[1]
Visual data transmitted in the form of a digital image is becoming a major method for visual communication in today's era, but the image obtained after the transmission is often corrupted with many types of noise. Noise is an important factor which, when get added to an image, reduces the quality and appearance. So in order to enhance the image quality, it must be removed with preserving the textural information and structural features of image. There are different types of noises exist which corrupts the images. Selection of the de- noising algorithm is application oriented [17]. Here in this paper, two filters were used; one is trim median filter used for estimation and removal of noise from image corrupted with Salt and Pepper and the other filter is Gaussian filter which is used for Gaussian noise estimation and reduction [16] . This is clearly a better algorithm because it is based on a modified decision based system. In this paper, the authors propose a modified decision based modified trim median filter algorithm for the restoration and effective suppression of gray scale and the color images that are highly corrupted by salt and pepper noise. The authors also calculate the presence of Gaussian noise in any noisy digital images. The authors implement a GUI for estimating the density of saltpepper noise in degraded images using joint entropy value and mutual information. The joint entropy value between the noisy image and the original image or other typical images was introduced in this paper to depict the inter-correlation.
The system is tested against the different color and grayscale images and its gives the better [2] Peak Signal-to-Noise Ratio (PSNR) and Image Enhancement Factor (IEF).
The image restoration and de-noising has a very important role in Digital Image Processing. Noise removal from highly corrupted images is one of the greatest challenges among the researchers, noise removal algorithms vary with the different applications, areas and the type of different images as well as noises. Before giving input to image processing tools, the most important task here is to de-noise the image first. In review paper [15], the author’s have already explained the noise interfacing and removing methods. In this paper, the authors are trying to show the comparison between median filter and modified decision based trimmed filter. The authors are testing this by using many sample images. In this paper, an approach has been proposed for estimating the density of salt-pepper noise in images using joint entropy value. The joint entropy between the noisy image and the original image or other typical images was introduced in this paper to depict the intercorrelation. It is revealed that, how the joint entropy [2] value changes along with the noise density, and is indicated that this relation is robust to individual image traits. Finally, this relation is represented in quantitative form and the authors took advantage of this relation to estimate the noise density.
In this paper, Noise can be removed by using Gaussian filter and median filter. Also in this system, a comparison between modified decision based trimmed filter and median filter are done by the authors. The performance and quality factors of the image are also calculated. This system will work on gray scale images.
In the survey paper [15], the authors have already discussed about noise interfacing and removing techniques. In this paper, the authors extend the previous system and add some GUI models for filter comparison and noise estimation. The main focus of this paper is for trimmed median filter, modified decision based unsymmetrical trimmed median filter algorithm has been used for getting the better results. The image quality enhancement factors are calculated for perfect output. The noise is estimated using joint entropy and mutual information between input original gray image and noisy image as well.
A modified decision based unsymmetrical trimmedmedian filter algorithm for the restoration of gray scale images and color images [6], that are highly corrupted by the salt-and-pepper noise is proposed in this paper. The proposed algorithm replaces [10] the noisy pixel by trimmed median value when the other pixel values, 0's and 255's are present in the selected window and when all the pixel values are be 0's and 255's, then the noise pixel is replaced by mean value of all the elements present in the selected window. This algorithm shows better results than [3] the Standard Median Filter (MF). The proposed algorithm is tested against different grayscale and color images [11], [14] and it gives better Peak Signalto- Noise Ratio (PSNR) and Image Enhancement Factor (IEF). The Modified Decision Based Unsymmetrical Trimmed Median Filter algorithm processes the corrupted images by first detecting the impulse noise. The processing pixel is checked whether it is noisy or noise free. That is, if the processing pixel lies between maximum and minimum gray level values, then it is a noise free pixel, it is left unchanged. If the processing pixel takes the maximum or minimum gray level, then it is a noisy pixel which is processed by MDBUTMF.
The steps of the MDBUTMF are elucidated as follows [1].
Step 1: Select a 3 x 3 2-D window for image. Assume that the processing pixel of image is Oij , which lies at center of window.
Step 2: If 0 < 0
Step 3: If Oij = 0 or Oij = 255, then it is considered as corrupted image pixel and two cases are possible as given below.
Case (i): If the selected pixel window has all the pixel value as 0, then Oij is replaced by the Salt noise (i.e. 255).
Case (ii): If the selected pixel window contains all the pixel value as 255, then Oij is replaced by the pepper noise (i.e. 0).
Image quality is a characteristic of an image [16] , [18] image that measures the perceived image degradation (typically, compared to an ideal or perfect image). Imaging systems may introduce some amounts of distortion or artifacts in the signal, so the quality assessment is an important problem.
Peak Signal-to-Noise Ratio, often abbreviated PSNR, is an engineering term for the ratio between the maximum possible power of a signal and the power of corrupting noise that affects the fidelity of its representation [5]. Because many signals have a very wide dynamic range, PSNR is usually expressed in terms of the logarithmic decibel scale.
PSNR is most commonly used to measure the quality of reconstruction of loss compression codecs (e.g., for image compression). The signal in this case is the original data, and the noise is the error introduced by compression. When comparing compression codecs, PSNR is an approximation to human perception of reconstruction quality. Although a higher PSNR generally indicates that the reconstruction is of higher quality, in some cases it may not. One has to be extremely careful with the range of validity of this metric; it is only conclusively valid when it is used to compare results from the same codec (or codec type) and same content.
PSNR is most easily defined via the Mean Squared Error (MSE). Given a noise-free m×n monochrome image I and its noisy approximation K, MSE is defined as also the PSNR (in dB) is defined as:
Here, MAXI is the maximum possible pixel value of the image. When the pixels are represented using 8 bits per sample, this is 255. More generally, when samples are represented using linear [9] Pulse Code Modulation (PCM) with B bits per sample, MAXI is 2B-1. For color images with three RGB values per pixel, the definition of PSNR is the same except the MSE is the sum over all squared value differences divided by image size and by three. Alternately, for color images the image is converted to a different color space and PSNR is reported against each channel of that color space, e.g., YCbCr or HSL.
Mean Square Error plays an important role in image quality and image compression.
The MSE is expressed as:
where MSE acronym of Mean Square Error, M x N is the size of the image, 
denotes the original image, Y shows the de-noised image, and η represents the noisy image.
The authors need to IEF factor to find the enhancement accuracy of image. The aim of image enhancement is to improve the interpretability or perception of information in images for human viewers, or to provide better input for other automated image processing techniques.
The IEF is expressed as,
Salt-and-pepper noise [8] has much more remarkable magnitudes than image signal, [12], [13] thus the gray levels of the noise pixels are usually digitalized as the extremism of entire grayscale area. In case of the 8-bits presentation, we can model the salt-pepper-noise if the noise density is φ :
where f, g denote the grayscale values for the original image and the noisy one, [7] respectively. P(x) denotes the probability that the value gets x. It implies that about N pixels are polluted by noise if the image has N pixels.
Estimation of the salt-pepper noise in a noisy image can be summarized with the following steps:
Note that the joint entropy value is calculated in step (1), the discretizing manner should be the same as the manner of obtaining the apriority knowledge, i.e., here the grayscale area of each image is divided into 50 statistical sections, too.
An original image with no noise will become a noisy image when it is polluted [4] by salt-pepper noise. The noisy image contains more and more noise points while the noise density becomes larger and larger. The imposed noise points usually are not significant to the original image because of their independent grayscale values, therefore the authors conclude that when the noise density becomes larger and larger, the joint entropy value between the noisy image and the original one (or other typical images) [6], [8] gets smaller and smaller because their inter-correlation gets weaker and weaker. To verify such conclusion, number of images are taken and dirtied them with each level noise.
The authors have created a GUI based model easy to control the image properties. In this paper, three sub-GUI model and a master GUI model are used. One is noise addition and filters control GUI and second is filter comparison and third is noise estimation. This model consists of various steps. Each steps take an important responsibility in this paper. At the starting of noise addition and filter control GUI model in the paper, we need some of the sample input and noisy images, where the image has different values of salt and pepper and Gaussian noise. So a noise adder system is needed. For this system, the original RGB image which is of any size and dimensions are loaded. After image loading using GUI model, the rgb2gray command is applied to gray conversion and then cmap property is applied and image in written in a particular folder. Different noise values are applied which is to be random. The salt and pepper noise and Gaussian noise in the gray converted image is applied. Then the median filter and Gaussian filter is applied. After applying the filter, the output which have minimum noise (noiseless) is considered. After getting an output, “imsubs” function is applied for subtraction of pixel between noisy image to original image. These results have the noise data. Figure 1 shows the GUI model for noise interface and removed using filter in various images.
Figure 1. A GUI model for Noise Interface and Removal using Filter in images
At the second GUI model, the authors try to compare between two filters, Median filter and Trim median filter. The noisy image is loaded, which has an unknown value of salt and pepper noise. Both the filters are applied and then a minimum noise output (Noiseless) is obtained. They estimate and analyse both the filters as in Figure 2.
Figure 2. Filter Comparison between Median Filter (MF) and the MDBUTMF
In this system, many sample images are considered and different noise parameters are applied to get image quality enhancement factors and error factors. Table 1 shows the results of comparison between MF and the MDBUTMF.
Table 1. Results of Comparison between MF and the MDBUTMF
In the third, GUI model is the main aim of this system, here the authors try to estimate the noise data using joint entropy and mutual information as represented in Figure 3. Five sample images are applied for testing the system as in Figure 4. The results of joint entropy and mutual information with different noise levels are given in Table 2.
Figure 3. Noise Estimation using Joint Entropy and Mutual Information
Figure 4. (a) Sample 1 Image, (b) Sample 2 Image, (c) Sample 3 Image, (d) Sample 4 Image, (e) Sample 5 Image
Table 2. Results of Joint Entropy and Mutual Information with Different Noise Levels
In this paper, two filters, namely median filter and trimmed median filter are compared successfully. The output finds a better result in trimmed median filter. This helps to clear maximum value of noise easily. The noise data and plots in the first GUI model are also estimated. This paper can be worked in very noise effective areas of image processing.
In this paper, the authors have a conclusion for estimation and filter comparison. It is concluded that the trimmed filter is better than the median filter. It is time consuming, but shows more accurate signals in image processing. The joint entropy and the mutual information estimator is applied for estimation of percentage of noise in input images. This system can be used in many areas where high density salt-and-pepper noise are present.
The author would like to acknowledge her gratitude to a number of people who have helped her in different ways for the successful completion of thesis. The authors take this opportunity to express a deep sense of gratitude towards her guide Mr. Devanand Bhonsle, Senior Assistant Professor (Electrical & Electronics), Faculty of Engineering & Technology, SSTC-SSGI, Bhilai for providing excellent guidance, encouragement and inspiration throughout the project work. Without his valuable guidance, this work would never have been a successful one. The author is also thankful to Mr. Chinmay Chandrakar HOD, ETC Department, and Dr. P. B. Deshmukh, Director, SSTC-SSGI, Faculty of Engineering & Technology, Bhilai, Mr. Somnath Baksi IEEE Student Co-ordinator for their kind help and cooperation. The author feels immensely moved in expressing her indebtedness to her parents whose sacrifice, guidance and blessings helped her to complete her work.