The fast growing of inter-net and multimedia services has generated a great deal of interest in storage and transmission of digital images and video streams. Image interpolation becomes an important technique because it can play a very important role in such image/video applications. Image interpolation tries to produce a higher-resolution image from an original image with a lower resolution. In the past, image interpolation was usually performed by simple pixel replication in a small neighborhood of the original pixel. However, such scheme is too simple which is usually unable to give us satisfactory solution....[ Read more ]

The fast growing of inter-net and multimedia services has generated a great deal of interest in storage and transmission of digital images and video streams. Image interpolation becomes an important technique because it can play a very important role in such image/video applications. Image interpolation tries to produce a higher-resolution image from an original image with a lower resolution. In the past, image interpolation was usually performed by simple pixel replication in a small neighborhood of the original pixel. However, such scheme is too simple which is usually unable to give us satisfactory solution.

In this thesis, we consider some new image interpolation approaches. The first one is to use Wiener and Volterra filters for doing interpolation. In contrast to those traditional interpolators, a unique feature of these interpolators is that they can make use of the information of desired pixel values and the image statistics to predict the missing pixels, while the prediction is optimal in the minimum-mean-square sense.

The next approach proposed in this thesis makes use of the DCT-domain zero-masking technique. The basic idea of it is to set a sufficient number of high-frequency transformed DCT coefficients of an image block to zero in order to derive the coefficients of the interpolator. This interpolation scheme finds solid supports from human visual system that is less sensitive to high frequencies than to low frequencies. Furthermore, the design of such interpolators is independent of the particular image under consideration.

The performances of all those interpolators are tested by Very-Low-Bit-Rate (VLBR) image compression using our JPEG interpolative coding system. It has been shown by our extensive simulation results that our interpolators can usually perform better than the traditional interpolators. Moreover, in this thesis, we have also proposed an Error Concealment (EC) algorithm by using pixel-interleaving and the DCT-domain zero-masking technique. Its superiority over other existing error concealment methods has been examined when there is packet loss during transmitting a JPEG-coded image in the ATM networks.