Moving Average Smoother - Essay Example

Laboratory Report Number Moving Average Smoother The moving average filter can be used and defined ina plethora of ways. For one, it can be referred to as the technique of smoothing an image by reducing the amount of intensity difference between one pixel to the next one. It can also be simply noted, from this application, as the smoothing of an image by blurring the image details. If the span is increased, on the other hand, the image obtained will be more blurred and containing fewer details. In essence, the moving average smoother can be employed to minimize noise in images. Experiment 1.1 Observations The obtained images after the experiment were all characteristically original images, smooth images and noisy images. There were two distinct kinds of images observed after the experimental procedure – the train images and the fork images. Both these original images obtained had high contrast with sharp edges and more detail. When some noise was added to the two original images, two types of noisy images for the train and the fork were observed. In addition, to smooth the images and reduce/ remove the noise in the noisy images, the moving average filter was employed on these particular images. The first span to smooth the noisy images was 10. The noise in both images was reduced, the images smoothed and the image details blurred when the first span was 10. With the moving average, however, the edges of both the images became blurred and when compared with the original images, the images with span 10 exhibited lesser contrast and detail. On the other hand, when the smoothed image with span 10 was compared with smoothed images with span 20 and 30, the smoothed images with span 10 exhibited the best results in terms of contrast and detail. The next span used was 20, where the noise in both images was eliminated and disappeared completely along with the images becoming smoother and the images edges more blurred hence a considerable difficulty discerning the image details. The resultant two images were, in addition, clearly more obscure. The third span used for both the image types was 30. Herein, the resulting images were smoother than the other images and more blurred too (in comparison with the other images). With span 30, the details of the smoothed images were unrecognizable with the edges more blurred than other smoothed images with span 10 and 20. Conclusively therefore, the moving average filter smoothes the image by blurring the image details with increase in average pixel value creating lower contrasts. In addition, smoothing reduces the noise level of the images. On the negative side however, the resulting images posses less contrast and detail compared with the original images. In a moving average filter, a high span value results in a more blurred image with edges increasingly blurred too. The moving average filter is thus notably not suitable for some images especially those images that have high noise levels. Experiment 1.2 In the three images, when the span value was increased with edge truncate command still on, the images appeared blurred with decreased contrasts, with the images smoothed. The blurring is hence increased with increasing span values with the edges remaining blurred however. If the span value is increased, the blurring and the smoothing covers all the image details with the noise and edges reducing gradually as can be witnessed from the above three kinds of filtered images (with the moving average filter) accomplished by removing the edge truncate. The three images are filtered using various span values. These resulting images are not smoothed as the noise persists on all the three of them. We can hereby conclude that the filter sixe is directly proportional to the edges obtained, that is, if the filter is a small one (25 by 25) the edge will be smaller, and if the filter is large (40 by 40 or 70 by 70), for instance, the edges also resultantly become larger. From the previous images also, we conclusively observed that the centers of the images are smoothed while the images with no edge truncates appear more blurred than the images with the edge truncates. In addition, the 3 images exhibit lesser contrast than their corresponding images with edge truncates. We can therefore conclusively state that if the span value is increased, the unsmoothed edges will increase which results to more obscure details on the image. Experiment 1.3 With Edge Truncate From the procedure, there were observed four different kinds of images. These were the original image, the noisy image, the filtered image with span value of 25 (that is, filter size 25 by 25) and lastly, the filtered image with span value of 40 (that is, filter size 40 by 40). From observations, the two filtered images were blurred and smoothed with the details lost and the edges in both the two filtered images blurred. The image with span 40 showed more blurring than that with span 25. In contrast, however, the noise is significantly reduced in both the filtered images with the noise appearing much less on the span 40 image (eliminated noise). In addition, as compared to the original and noisy images, the filtered images exhibited more white and black colors. The original image had more details and the best contrast of all the four images. Without Edge Truncate In this case, the same filtered images were filtered using the without edge truncate command. There are witnessed images with large edges with noise when the span value is high. The noise can be seen in the edges in these high values since the images produced are blurred in the center. Therefore, the contrast should be reduced as the center of the image with high span is regarded as a very blurred region with very little details observable therein. With a low span value on the other hand, the image has blurred details and is smoothed, with the blurred region covering most of the image. However, this blurring does not affect the edges of the image. On the other hand, with the command edge truncate, the blurring covers all area in the image and the image itself is smoothened. In addition, the edges of the image are blurred and the noise reduced. As for the without truncate command, the blurring does not cover all the areas of the image and the image is smoothened. The edges of the image remain free from blurring and with high span values the noise appears on the edges not covered by the blurring. If the span is high, this kind of blurring and smoothing is impacted by the center of the image. The high span value results more obscure images with the possibility of losing images. Also, worthy of note is the fact that the filter size can have some impact on the final quality of the image. Median Filter The median filter, also referred to as the non-linear filter, considers each pixel in the image. Apart from functioning to reduce and do away with noise in an image, the median filter works to substitute the pixel values with the median of those values. Median filters also work to preserve the sharp edges in the images. The median filter/ non-linear filter technique is a more robust method than linear filters, therefore. Experiment 2.1 In the procedure, salt and pepper noise have been added to the original image. The result obtained is a noisy image with a shot-noise-strength-3000. The noisy image is then filtered using the moving average filter with span 5 and the median the median filter, also with span 5. For the image obtained from the moving average filter, the filter generates a new value dependent on the pixels in each kernel resulting in a new unrealistic value. This image from the moving average filter is a low contrast image and has some noise. On the other hand, the image obtained from the median filter is smoother and better than the moving average filter image. Additionally, use of the median filter means that a single very unrepresentative pixel in the surrounding will not have any significant impact on the median value and therefore the median filter is deemed far superior to the moving average filter. The median filter preserves the sharp edges of the image. Experiment 2.2 In this set-up, the shot strength is increased to 30,000 with the result showing more noise in the image. In a sequence, the moving average span was altered by 5, 10 and 15 and then the median span changed by 5, 10 and15 too. These changes were then applied to the image. The resulting images indicated that black object lose their original colors and become brighter when the moving average filter with span 5 was used. On the other hand, when the median filter with span 5 was used instead, the image maintained the original value. This was so because the median filter, instead of averaging or mixing the original with the noisy pixel, uses pixels to give a new value. The image of the span 10 value of the moving average filter becomes more obscure while the image of the span 10 value of the median filter has its noise eliminated and the black objects still maintaining some of their sharp edges. A blurry image is yielded when the span is changed to 15 with the moving filter image. As for the median filter image, at span value of 15, the amount of noise is completely eliminated with some slight loss of detail with the edges shifted. The observations are that the filtered images with the median filter are much better than the images produced with the moving average filter. This is because the median filter allows more pixels through and removes the noise without significant impact on the original pixels as the filter also preserves the sharp edges. The moving average filter on the other hand unpalatably blurs the image details and edges as the sharp edges are lost in the process. Experiment 2.3 If the amount of noise in the image is significant, the median filter becomes ineffective. This ineffectiveness of the median filters has the implication that more than half of the pixels in the filter matrix are noisy pixels and therefore the median value will be the noisy pixel in which case the median filter is rendered useless and ineffective. As is observable from the images of the experiment, the moving average filter can yield better images than the median filter if the shot noise is greatly increased. In such an instance, use of a median filter can create more noise which essentially implies that if the shot noise was raised to 50,000 and above, the median filtering could be ineffective. Experiment 2.4 Pit scan images with shot noise of 30,000 are obtained then the noisy image filtered with different span values with the moving average filter and the median filter. The span values used were 5, 10 and 15. The results obtained indicated that the median filter was better than the moving average filter with these kinds of images. However, if the noise is increased to the high level, the median filter functioning is limited. The median filter, for the pit scan images, is better than the moving average filter for two main reasons. For one, the median filter is more robust than the moving average filter in that a single unrepresentative pixel in a neighborhood does not considerably impact the median value. Secondly, the median filter preserves sharp edges more effectively than the moving average filter as it does not make new unrealistic pixel values when the filter straddles an edge. 3.3 Convolving with User Defined Kernels There are four different types of images. These are the original image, the noisy image, the moving average image and the kernel filtered image. The kernel filtered image is obtained by filtering the noisy image by the Gaussian filter. This image exhibits more noise and is much sharper than the moving average image. The moving average image is however much more blurring than the kernel filtered image. Compared to the original image, the kernel filtered image which is the brightest of all the four images, has lesser contrast. In terms of clarity, the kernel filtered image is clearer than the noisy image and the moving average filtered image. The edges of the kernel filtered image, compared to the moving average filtered image, are sharper. Therefore in comparison to the moving average filter, the user defined kernel filter can enhance image quality by significantly reducing the noise while simultaneously maintaining the image detail and keeping the edges comparatively entire. The noise from kernel filtered images can however not be completely eliminated and therefore some contrast is lost in the process alongside some of the image details and information. Read More