">
">
">
">
Removing Noise in Digital Images Copyright 2003 - JMA - All Rights Reserved |
Noise in digital images can be defined as a random pattern of pixels that give the image a grainy appearance similar to static on a television screen. noise can be attributed to a variety of reasons depending on the source of the image. For this discussion we will focus on two sources of noise. The noise inherent in digital cameras and the noise introduce from scanning high speed film. Digital Camera Noise - In a digital camera light is collected and transferred to a charge coupled device (CCD). The CCD is a light sensitive device containing millions of pixels that store and display image data by converting the light into an electrical charge. The level of intensity of each pixel corresponds to a color in the color spectrum and thus the sum of these millions of pixels form the image. However, depending on the length of the exposure and the ISO setting in the camera, these pixels can also produce noise. The most common types of noise are Thermal noise, also nown as Dark Current noise and Amplification noise. Thermal noise occurs when the exposure time exceeds 1 second. This s due to the different sensitivity of each pixel and its leakage rate. Exposing a CCD long enough will cause the pixels to light up even without light. Amplification noise occurs when the ISO value is changed to a higher number. Digital cameras are designed to work best at one factory setting which is usually the lowes ISO value. Since there is no film in a degital camera, raising the ISO value simply increases the amount of amplification that will be applied to the readout circuitry and because the perfect circuit is yet to be designed, amplifier noise is always added to the image in varying amounts based on the level of amplification. Amplification nose is mostly found in the Blue channel of an RGB image because the CCD, like the human eye is less sensitive to the color Blue. Noise from High Speed Film - With film, the grainy effect is caused by silver halide crystals that come together as the film is developed. When the film is scanned, these crystals are enlarged and become more visible as a random pattern of pixels distributed over the entire image. Thus the higher the ISO rating of the film, the larger the grain size and subsequently the more the noise is visible when the image is digitized. In this tutorial, we wll discuss how to apply the noise filters to the individual RGB channels, a technique that will produce better results than can be achieved using global noise filtering techniques. The subject for our tutorial will be an image scanned from a very grainy 35 mm film negative with noise levels far greater than you will probably ever see with your digital camera. We will filter out the noise, brighten the image, balance the colors and enhance the overall appearance of the image. |
Figure 1. The Raw Image. The image is a scan of a 35mm film negative, ISO 400, Brand X film. As we can see, the noise is excessive throughout the image and most nticeable in the sky, the shed and the cart. Before we apply any noise reduction techniques, we want to identify what global changes can be made to fix the overall appearance of the image. These are fixes that involve color changes, brightness, balance and contrast. These fixes exaggerate filtering artifacts and should be applied prior to any noise reduction filters. |
Figure 2. Adjusting the Color The Blue color cast has created an imbalance in the Red tones making them appear Purple. We want to bring these colors back to normal. To do so, we use the Auto Color Balance command. On the Photo Toolbar select Enhance Photo > Automatic Color Balance. When the dialog window appears, set the Strength to 30 and the Color Temperature to 4500K. This will restore the Red colors to a more natural tone without significantly changing the other colors in the image. |
Figure 3. Adjusting the Contrast There are several commands that will help you to adjust the contrast in this image. These are the Curves and Levels commands. An easier way is to use the Auto Contrast feature available in PSP. In the Photo Toolbar select Enhance Photo > Automatic Contrast Enhancement. In the dialog window set the Bias to Lighter, the Strength to Normal and the Appearance to Natural. Adjusting the contrast will balance the light and dark areas in the image and bring out the details. |
Figure 4. Splitting the Channels As we mentined earlier, noise resides mostly in the Blue channel of the image. To access the Blue channel we need to split the image into the three color channels that make up the RGB color. To split the RGB image, go to the Main Menu bar and select Image > Split Channel > Split to RGB. PSP will separate the image into three Grayscale images representing the individual Red, Green and Blue channels. |
Figure 5. Blue Channel Adjustment The first channel we want to apply the noise filter to is the Blue channel. Place the mouse pointer on the Blue channel title bar and left click the mouse button to make the Blue channel image the active image. To this channel we want to apply the Edge Preserving Smooth filter. Go to the Main Menu bar and select Adjust > Add/Remove Noise > Edge Preserving Smooth. In the dialog window enter 10 for the Amount of smoothing. |
Figure 6. Red Channel Adjustment Repeat the procedure for the Red channel. Place the mouse pointer on the Red channel title bar and left click the mouse to make the image active. Once again, apply the Edge Preserving Smooth filter, but this time set the amount of smoothing to 5 becaus the Red channel is not as noisy. In fact it's about half. |
Figure 7. Recombining the Channels to RGB Once the filter has been applied, we can re-combine the channels back into an RGB image. To do so, go to the Main Menu bar and select Image > Combine Channels > Combine from RGB. PSP will create a new RGB image with the filtering applied. Notice how the filter has already improved the image. But we are not done yet. Close all the Grayscale images and the original RGB image (the unfiltered image). Now let's apply the Moire filter. One of the features of the Moire filter is that it will clean up stray interference caused by the Green channel when images are re-combined into RGB. To apply the filter, go to the Main Menu bar and select Adjust > Add/Remove Noise > Moire filter. When the dialog window opens, set the Fine Details value to 1 and the Remove Bands to 0. |
EPS Filter |
Unfiltered |
Figure 9. Making the Sky Selection The noise filtering we have applied so far has removed most of the noise in the image, but t is still pronounced in the sky area. Areas that contain soft smooth color require a higher degree of filtering and applying higher levels to the whole image would destroy most of the detail. We need to separate the sky area from the rest of the image. To do so, we use the Magic Wand and select the sky as shown. For this image the wand tolerance was set to 20 and the atch mode was set to RGB. After the selection is made Feather it by 6 pixels to smooth out the transition when the noise filter is applied. Then go to the Main Menu bar and select Adjust > Blur > Average to smooth out the entire sky. |
Figure 10. Saturating the Colors At this point we want to saturate the colors a bit and bring back some of the intensity without making them look phony. Here you can apply the Color Bance feature or for a quicker way you can use the Auto Color Saturation feature. On the Photo Toolbar, select Enhance Photo > Auto Saturation Enhancement. When the dialog window appears set the Bias and Strength to Normal. |
Figure 11. Sharpen the Image With all the color and filter changes complete we can now apply some sharpening to the image to bring back some of the details. On the Main Menu bar select Adjust > Sharpen > Unsharp Mask (USM). When the dialog window opens set the Radius to 0.6, and leave the Strength and Clipping to their default values. With these types of images it is important not to apply too much sharpening or some of the noise may re-appear. Zoom in on the critical areas to see the effect of the USM filter. |
Figure 12. Before and After Comparing our before and after images we can see that the noise has been removed. The image is brighter with more natural looking colors and details. Remember that noise in digital images is unavoidable. With proper care it can be minimized and with the tools available in PSP it can be virtually eliminated. |
Before |
After |
