Adaptive Image Process

We have already seen why Wiener filter and its variations blur the image significantly is that a fixed filter is used through out the entire image. The Wiener filter was developed under the assumption that the characteristic of the signal and noise do not change over different regions of the image. This resulted in space- invariant filter. In a typical image however, image characteristic differ considerably from one region to another. For e.g. walls & skies have approximately uniform background intensities while buildings & trees may have large detailed variations in intensity. Degradations may also vary from region to region. It is reasonable to adapt the processing to the changing characteristics of the image degradations. Idea of adapting the processing is useful not only in image restoration but also in image enhancement .

Two approaches to adaptive image process have been developed.

1) In one approach called " pixel by pixel processing ", process is adapted at each pixel.

At each pixel, the process method is determined based on the local characteristic of the image, degradation, and any other relevant information in the neighborhood region surrounding that pixel.

Advantage:- since each pixel is processed differently, this process is highly adaptive and does not suffer artificial discontinuity in the processed image.

Disadvantage:- However approach is computationally expensive and is typically implemented only in the spatial domain.

2) Another approach called subimage-by subimage or block-by-block process. In this method an image is divided into many subimages and each subimage is processed separately and then combined with the others.

The size of the subimage is typically between to pixels. For each subimage, a space invariant operation appropriate to subimage is chosen on the basis of the local characteristic of image, degradation and any other relevant informing in the subimage region. This space invariant operation offers more flexibility in implemention compared to pixel by pixel method.

For e.g. A LPF may be implemented in spatial or frequency domain. In addition, subimage-by subimage processing is generally computationally less expensive than pixel-by-pixel since the type of processing to be performed must be determined only once for the whole subimage. Because the type of process changes abruptly as we move from one subimage to next, artificial discontinuity may appear along boundaries of adjacent subimages. This is referred to as blocking effect. It is an objectionable feature in low SNR environment and in low bit rate transfer of coding. In image restoration in high SNR environment blocking effect may not be visible.