By Mirsad Makalic
Supervisor: Dr. Peter Tischer
Introduction | Submissions | Deblocking Filter | Super Resolution Filter
The technological advancement in telecommunication and computer technologies has greatly increased the demand for video and image compression. Multimedia applications that require both sound and video contain large amounts of information that is often transmitted through low-bandwidth communication channels. Lossy image compression is by far the most popular form of image compression due to the very good compression ratios while maintaining good image quality. The most common lossy image compression scheme is the block based transform approach. This scheme utilizes the discrete cosine transform (DCT) on an 8x8 block basis. The coarse quantization of the DCT coefficients introduces artifacts into the reconstructed image. The most common artifact is blockiness. This is simply a discontinuity between 8x8 block boundaries due to the incorrect reconstruction of DCT coefficients. The purpose of deblocking filters is to remove the blockiness from the reconstructed image. The most significant problem with a large number of the current deblocking filters is that they introduce blurring into the image. This is usually because the filter does not take into account the structure of the image. The developed deblocking filter attempts to solve this problem by locally segmenting the image before filtering. This allows the filter to take into account the actual structure of the image, thus, reducing the blurring from the filtered image. The second part of the project was to adapt the solution to the deblocking filter to a super resolution problem.
Figure 1 - A low bit rate JPEG image with severe blocking artifacts