation of edges. [18] investigated a new algorithm based on adaptive [20] Presented a combined method based on the median filtering and wavelet threshold function improved wavelet threshold function and median (AMF-WT) to adjust the template size adaptively filtering. In this study, the detail coefficients (high- according to the noise concentration. In their study frequency) mixed with Gaussian noise are denoised adaptive median filtering used to denoising image after the wavelet decomposing of the image. After from two-level reconstructing the wavelet coefficients, two times of decomposition is done on the image noisy by wavelet median filtering is performed for the reconstructed to denoising the high-frequency coefficients with images to obtain the denoised image. The proposed improved the threshold function then reconstructed method denoised the image under the condition of with low-frequency coefficients. The performance measurement proposed method PSNR measurement parameter is experimental results of proposed algorithm show that associated. The experimental result shows that the the denoising effect of the improved threshold investigated algorithm is superior to the adaptive function is superior to hard threshold and soft median filter and median filter in eliminating salt-and- threshold, therefore presented method can effectively pepper noise and can retain more image details and remove the mixed noise in the image, such as edge information. The improved threshold function Gaussian noise, salt-and-pepper noise and speckle designed has a better denoising effect than a hard noise, it has a strong adaptability, a stable denoising threshold function and soft threshold function. effect and can be applied in practical engineering [19] suggested a new method to enhance visual quality image denoising. and effectively eliminate speckle noise of the image [21] suggested denoising image approach basic on salt-and-pepper noise, while other existing methods during some the better parameter visual such quality as of PSNR. the The doi : 10.25007/ajnu.v9n1a587 160 Academic Journal of Nawroz University (AJNU) discrete wavelet transform using Soft Thresholding filter used as preprocessing filter to preserve the useful and Wiener Filter. In their proposed method, Wiener information in the Filter applied on the Approximation coefficient and logarithmic transformation (Log) is applied to convert Soft Thresholding performed on the Detail coefficients speckle noise are remaining in the filtered image into based on db4 wavelet. The performance of the additive presented algorithm evaluated in terms of PSNR and thresholding were used to remove the noise in the MSE. Competitive performance compared to other approximation methods frequency). has been observed. The value of noise, The image while of noisy images, guided sub-image filter and (high-frequency, performance of the a soft low- proposed measurement parameters such as PSNR and MSE for algorithm and the conventional noise reduction denoising image of the proposed method compared methods are compared using the quantitative values with the noisy images which contain AWGN and salt of parameters RMSE, an equivalent number of looks & pepper noise. (ENL), PSNR and SSIM. The experimental results [22] proposed a novel method for speckle noise indicate that the proposed algorithm shows the best reduction of time series SAR images based on wavelet performance regarding speckle noise removal and transform and Kalman filter to preserve the spatial edge preservation in SAR images compared to resolution, and reduce the time of processing in the conventional filtering methods. field of remote sensing. In their study, biorthogonal [24] suggested a new approach to denoising highly spline wavelet and Birge-Massart strategy were chosen distorted images affected by speckle noise. This to obtain denoising thresholds. While two assessment proposed approach is carried out in a homomorphic parameters used which are Equivalent Number of framework using bacterial foraging optimization Looks (EEI) (BFO) cascaded with wavelet transformation and parameters to evaluate the capability of the proposed wiener filter. In their proposed method, the wavelet method its packet decomposition is used to identify and remove effectiveness to preserving the edge structures. The the noise from affected pixels. For pre-processing experimental results of proposed algorithm show that purposes, the Wiener filter used. The BFO algorithm is the method has efficient performance compared to used to reduce the amount of error between the another method for speckle reduction of time series speckled image and the not speckled output image SAR images in the terms of smoothing homogeneous from the homomorphic framework after processing. areas and preserving edges and details and also can The multiplicative noise (speckle) was transformed preserve the major edge structures and the spatial into the additive form by log transformation of the resolution while reducing the speckle noise from time original image. PSNR and MAE are two measurement series SAR images. criteria used in the proposed method of quality [23] proposed a combination algorithm of speckle evaluation of the image. The simulation result of the reducing soft proposed approach shows that the combination of thresholding and a guided filter to effectively remove DWT and wiener filter performs best results in the speckle noise from SAR images while preserving edge form of information preservation of the image by information. In their suggested technique an SRAD removing noise from images, while the combine BFO (ENL) to and Edge-Enhancing reduce the anisotropic doi : 10.25007/ajnu.v9n1a587 speckle diffusion Index noise and (SRAD), 161 Academic Journal of Nawroz University (AJNU) and AMF achieve superior visual and statistical illustrates that the presented approach has more results. effective and efficient than the DTCWT to maintain [25] presented a modified technique that based on more rich details in images. dual-tree and [27] investigated a hybrid image denoising algorithm interpolation. In their proposed method, DTCWT based on wiener filter is to remove the additive white cascaded complex wavelet transform generate different Gaussian noise (AWGN). They develop an image while denoising denoising algorithm based on wiener filter and its approach for the image is used based on dual-tree method noise thresholding using discrete wavelet complex technique. transform to deal with edges and details in images Proposed method focused on the enhancement of the more efficiently. The performance of the proposed resolution of a low-resolution medical image, the Low method compared with Standard Median Filter (SMF), Resolution (LR) of input images decomposed into Relaxed Median Filter (RMF), FPDEF and Bilateral different six Filter (BF), while the filtered images are evaluated decomposition levels. Different quality measurement using quantitative measures Root Mean Square Error criteria such as PSNR, SSIM, and MSE are used to (RMSE), PSNR & Image Quality Index (QI) to measure assess resolution enhancement proposed technique image efficiency. Experimental results of the proposed and to determine the quality of denoised images. algorithm indicate achieves much better performance Findings show that the proposed approach has a better in removing the additive white Gaussian noise from smoothness and accuracy balance of image than the the image with minimum edge blurring compared to DWT and is less redundant than the SWT. other filters. [26] investigated a new denoising image algorithm [28] presented a hybrid filter that combines the using a double-density dual-tree complex wavelet Modified Median Wiener filter (MMWF) and Absolute transform (DDCWT) based on a modified threshold Difference and Mean filter (ADMF). The MMWF filter function for Computed Tomography (CT) image. In merged the complementary qualities and abilities of their proposed study they decomposed the noisy the Median (MF) and Wiener Filter (WF). To improve image the quality of the image, the proposed method frequency structure bands wavelet into used for and frequency high to analysis, wiener filter sub-bands frequency using and low-frequency components during DDCWT while the adjusted suppressed threshold used for DDCWT coefficient in the next step Tomography (CT) medical images for better disease and as usual, it was rebuilt image by reconstructing diagnosing. This high performance compared frequency and low-frequency components the Gaussian noise hybrid with in Computer proposed algorithm similar techniques through the inverse decomposition of DDCWT. Hard depending on two commonly measured in terms of threshold function selected with the chosen universal (PSNR) and (MSE). The simulation result shows that threshold to denoising CT images. For assessing the the hybrid suggested technique article is best suited effectiveness for removing Gaussian noise in CT images modality. of the proposed method, two measurement parameters computed visual quality [29] suggested a new method of image denoising between the original image and the denoised image, based on using median filter (MF) in the wavelet which are (PSNR) and (MSE). The experimental result domain. The hybrid proposed algorithm depended on doi : 10.25007/ajnu.v9n1a587 162 Academic Journal of Nawroz University (AJNU) to apply the DWT only or the median filter only and filter with a high-resolution estimate that will evaluate mixing between them on a noisy image. In their the signal power while preserving the edge details. suggestion a spatial filter which is median filter used Basic on their proposed method, the local covariance is to remove Gaussian noise from noisy scaled image to used blurring and smoothing edges and details, while the coefficients with low resolution while to estimate the wavelet transform applied to analyze the image, WT signal variance in the wiener filter the high-resolution working on the frequencies of sub-bands split from an values used. The results are quantitatively analyzed image. The denoising images are evaluated depending using on two measurement parameters which are MSE and Experimental results indicate that the proposed PSNR. The simulation result of the proposed approach algorithm shows that the combination of DWT and Median filter subjective output; consequently, the proposed method performs significant effects and can recover much activated the reduction of image noise while edge more detail of the original image. details preserving. In contrast with other approaches, [30] suggested a new denoising algorithm to restore the denoising image details have provided better the image corrupted by Gaussian noise based on the visual quality. bilateral filter and its method noise thresholding based [32] proposed a new method for image denoising. The on the dual-tree complex wavelet transform. In this proposed method based on a combination of DWT and proposed technique the bilateral filter removes the MF. In their recommended way, the WT analyzes the noise as well as some details of the image, so these image because it can split the image into sub-bands details are estimated accurately in the wavelet domain and function separately on each sub-band frequency. with MMSE so that the edges and other features of the Also, the noise ratio at the noisy image estimated by a original image are preserved properly, while the sum robust of the bilateral filter output and the detail image will measurement parameters are used to determine the give the final denoised image. BayesShrink has been noise in the image, which is (MSE) and (PSNR) to used in this proposed method as a selection of evaluate the proposed image denoising method. threshold technique to compute each sub-band level in Experimental results of the submitted method show the wavelet decomposition, while PSNR, SSIM, and efficiency denoising images and obtained the best UIQI index values used as measurement parameters to results for image denoising process by using different performance types of wavelet transform filters. proposed method depending on to get high-resolution Peak Signal-to-Noise substantially median coefficients Ratio improves estimator (PSNR). objective parameter. from and Two denoising image. Experimental results show that the [33] proposed a scheme based on Wiener filtering in proposed algorithm is superior compared to another the wavelet domain. In the proposed system, the CT filter algorithm in terms of visual quality which is image is denoised using the concept of Wiener improved significantly as the accuracy of the wavelet filtering and method noise in the wavelet domain. The decomposition and perfect reconstruction of the image concept of their proposed technique presents that the with preserve essential details of an image such as edge extraction improves the quality of denoised CT edges, textures, etc. images [31] proposed a new technique based on the Wiener (DWT) concerning noise suppression and preservation doi : 10.25007/ajnu.v9n1a587 through discrete wavelet transformation 163 Academic Journal of Nawroz University (AJNU) of the structure. With the motivation of combining two detailed parts to increased intensity of the pixels. Then, methods, wiener filtering and thresholding in the they convolved WHFC with detailed coefficients of wavelet domain are performed using the concept of wavelet. The resultant image was obtained during the method noise. To measure the performance of the inverse wavelet transform process, while the resulting suggested way, the performance metrics (PSNR, SSIM) image is visually blurred. Adaptive Wiener filter used used. proposed to minimize the blurriness of the ensuing image methodology gives noise suppressed as well as edge depending to achieve the maximum PSNR. The preserved image. The experimental result of the experimental result shows that this method is perfect proposed method indicates that the quality of CT for noise reduction gives higher PSNR and better visual images enhanced in terms of noise reduction as well as quality of the image. structure preservation. Therefore, the small image [36] Proposed a denoising approach basing on dual- details preservation and no visual artifacts created. tree complex wavelet and shrinkage with the Wiener The performance of the proposed algorithm has filter technique applied for medical images. In this excellent potential to serve for denoising the computed study, tomography images. threshold technique used to select the threshold value. [34] investigated a novel wiener filtering and adaptive They estimate the noise level while hard and soft soft thresholding of wavelet transform coefficients, thresholding functions are used to shrink wavelet which are a combination of spatial and frequency coefficients and compare the efficiency of denoising domain techniques for removing speckle noise. The images based on Peak Signal to Noise Ratio (PSNR) algorithm uses a Wiener filter as a preprocessing stage and Structural Similarity Index Measure (SSIM) to in the spatial domain, and adaptive soft thresholding evaluate the proposed technique. The findings of the of Wavelet transform coefficients in the frequency proposed method show that the denoising images domain. In their proposed algorithm, the threshold at have a better balance of smoothness and accuracy than each sub band is calculated automatically from the the DWT and are less redundant than the Stationary variance (standard deviation) of the subbands. Two- Wavelet Transformation (SWT). level decomposition of the wavelet transform is [13] presented a Hybrid filter is it composite of various considered, and thresholding is applied only to detail filters to remove a mixed type of noise from a digital coefficients. Results compared in terms of Parameters image, a combination of three filters median filter, PSNR, and MSE. Experimental results presented that wiener filter, and bilateral filter. In their proposed the proposed method is efficient in removing speckle algorithm five types of method applied on noisy image noise compared to other algorithms. to get denoising image depending on the performance [35] proposed algorithm is based on wavelet transform of some parameters such as MSE, signal to noise ratio that denoised the noisy image by adding weighted high (SNR), PSNR, RMSE, and SSIM. The experimental pass filtering coefficients (WHFC) in the wavelet result shows that the proposed method can recover domain. They implemented a smart system for much more detail of the original image and provides a estimating the iterative noise variance, which denoised successful way of image denoising compared with the noisy image. They are applying WHFC on the another process in the study. The resultant image of the denoising methods using the universal doi : 10.25007/ajnu.v9n1a587 164 Academic Journal of Nawroz University (AJNU) noisy image that contents mixed noise, to remove The summary of the selected researches that deals with Wavelet mixed noise which exists in the image. Therefore, transformation with a spatial domain filter are shown denoising image with one type of filter inefficient in Table (1) below. to remove noise, while the combined denoising 5. Discussion method with various filters is needed to be more Many factors play a critical role to obtain denoising efficient to eliminate the mixed noisy image as image entirely, such as the threshold and parameter, possible. the image denoising by the using the type of noise and field of the image also have an active role noisy image (window size) to denoising image characteristics must be considered to evaluate the while others concentrate on the edges details and performance critical features of the noisy image. of and eliminating the noise. proposed assessing All Numerous studies focused on the template of the these reviewing in • technologies. By proposed and the • Many techniques developed a simple method to implemented methods of work covered in this remove noise concerning the time of processing to research, the following points can set: get a denoising image, while others developed • Many research focused on medical images complex methods such as DTDWT using a parallel (Computerized Tomography (CT), Ultrasound and combination of two DWTs to obtain denoising X-ray) and another focused on earth science image image which achieved an effective result. Still, (Synthetic Aperture Radar (SAR)) while other there is a lack of image details and take long methods focused on standard image (Lena, processing time. Cameraman, kid, etc.) to remove noise in noisy • parameters such as WSVODP, which is not used in it. suitable for all types of images, especially CT Numerous measurement parameters are value of denoising image compared with noisy image, some of them are new and others are the tradition. The selection of the filter is playing a critical role in the denoising methods. Also, the choice of the filter must be compatible with the type of noise, for example, the performance of Median filter to denoise Salt & Pepper noise is better than other filters, while the performance of the Wiener Filter to denoise Speckle and Gaussian noise is better than Median filter and others filter and so on. • Some research has used new measurement image depends on the field in which the image computed to evaluate technique depending on the • • There are several types of noises exist in the image at the same time, so many works focused on the doi : 10.25007/ajnu.v9n1a587 images, while EEI, ENL parameters suitable for SAR images. Academic Journal of Nawroz University (AJNU) 165 doi : 10.25007/ajnu.v9n1a587 166 Academic Journal of Nawroz University (AJNU) Table (1): The summary of the selected researches that deals with the image denoising by the using Wavelet transformation with a spatial domain filter. Authors and Problem year of Domain Publication measuremen t parameter Denoising to Type of Filtering Advantages Disadvantages Short description performance noise Techniques proposed method (Chen et al. 2019) [17] Variable stripe noise variable in the water stripe vapor band noise remote sensing images of the AGRI affects many application s when one full disk image is separated into ten sub-images for transformin g as soon as possible. (Qian 2019) [18] Hard salt-and- Adaptive threshold pepper Median and soft Filtering threshold and has Wavelet effecting Threshold in image function as (AMF-WT) oscillation and boundary blurring. doi : 10.25007/ajnu.v9n1a587 Adaptive Wavelet Filter reduce the variable stripe noise caused by different SRF Used for specific WSVODP,HIS Introduces a new purpose which is D parameter termed Advanced weight sum Geostationary variance of digital Radiation Imager number probability (AGRI) (WSVODP), which is used to indicate the appropriate wavelet filter coefficients. maximum template can avoid losing more image details and edge information when the noise concentratio n is very high, and the remaining noise is eliminated by wavelet threshold function Maximum PSNR template which is set very large effected, the details and edges of the denoised mage cannot be preserve well. Improve threshold function which has better denoising effect than hard threshold function and soft threshold function. 167 Academic Journal of Nawroz University (AJNU) (Nishu et al. 2019) [19] (Qian 2018) [20] obtain speckle better noise visual quality of the speckle noise reduction and edge preservatio n Remove difficult form of noise which is speckle noise to obtain better visual quality of the reconstructed image by asking a good equalization in speckle noise reduction and preserve of edges. A single Mix noise Improved effectively denoising such as wavelet eliminate method Gaussian threshold mixed noise, cannot noise, function and that it can achieve salt-and- median get better good effect pepper filtering. visual effect, of noise and less loss of removing speckle detail noise information those . also dealing mixed with noises such fingerprint as salt- andimages and pepper printed circuit noise, board (PCB) Gaussian Images in noise and complex noise speckle environment. noise from images. (Longkume Remove r and Gupta noise 2018) [21] while it preserves the importan t elements of the image. Gaussia n, Salt and Pepper noise Speckle Reducing Anisotropic Diffusion (SRAD) in combination with the Discrete Wavelet Transform (DWT) using Bayesian Threshold discrete wavelet transform using Soft Thresholdi ng and Wiener Filter remove noise while it preserves edge and details combine more PSNR, RMSE, than on SSIM technique, which is take long time process Mixed noise PSNR request more than on filter to reduce noise as well, using median filter not effect well when dialing with another noise Denoising PSNR, MSE image dependent on Detail coefficients which is based on db4 wavelet only the uncompressed noisy image has processed by SRAD filter, then log is performed on the filtered image to convert speckle noise to Gaussian noise. While DWT using Bayesian thresholding has been done on the log compressed image to obtain denoising image. Improve threshold function is designed, which is constructed to avoid the defects of hard Threshold and soft threshold. It is not only continuous and derivable, but also has less compression of wavelet coefficients, which solves the oscillation and boundary Fuzziness to a certain degree after reconstruction. Discrete wavelet transform applied on the input image Wiener Filter aoolied on the Approximation coefficient and perform Soft Thresholding applied on the Detail coefficients. doi : 10.25007/ajnu.v9n1a587 168 Academic Journal of Nawroz University (AJNU) (Aghabalaei et reducing Speckle al. 2018) [22] the speckle noise noise from time series SAR images to preserve the spatial resolution as well wavelet transform and Kalman filter Reduce the speckle noise and its effectiveness to preserving the edge structures as well as it reduces the time of processing. Only provide Synthetic Aperture Radar (SAR) imaging systems ENL, EEI (Choi and Jeong 2018) [23] edge Speckle information noise loss when removing speckle noise in SAR images soft thresholding and a guided filter remove such difficulties by eliminating the speckle noise while avoiding the loss of edge information and critical features Only provide Synthetic Aperture Radar (SAR) imaging systems RMSE,ENL, PSNR and SSIM (Dass 2018) [24] Speckle speckle noise noise degrades the visual evaluation of ultrasound images. So the main challenge of not speckling is to preserve all the fine details and the edges of the ultrasonographi c images bacterial foraging optimization (BFO) cascaded with wavelet transformati on and wiener filter gives Cover only recognizable medical images and acceptable restoration of ultrasound images in presence of speckle noise PSNR, MAE doi : 10.25007/ajnu.v9n1a587 all of the images were co-registered with respect to one of the Images, and then the resampling process was performed. Secondly, the first image was despeckled using the biorthogonal spline wavelet and BirgeMassart strategy. SRAD filter used as preprocessing filter to preserve the useful information in the image of noisy images, a logarithmic transformation (Log) is applied to convert speckle noise are remaining in the filtered image into additive noise, while guided filter and soft thresholding were used to remove the noise in the approximation subimage (highfrequency, lowfrequency). The wavelet packet decomposition is used to identify and remove the noise from affected pixels. For pre-processing purposes, the Wiener filter is used and BFO algorithm is used to reduce the amount of error between the speckled image and the not speckled output image from the homomorphic framework after processing 169 Academic Journal of Nawroz University (AJNU) (Tayade and Wavelet Bhosale 2018) techniques [25] are effective to remove the noise due to their ability to capture the energy of a signal in a few energy transform values. (Luo et al. Improves 2018) the [26] performanc e of DWT with important properties: direction selectivity and nearly shift invariant. Not specific dual-tree complex wavelet and Wiener filter enhancement of the resolution of a low-resolution medical image only medical images Covered and more complex Not specific double density dualtree complex wavelet transform (DDCWT) It obtains Highly complex PSNR, MSE direction and get more time selectivity in process higher dimensions and nearly shift invariant. (Kannan 2017) remove the [27] noise while preserving the edges and details in the images who effected by Gaussian noise additive white Gaussian noise wiener filter and its method noise thresholding using discrete wavelet transform achieves much better performance in removing the additive white Gaussian noise from the image with minimum edge blurring compared with the original wiener filter Only efficient on image who contain the additive white Gaussian noise PSNR, SSIM, and MSE RMSE, PSNR, QI DTCWT cascaded structure used to generate different frequency bands for analysis, while denoising approach for image is used based on dual-tree complex wavelet and wiener filter technique They used DDCWT decompose noisy CT image into high frequency and low frequency components. In the next, a modified threshold is used for DDCWT coefficient. Finally, the denoised image is obtained by reconstructing high frequency and low frequency components through inverse decomposition of DDCWT. Wiener filter is often assumed to be unsuitable for images containing edges and details, while to deal with edges and details in images, the method noise thresholding used discrete wavelet transform. doi : 10.25007/ajnu.v9n1a587 170 (Chithra and Santhanam 2017) [28] Academic Journal of Nawroz University (AJNU) suppressin Gaussian Modified remove g the noise Median speckle noise Gaussian Wiener filter from the noise from (MMWF) Ultrasound CT images and Absolute images to in order to Difference preserve edge improve and Mean the quality filter of the (ADMF). images for better disease diagnosing. (Ramadhan et choose a al. 2017) [29] suitable image denoising method to restored details of an image with noise (Majeeth and Babu 2017) [30] Gaussian median filter Has ability to efficient for one noise (MF) and recover much type of noise DWT more detail of the original image. removes Gaussian bilateral filter noise in an noise and its image method noise while thresholding preserves based on the the edges dual-tree complex wavelet transform doi : 10.25007/ajnu.v9n1a587 Cover only PSNR, MSE medical image which is effected by Gaussian noise give high directional selectivity and perfect reconstruction of the edges and other essential details of an image which are preserved Using specific filter to remove just specific type of noise MSE, PSNR PSNR, SSIM, UIQI MMWF used to remove Gaussian noise in the image. This MMWF merges the complementary qualities and abilities of Median (MF) and Wiener Filter (WF). While ADMF used to remove speckle noise from the Ultrasound images). An attempt is made to test whether this filter is used to remove Gaussian noise from the image. median filter used to remove Gaussian noise from noisy scaled image to blurring and smoothing edges and details, while the wavelet transform applied to analyze image bilateral filter removes the noise as well as some details of the image, so these details are estimated accurately in the wavelet domain with MMSE so that the edges and other features of the original image are preserved properly, while the sum of the bilateral filter output and the detail image will give the final denoised image. 171 Academic Journal of Nawroz University (AJNU) (Wang et al. 2016) [31] use a large Gaussian wiener filter window noise with a highsize lead to resolution degrades estimation the image detail Determines and provide the signal power to preserving the edge information Suitable for small PSNR or specific window size of image (Ismael et al. 2016) [32] Enhance the image quality after degraded by the noise. Provide clearly image to preserve edge and details Deal with specific MSE, PSNR amount of the standard deviation of original images. Provide edge extraction to improve the quality of denoised CT images through discrete wavelet transformation (DWT) with respect to noise suppression and structure preservation. Does not provide PSNR, SSIM all CT reconstructed images. Gaussian DWT and Noise MF (Diwakar and When the Gaussian DWT and Kumar 2016) radiation noise WF [33] dose is low, the quality of CT image is degraded in terms of noise. Therefore, improve the quality of noisy CT images, it can be helpful to avoid the need of higher dose CT images. the local covariance is used to get highresolution coefficients from coefficients with low resolution while to estimate the signal variance in the wiener filter the high-resolution values are used Wavelet transform is used to analysis the image due to the ability to split the image into subbands and work on each sub-band frequencies separately. While, the RME has been used to estimate the noise ratio at the noisy image. Two intermediate result are obtained the first intermediate result is subtracted from the input noisy image and processed using wavelet packet thresholding. The outcome of wavelet packet thresholding is second intermediate result. Both intermediate results are added to gain the final denoised CT image. doi : 10.25007/ajnu.v9n1a587 172 doi : 10.25007/ajnu.v9n1a587 Academic Journal of Nawroz University (AJNU) 173 Academic Journal of Nawroz University (AJNU) (Diwakar and When the Kumar 2016) radiation [33] dose is low, the quality of CT image is degraded in terms of noise. Therefore, improve the quality of noisy CT images, it can be helpful to avoid the need of higher dose CT images. (Mohan et al. Reduce the 2016) [34] speckle noise without sacrificing the information content. Gaussian DWT and noise WF Provide edge extraction to improve the quality of denoised CT images through discrete wavelet transformation (DWT) with respect to noise suppression and structure preservation. Does not provide PSNR, SSIM all CT reconstructed images. Two intermediate result are obtained the first intermediate result is subtracted from the input noisy image and processed using wavelet packet thresholding. The outcome of wavelet packet thresholding is second intermediate result. Both intermediate results are added to gain the final denoised CT image. speckle noise Has ability to significantly reduce the speckle noise Only cover (SAR) PSNR, MSE field (Saluja and Boyat 2015) [35] How to remove the noise from the corrupted image as well as preserve the edges and other detailed features Add WHFC and Remove the Consider only additive Adaptive noise from the Gaussian noise white Wiener filter noisy image Gaussian which gives better visual appearance. (Naimi et al. 2015) [36] enhanceme nt the discrete wavelet transform with important additional properties to preserve the edges and details of noisy image Wiener filter used as a preprocessing stage in the spatial domain and adaptive soft thresholding of Wavelet transform coefficients used in the frequency domain Apply WHFC on the detailed parts to increased intensity of the pixels. Then, they convolved WHFC with detailed coefficients of wavelet. The resultant image was obtained during inverse wavelet transform process, while the resultant image is visually blurred. Adaptive Wiener filter used to minimize the blurriness of the resultant image They estimate the noise level and then for DWT, SWT and DTCWT based denoising they used ‘db4’ family wavelets. wiener filtering and adaptive soft thresholding dual-tree complex wavelet and shrinkage with the Wiener filter Improve better balance of smoothness and accuracy than the DWT and are less redundant than the SWT Applied only for medical image using only db4’ family wavelets. PSNR PSNR, SSIM doi : 10.25007/ajnu.v9n1a587 174 Academic Journal of Nawroz University (AJNU) 6. Conclusion International Journal of Mathematical, Engineering Many techniques based on wavelet transform hybrid and Management Sciences, 2018. with some spatial domain filters such as Wiener filter, 3. Nature Median filter, anisotropic filter, bilateral filter and etc., Inspired Optimization Algorithms "Image Denoising Techniques: A Brief Survey". Advances in have been implemented to preserve details and edges Intelligent Systems and Computing 741, springer, of the image as much as possible. Therefore, this work present a reviewed approaches of image denoising of 3. Singh, L. and R. Janghel, Harmony Search and 2018. 731-740. 4. 4. Alisha, P.B. and G.S. K, Image Denoising Techniques- various methodologies which adopted to obtain the An optimum image denoising performance. The above Communication Engineering (IOSR-JECE), 2016. survey of different type of methods, show that some of 11(1): p. 78-84. them depended on using the primary WT (DWT), and 5. 6. thresholding, Bayesian soft thresholding, thresholding). 7. 6. Song, Q., et al., Image Denoising Based on Mean Filter 7. Wang, G., Z. Wang, and J. Liu, A New Image Denoising Method Based on Adaptive Multiscale Morphological Edge Detection. Mathematical Problems Different in Engineering, 2017. traditional and new measurement parameters have 8. 8. Xiaoa, F. and Y. Zhanga, A Comparative Study on Thresholding Methods in Waveletbased Image Denoising. These various methods still lag behind in the better Elsevier (Advanced in Control Engineeringand visual quality of the image. More performance parameters can be calculate to study the behavior of 5. Fan, L., et al., Brief review of image denoising and Sensor Application (AITS), 2015. hard been used to evaluate methods. and Conference on Advanced Information Technology image that to be close to original image as much as thresholding, Electronics and Wavelet Transform. IEEE 4th International pepper, speckle) in noisy image to obtain denoising (universal of for Industry, Biomedicine, and Art), 2019. using filters to remove single or mixed type possible depending on various threshold techniques Journal techniques. Springer Singapore (Visual Computing some of them depending on complex WT (DTCWT) (multiplicative) of noise (Gaussian noise, and salt- Overview. Information Science), 2011. 15. 9. 9. Kethwas, A. and B. Jharia, Image de-noising using the hybrid techniques by making a good compromise Fuzzy and Wiener filter in Wavelet domain. IEEE in various noise reduction fields. A better hybrid International Conference on Electrical, Computer technique with a suitable filter model can be designed and Communication Technologies (ICECCT), 2015. to remove single and mixed noise in multiple images in all areas gives a better visual appearance, without destitution the details of the image and take into account the processing time of the denoising image. 7. References 1. Technology and Engineering (IJRTE), 2019. 7(5S4). 2. Koranga, P., et al., Image Denoising Based on Wavelet Transform using doi : 10.25007/ajnu.v9n1a587 Visu Thresholding Technique. Jagadesh, T. and R.J. Rani, A novel speckle noise reduction in biomedical images using PCA and wavelet transform. IEEE International Conference on Wireless Communications, Signal Processing and Networking (WiSPNET), 2016. 11. 11. 1. Kommineni, V.R.R. and H.K. Kalluri, Image Denoising Techniques. International Journal of Recent 2. 10. 10. Agarwal, S.K. and P. Kumar, Denoising of A Mixed Noise Color Image Through Special Filter. International Journal of Signal Processing, Image Processing and Pattern Recognition, 2016. 9(1): p. 159-176. 175 Academic Journal of Nawroz University (AJNU) 12. 12. Kaur, G. and R. Kaur, Image De-Noising Using Wavelet Transform and Various Filters. International Journal of Research in Computer Science, 2012. 2(2): Kumar, P. and S.K. Agarwal, A Color Image Denoising By Hybrid Filter for Mixed Noise. International Journal of Current Engineering and 23. 23. Choi, H. and J. Jeong, Despeckling Images Using Based Noise Reduction Techniques. IEEE Sensors Journal 2018. 18(8): p. 3131 - 3139. 24. 24. Dass, R., Speckle Noise Reduction of Ultrasound Images Using BFO Cascaded with Wiener Filter and Technology (ijcet), 2015. 5. 14. 14. Geoscience and Remote Sensing Symposium, 2018. a Preprocessing Filter and Discrete Wavelet Transform- p. 15-21. 13. 13. and Kalman Filter. IGARSS 2018 - IEEE International Lin, L., An Effective Denoising Method for Images Discrete Wavelet Transform in Homomorphic Region. Contaminated with Mixed Noise Based on Adaptive Procedia Median Filtering and Wavelet Threshold Denoising. J Inf Conference on Computational Intelligence and Data Process Syst (JIPS), 2017. 14(2): p. 539-551. Science (ICCIDS 2018). 2018. 132: p. 1543-1551. 15. 15. Fan, L., et al., Brief review of image denoising techniques. Visual Computing for Industry, Guhathakurta, R., Denoising of Image : A Wavelet Based Approach. 8th Annual Industrial Automation and Electromechanical Science, International Tayade, P.M. and S.P. Bhosale, Medical Image Denoising and Enhancement using DTCWT and Wiener filter. International Journal of Advance Research, Biomedicine, and Art, 2019. 16. 16. 25. 25. Computer Engineering Conference Ideas and Innovations in Technology (IJARIIT), 2018. 4(4). 26. 26. Luo, P., et al., CT Image Denoising Using Double Density Dual Tree Complex Wavelet with Modified (IEMECON), 2017. Chen, B., et al., Adaptive Wavelet Filter With Edge Thresholding. IEEE 2nd International Conference on Compensation for Remote Sensing Image Denoising. Data Science and Business Analytics (ICDSBA), 2018. 17. 17. 27. 27. IEEE, 2019. 7: p. 91966 - 91979. 18. 18. Qian, Y., Removing of Salt-and-pepper Noise in Kannan, K., A New Hybrid Image Denoising Algorithm Based on Wiener Filter and Wavelet Images Based on Adaptive Median Filtering and Thresholding. International Journal of Research in Improved Threshold Function. 2019 Chinese Control Information Technology, 2017. 1(2): p. 37-45. 28. 28. And Decision Conference (CCDC), 2019. Chithra, K. and T. Santhanam, Hybrid Denoising Nishu, I.Z., et al., A New Image Despeckling Technique for Suppressing Gaussian Noise in Medical Method by SRAD Filter and Wavelet Transform Using Images. IEEE International Conference on Power, Bayesian Threshold. IEEE International Conference on Control, Signals and Instrumentation Engineering Electrical, (ICPCSI), 2017. 19. 19. Computer and Communication 29. 29. Engineering (ECCE), 2019. 20. 20. Qian, Y., Image Denoising Algorithm Based on Ramadhan, A., F. Mahmood, and A. Elci, Image Denoising by Median Filter In Wavelet Domain. The Improved Wavelet Threshold Function and Median Filter. International IEEE Applications (IJMA), 2017. 9(1). 18th International Conference on 30. 30. Communication Technology (ICCT), 2018. 21. 21. Longkumer, M. and H. Gupta, Denoising of Images Using Wavelet Transform,Weiner Filter and Soft Thresholding. International Research Journal of Engineering and Technology (IRJET), 2018. 5(6). 22. 22. Aghabalaei, A., et al., Speckle Noise Reduction of Journal of Multimedia & Its Majeeth, S.S. and C.N.K. Babu, A Novel Algorithm to Remove Gaussian Noise in an Image. IEEE International Conference on Computational Intelligence and Computing Research (ICCIC), 2017. 31. 31. Wang, J., et al., Wiener filter-based wavelet domain denoising. Elsevier - Displays, 2016. 46: p. 37-41. Time Series SAR Images Based On Wavelet Transform doi : 10.25007/ajnu.v9n1a587 176 Academic Journal of Nawroz University (AJNU) 32. 32. Ismael, S.H., D.F.M. Mustafa, and D.İ.T. OKÜMÜŞ, A New Approach of Image Denoising Based on Discrete Wavelet Transform. IEEE World Symposium on Computer Applications & Research, 2016. 33. 33. Diwakar, M. and M. Kumar, Edge Preservation Based CT Image Denoising Using Wiener Filtering and Thresholding in Wavelet Domain. IEEE Fourth International Conference on Parallel, Distributed and Grid Computing (PDGC), 2016. 34. 34. Mohan, R., S. Mridula, and P. Mohanan, Speckle Noise Reduction in Images using WienerFiltering and Adaptive Wavelet Thresholding. IEEE Region 10 Conference (TENCON), 2016. 35. 35. Saluja, R. and A. Boyat, Wavelet based Image Denoising using Coefficients and International Weighted Adaptive Conference Highpass Wiener Filtering Filter. on IEEE Computer, Communication and Control (IC4-2015), 2015. 36. 36. Naimi, H., et al., Medical image denoising using dual tree complex thresholding wavelet transform and Wiener filter. Journal of King Saud University – Computer and Information Sciences, 2015. doi : 10.25007/ajnu.v9n1a587