Reduction of Blurring and Additive Random Noise

In processing an image, it is important to keep intermediate results with high accuracy to avoid a possible harmful effect of quantization.

A small amount of quantization noise may not be visible by itself but can be amplified by subsequent processing. For example, suppose we process an image degraded by blurring and additive random noise with a noise reduction system followed by a deblurring system. If the output of the noise reduction system is quantized to 8 bits/pixel, the quantization noise will not be visible at this stage. However, the subsequent delburring system, which is typically a highpass filter, can amplify the quantization noise, and the noise may become clearly visible in the final result. Since the effect on the final result due to quantization in the intermediate results often is not straightforward to analyze, it is worthwhile to keep intermediate results with high accuracy.

In this section, we have discussed the problem of restoring an image degraded by two specific types of degradation. The idea of reducing one type of degradation at a time may be applicable to other types of degradation. Specifically, when an image is degraded by degradation1, followed by degradation 2, and followed by degradation 3, one approach to consider is to reduce degradation 3 first, degradation 2 next, and finally degradation 1. Once the overall system consisting of subsystems is developed, it may be made more computationally efficient by rearranging the subsystem. Such an approach, although not always optimal, often simplifies the restoration problem and is in some cases is an optimal approach in that it leads to the same solution as the approach that treats all the degradations simultaneously.