Computational scrutiny of image denoising method found on DBAMF under SPN surrounding

Received Sep 3, 2019 Revised Feb 27, 2020 Accepted Mar 8, 2020 Traditionally, rank order absolute difference (ROAD) has a great similarity capacity for identifying whether the pixel is SPN or noiseless because statistical characteristic of ROAD is desired for a noise identifying objective. As a result, the decision based adaptive median filter (DBAMF) that is found on ROAD technique has been initially proposed for eliminating an impulsive noise since 2010. Consequently, this analyzed report focuses to examine the similarity capacity of denoising method found on DBAMF for diverse SPN Surrounding. In order to examine the denoising capacity and its obstruction of the denoising method found on DBAMF, the four original digital images, comprised of Airplane, Pepper, Girl and Lena, are examined in these computational simulations for SPN surrounding by initially contaminating the SPN with diverse intensity. Later, all contaminated digital images are denoised by the denoising method found on DBAMF. In addition, the proposed denoised image, which is computed by this DBAMF denoising method, is confronted with the other denoised images, which is computed by standard median filter (SMF), gaussian filter and adaptive median filter (AMF) for demonstrating the DBAMF capacity under subjective measurement aspect.


RELEVANT RESEARCHED OF DENOISING METHODS FOF SPN
Because of diverse reasons such as fault in synchronization of analog to digital process, malfunction of CCD chip, fault in addressing of storage process, and fault in transmission, etc., the impulsive noise [1][2][3][4][5] can be separated into two main categories: Salt and pepper noise (SPN) and random magnitude impulsive noise (RMIN) from algebraic formulation aspect. Hence, may SPN denoising methods [1][2][3][4][5][6][7][8][9] have been examined for more than two decades due to demanding of modern digital image applications [10][11][12][13][14][15][16][17][18][19][20][21][22][23]: retina classification, super resolution, emotion classification, etc. At first, the SMF (Standard Median Filter) [8,24] is discovered in 1975 for excluding SPN (Salt and Pepper Noise) and, later, is become one of the most capable and practical denoising methods from the fact that this method is low computation and high capability. The denoised method based on Gaussain filter [24] is desired for well applying on Gaussain noise but this method has poor performance for SPN. Later, adaptive median filter (AMF) [7,25], which is improved from the SMF denoising method by varing its window size, is proposed and its performance is better than the SMF denoising method. After two decades, the modern decision based adaptive median filter (DBAMF) [3,25], which is improved from the SMF-denoising method, is discovered for denoising RMIN in 2010.
The DBAMF method is formulated from two main schemes: noise classification schemes (that is found on ROAD (Rank Order Absolute Difference)) and noise exclusion schemes (that is found on SMF [5,25] method is implemented on SPN at diverse intensity. Consequently, this analyzed report focuses to examine the similarity capacity of denoising method found on DBAMF [3] for diverse SPN surrounding in order to analytically understand its upper bound of its performance and its limitation for future implementations.

THE PRIMARY CONCEPT OF DBAMF
The distorted portrait is mathematically explained as Y and the portrait intensity is mathematically explained as   , y i j . The DBAMF scheme [3,25] can be separated into two primary schemes: noise recognizing scheme and noise repairing scheme, which can be comprehensively reviewed as upcoming.

The primary concept of noise recognizing scheme
The performing arithmetic concept of the noise recognizing scheme can be reviewed as.
 From NROAD, if the statistical mean of NROAD, which fluctuates between 0.00 to 1.00 for all pixels in the processed portrait, is greater than a stable constant 0 T [3,25] then the processed portrait pixel is recognized as the distorted pixel, otherwise then the processed portrait pixel is recognized as the noise-free pixel. Therefore, the the Noise Detected Matrix can be comprehensively clarified as upcoming.
From the above noise recognizing scheme of the DBAMF, we can comprehensively display this processing scheme in the upcoming flowchart as Figure 1.

The primary concept of noise repairing scheme
The performing arithmetic concept of the noise repairing scheme can be reviewed as.
 Determine the calculated square region 33  From the calculated square region of NDM, if the total noise-free pixels is fewer than three pixels then the dimension of the calculated square region 33  W is expanded by one and the Step 2 is reexecuted.
 From the calculated square region of NDM, if the total distorted pixels is more than two pixels then the repaired pixel is executed by as upcoming.
 The duplicated sheme is re executed for every pixels in the processed portrait pixels. From the above noise repairing scheme of the DBAMF, we can comprehensively display this processing scheme in the upcoming flowchart as Figure 2.

RESULTS AND DISCUSSION
In this examining of the DBAMF denoising capacity, the calculation software in this analyzed report is MATLAB program that is run on workstation computers with the hardware detail: the CPU is Intel i7-6700HQ and the internal memory is 16 GB and all workstation computers simulate on diverse portraits, which are contained of Airplane, Pepper, Girl and Lena, at numerous SPN densities where all diverse portraits that are distorted by adding synthesized SP noise. All distorted portraits are repaired for obtaining the finest quality and best PSNR by executing the image denoising method found on DBAMF for first noise recognizing scheme (in order to recognize whether the pixel is noise-free or noisy) and, later, noise repair scheme (in order to repair only the noisy pixels).

The experimental investigation of noise recognizing scheme
This simulated experiment section investigates the optimized stable constant 0 T for providing the finest quality and best PSNR as shown in Table 1-4. The stable constant 0 T , which fluctuates between 0.00 to 1.00 for all pixels in the processed portrait, ultimately impacts to the denoising capacity of DBAMF method. Consequently, this computer examining comprehensively determines the stable constant 0 T , which make the finest quality and best PSNR when each distorted portrait is executed by denoising capacity of DBAMF method. The numerous digital portraits (which are contained of Airplane, Pepper, Girl and Lena) are used to analyze the stable constant 0 T by varying from 0.00 to 0.50 at 0.025 incremented steps as displayed in Table 1 to Table 4, respectively.  From these computer examining of Girl in Table 1, the optimized pre-specified constant 0 T is about 0.1153 0.0005 or be fluctuated from 0.075 to 0.150 for making the finest DBAMF denoising capacity  From these computer examining of Pepper in Table 2, the optimized pre-specified constant 0 T is about 0.10690.0010 or be fluctuated from 0.025 to 0.150 for making the finest DBAMF denoising capacity.  Table 3, the optimized pre-specified constant 0 T is about 0.11250.0042 or be fluctuated from 0.050 to 0.175 for making the finest DBAMF denoising capacity.  From these computer examining of Airplane in Table 4, the optimized pre-specified constant 0 T is about 0.10970.0014 or be fluctuated from 0.050 to 0.150 for making the finest DBAMF denoising capacity.

The experimental investigation of image denoising method found on DBAMF
This analyzed report focuses to examine the computational scrutiny of image denoising method found on DBAMF under SPN surrounding. In this examining of the DBAMF denoising capacity, four analyzed images, which are contained of Airplane, Pepper, Girl and Lena, are used to analyzed by initially adding synthesized SP noise for creating numerous distorted portraits. Later, all distorted portraits are repaired for obtaining the finest quality and best PSNR by executing the image denoising method found on DBAMF. From the above examining, the denoising methods by applying AMF [7,25] and DBAMF produce the finest quality and best PSNR than other denoising methods for instant SMF and Gaussian filter. However, the DBAMF has marginally improved than AMF from that fact that the DBAMF is initially developed solely for random magnitude impulsive noise (RMIN) but the AMF is developed solely for SPN. From these computer examining of the denoising capacity in Table 5(a)  for Lena and Pepper and Table 5(b) for Girl and Airplane, althrogh the DBAMF denoising method is originally desired for RVIN, the DBAMF denoising method can provide the fine results (the denoised images with high quality). From these inverstigation, the DBAMF method and adaptive median filter (AMF) can produce the denoised image with finer quality and high PSNR, which is confronted with the other denoised images, which is computed by standard median filter (SMF) and Gaussian Filter. Due to circumspection of sheet of paper, some  Figure 4.   Figure 4. The analysis report of noise repairing method of the DBAMF

CONCLUSION
This analyzed report focuses to examine the similarity capacity of denoising method found on DBAMF for diverse SPN Surrounding. In order to examine the denoising capacity and its obstruction of the denoising method found on DBAMF, the four original digital images, comprised of Airplane, Pepper, Girl and Lena, are examined in these computational simulation for SPN surrounding by initially contaminating the SPN with diverse intensity. The first constribution of this report is the optimized stable constant, which is determined from computer simulation at SPN surrounding. Later, the second constribution of this report is the overall capacity of denoising method found on DBAMF, confronted with the other denoised images (SMF and gaussian filter and adaptive median filter (AMF)) under the SPN with diverse intensity.