IMAGE interpolation is getting more and more attentions because it plays an important role in many video applications such as image compression, interlaced- to-progressive conversion, and pyramidal image coding. In the past, image interpolation was usually performed by pixel replication in a small neighborhood of the original pixel. Later on, higher-order linear filters were used to enhance the performance. However, all these filters are only based on the neighborhood pixels to interpolate the missing pixels. Therefore, when an image is transmitted at a high compression rate, the distortion in the received image is very serious. As a result, we may not be able to get a good interpolated image based on the highly distorted neighbor pixels....[ Read more ]

IMAGE interpolation is getting more and more attentions because it plays an important role in many video applications such as image compression, interlaced- to-progressive conversion, and pyramidal image coding. In the past, image interpolation was usually performed by pixel replication in a small neighborhood of the original pixel. Later on, higher-order linear filters were used to enhance the performance. However, all these filters are only based on the neighborhood pixels to interpolate the missing pixels. Therefore, when an image is transmitted at a high compression rate, the distortion in the received image is very serious. As a result, we may not be able to get a good interpolated image based on the highly distorted neighbor pixels.

In this thesis, we develop three new approaches, namely neural network approach, Wiener filtering approach, and subband Wiener filtering approach. In contrast to the traditional linear filtering methods described above, a unique feature of these interpolators is that they are able to make use of the information of desired pixels (i.e. the ideal pixels which we want to interpolate) and the image statistics to predict the missing pixels, while the prediction is optimal in the least-mean-square sense.

We use the JPEG based interpolative image coding scheme [l] as an example to test our new interpolators. It is shown by the simulation results that our interpolators give better performance than traditional interpolators.