Improved image decompression for reduced transform coding artifacts

Abstract

Abstract--The

Variable projection for near-optimal filtering in low bit-rate block coders

“…Before examining the experimental results, we shall comment about the computational burden involved in each iteration of Brent's algorithm. Each such iteration requires coding the image after applying the filter , followed by a decoding stage and computation of the residual in (9). In our experiments on typical 512 512 images, the execution time per iteration was 3.92 s, when executed on a P4 2-GHz machine, using Matlab.…”

“…For the design of the lowpass filter , we use the window method with a Hamming window to design a separable filter with cutoff frequency . Consequently, a suboptimal solution of (5) is given by the lowpass filter , where is given by (9) This is a univariate minimization problem, which we solve using Brent's classic algorithm for minimization without derivatives [20]. We note that the optimization problem (9) is solved once at the encoding stage, and the resulting decimation filter need not be sent to the decoder as it is not used in the decoding process.…”

“…Among these we count techniques based on adaptive filtering [4], [5], projection on convex sets (POCS) [6]- [8], Markov random field modeling [9], [10], and many others (see [11] for a review). Most of these methods work well when the image is compressed at high or medium bit rates, but the majority of them fail to produce satisfactory results when applied on images compressed at low bit rates.…”

Post-Processing for Restoring Edges and Removing Artifacts of Low Bit Rates Wavelet-Based Image

“…For block-based discrete cosine transform (DCT) coding, the coded images usually suffer from the blocking effect across block boundaries and the ringing effect around edges. Due to the widespread use of DCT-based coding, many post-processing techniques have been proposed, among which most of them aim at suppressing the blocking effect as methods (4) (5) , and some can also reduce the ringing effect as methods (6) (7) . Recently, the discrete wavelet transform (DWT) has attracted considerable attention for image coding due to its unique joint space-frequency characteristics.…”

Improving MPEG-4 coding performance by jointly optimising compression and blocking effect elimination

“…In order to remove this blocking artifacts while maintaining the compatibility with current industrial standards, various post-processing techniques such as block-boundary filtering [2][3][4][5], maximum a posteriori (MAP) methods [6], projections onto convex sets (POCS) [7][8][9][10], transform domain filtering [11,12] and others [13][14][15] have been proposed. Most of these post-processing techniques were proved to be very effective to eliminate the blocking effect.…”