Image Processing: The Two Main Types of Noises


Whenever production of digital images is done, several activities may be involved. These include introduction of various types of noise into the digital image. More often, this is quite challenging to the producer since introduction of noise into images affects quality of visual image in a substantial manner. There are various types of noise, among these include Gaussian and Salt n Pepper noises. The other approaches that have been discussed include filtering techniques, among others. In filtering techniques elements such as mean as well as wiener filter have been discussed. For instance, by taking the mean filtering techniques, several aspects have been considered. These include arithmetic as well as geometric methods. In addition, various parameters of noise image such as variance, mean and density have been observed to affect the visual quality of the output in a considerable manner. Therefore, their particular effects have been analyzed thoroughly by altering parameters and then comparing the resulting output. After all, effects of change on size of window on the final image have also been considered. That is, windows of dimensions 3 x 3, 5 x 5 and 7 x 7 were treated separately. For successful execution to be achieved, it is quite essential to involve use of customized MAT LAB functions. Moreover, self-designed codes can also be utilized for displaying output images as desired. Furthermore, they can be employed in specifying filtered windows. This paper will try to explore the two main types of noises namely, Salt n Pepper and Gaussian.

Random Noise Generation

In this case, both Gaussian noise and Salt n Pepper noise are considered since they are the most common types of noise in digital media. The process involves introduction of all the noisy images with all its parameters into the original image. This can be achieved using MAT Lab’s “imnoise” function. For a good investigation to be carried out, the two noises are usually used. For instance, in thius case, Gaussian noise was used as an input in one of the samples while Salt n Pepper was utilized in the remaining sample. This was done in order to investigate the output in this procedure. Both noises (Gaussian and Salt n Pepper) were therefore introduced into the original image through a randomized distribution. In this process, mean and variance of the Gaussian noise were employed as inputs to the “imnoise” function.

Throughout the experiment, process of introduction of noise was accompanied by maintaining mean at 0.0 while at the same time varying variance as shown, that is from 0.01, 0.1, 0.2 to 0.5. Ultimately, It was observed that by increasing variance of noise signal, the corresponding result was decrease in visual quality of the resultant image. This made it much more difficult to identify the many distinct features of the image. For instance, from the observation, a variance of 0.01 gave a resulting image in which tiny details of the image were still observable in visual clarity. However, in this case, increase in variance of the Gaussian noise led to blurring of features of the image. This included even the most conspicuous of all.

From the above shown figures, it can be seen that the higher the values of mean the brighter the image. Similarly, when a closer observation is made on the second image, it is found that visual quality aspect of white features of the image that appears to the left becomes fainter. Further on, when a mean of 0.5 is used, the there is no white features are observable. Instead, the black features become more dominant. Clearly, it can be said that increase in variance of Gaussian noise leads to a corresponding increase in noisy features in the image. This reduces visual quality of the resulting image. On the other hand, higher mean values of the Gaussian noise leads to increased brightness of the resultant image.

Again, the process above was repeated, but this time using Salt n Pepper noise in lace of Gaussian noise. This was carried out as follows; Salt n Pepper noise was introduced onto an original image by use of “imnoise” function. This was done to investigate its effect on visual quality of the image. As opposed to the first case, density “d” of noise was chosen as the variable. Therefore, different values of density “d” of noise were employed in determining extent of distortion of the pixels by noise in the original image. For example, when Salt n Pepper noise with a density of 0.1 was introduced into an image, the resulting activity was a distortion of 10% of the pixels. It is also quite essential to note that since Salt n Pepper noise constitutes of white (salt) and black (pepper) pixels, care should be taken to ensure that their probabilities are maintained at equal levels before introduction into the image. The figures below show original images alongside three other noisy images in which Salt n Pepper noise has been introduced with the value of “d” being varied from 0.1, 0.2 and 0.5.

Image De-noising by Mean Filters

This section involved introduction of Gaussian randomized noise into the original image. This was done to get a noisy image. The extra noise was then removed by using the two mean filters known as arithmetic and geometric. Therefore, the next step involved observation and analysis of visual quality of the de-noised image. This was then compared with other images that resulted after the mean filter was varied from arithmetic to geometric. As is required, the effect of changing window size from 3 x 3 to 5 x 5 was also noted. To ensure that both the 3 x 3 and 5 x 5 image windows performed as required, a padding exercise was done on the noise image. This was attained using MAT LAB. The resulting images are as shown below. Although arithmetic mean filtration was employed with the first image using a 3 x 3 window, a 5 x 5 window was used for the second

Several problems were encountered when filtration process was conducted using mean filters. However, the main one includes the fact that it did not give a very good product. This was quite astonishing since the image produced by use of the 5 x 5 window was smoother than that produced using the 3 x 3 window. Overall, the much suitable 5 x 5 image had most of its black features brightened up. Nonetheless, these features could still be observed from the image produced by use of the 3 x 3 window. The following figures tell what happens when geometric filter is used to de-noise the image by use of both 3 x 3 and 5 x 5 windows. An observation made on the 3 x 3 window produced better visual quality image than those produced by the 5 x 5 window. Moreover, because window 5 x 5 was large, over-smoothening of the image occurred, in filtration process. Subsequently, an impression of residual noise could be seen on the black features of that image.

In comparison, it was observed that geometric mean filters produced images of better visual quality than those produced by arithmetic mean filters. Furthermore, black elements could easily be recovered using geometric filtration process. This could be done in a better way than that of arithmetic process. Moreover, a similar scenario was observable when each of the two window sizes was employed.

From the experiment above, it can be deduced that the 3 x 3 window produced the best visual quality image. However, some traces of pepper could still be seen on it. Likewise, In case of the 5 x 5 window, resulting image was outstanding although smoothening up of the image was still difficult. Besides, a 7 x 7 window on the other hand restored the top quality image. Nonetheless, this was not perfect as it showed over-smoothed features, which had the capacity to compromise some features of the image colors. This is quite evident in the 3-pins located at top side of the image. A different observation that could be seen was on the 7 x 7 window, in which treated features that were smaller than its block we considered as part of noise. This resulted in lower visual quality of the final image. This is well illustrated in the 8-dots that can be observed around the white circular feature that appears on top of the image. Consequently, it can be deduced from these results that the 5 x 5 window is offer the most suitable and optimized output. This is mainly because of it has the capacityto provide a more balanced output. In fact, its output neither is over-smoothed nor contains excessive traces irremovable noise.


Clearly, the size of window is found to affect output of de-noised image in different ways. The effect can be described to be in both geometric mean filter. Arithmetic, as well as in other methods of filtering. Moreover, the value of noise parameters have been found to influence the resulting visual quality of de-noise image. In essence, visual quality is proportional to the value of noise parameters which is utilized. Another important factor is the kind of filter that can be used. This is mainly because different kinds of filters determine the type of noise that can be removed from the image. Nonetheless, it is quite important to note that de-noising of images were successful in all cases, although there was a notable difference in visual quality of the digital images.

Cite this paper

Select style


Premium Papers. (2022, April 5). Image Processing: The Two Main Types of Noises. Retrieved from


Premium Papers. (2022, April 5). Image Processing: The Two Main Types of Noises.

Work Cited

"Image Processing: The Two Main Types of Noises." Premium Papers, 5 Apr. 2022,


Premium Papers. (2022) 'Image Processing: The Two Main Types of Noises'. 5 April.


Premium Papers. 2022. "Image Processing: The Two Main Types of Noises." April 5, 2022.

1. Premium Papers. "Image Processing: The Two Main Types of Noises." April 5, 2022.


Premium Papers. "Image Processing: The Two Main Types of Noises." April 5, 2022.