This thesis presents several new blocking effect reduction techniques based on DCT for very low bit rate DCT-coded images. Some common techniques, such as POCS, the use of an interleaved image block before encoding, and the application of a linear space invariant low-pass filter over the decoded image, developed in these two decades. However, there are some drawbacks in each of the existing methods. These include a very long convergence time for POCS, the difficulty of adopting the international standard for using an interleaved image block before encoding, and blurring of the image details for using a linear space invariant filter....[ Read more ]

This thesis presents several new blocking effect reduction techniques based on DCT for very low bit rate DCT-coded images. Some common techniques, such as POCS, the use of an interleaved image block before encoding, and the application of a linear space invariant low-pass filter over the decoded image, developed in these two decades. However, there are some drawbacks in each of the existing methods. These include a very long convergence time for POCS, the difficulty of adopting the international standard for using an interleaved image block before encoding, and blurring of the image details for using a linear space invariant filter.

Because of these drawbacks, we develop a new algorithm to reduce the blocking effect. Our algorithm is fast, simple and straightforward, retains the original bit rate, and preserves the varying characteristics within the image. Our algorithm can be used to change some of the DCT AC coefficients of the shifted image block, and then apply IDCT to obtain a processed image.

There are two proposed methods to change the DCT coefficients. One involves setting some of the DCT AC coefficients to zero. This is called the zero-masking technique. The other method is to multiply the DCT AC coefficients by certain weighting. This is called DCT coefficients weighting technique.

Also, we propose two methods for shifting the image block. One method involves shifting, individually, the image horizontally and vertically. That is, to first shift the image blocks either horizontally or vertically, applying the zero-masking technique or DCT coefficients weighting technique on the shifted image block and, finally, repeating the above procedure in another direction. We call this processing method the one-dimensional processing technique. Another way to shift the image block is to do in both horizontally and vertically at the same time. Similarly, we apply the processing technique on this shifted image block and obtain the processed image. We call this processing method the two-dimensional processing technique.

A lot of simulations prove that our proposed algorithms give both qualitative and quantitative improvements.