Wavelet transform IP block aids image compression

IMEC has developed a local wavelet transform (LWT) IP block to reduce the memory requirements in wavelet-based image compression.

IMEC has developed a local wavelet transform (LWT) IP block to reduce the memory requirements in wavelet-based image compression.

The LWT, based on block-based wavelet techniques, reduces the memory access cost by roughly a factor of three while producing results equal to those of non-block-based wavelet techniques.

Wavelets are increasingly used in image compression schemes because of their intrinsic scalability features, which enable new functionalities such as, progressive transmission, automatic adaptation to network and terminal resources etc.

The LWT additionally supports features like region-of-interest coding in an efficient way.

Wavelets are the technology underlying recent compression standards including JPEG2000 and MPEG-4 visual texture coding.

Currently, they are also being considered for scalable video coding for future versions of MPEG-4 and for space-based compression in the context of the CCSDS (Consultative Committee for Space Data Systems) standardisation.

Mainly due to memory cost (especially for large image sizes), wavelets typically come with a higher implementation cost than traditional block-based techniques such as the widely used JPEG.

IMEC's LWT block computes the wavelet transform on a block basis, resulting in a reduction of the memory access cost by roughly a factor of three, while producing results equal to non-block-based wavelet techniques.

IMEC's LWT functionality has been designed in an instruction-based custom processor with a dedicated instruction set for maximum flexibility of image size, wavelet filter type, block size etc.

The LWT processor requires on average only three to four cycles per pixel, compared with six for state-of-the-art (non LWT) optimised implementation approaches.

To demonstrate its features, IMEC has integrated its LWT processor into a high-speed wavelet-based encoder for large high-quality images (such as remote sensing).

The FPGA realisation of the LWT allows processing of more than 10Msample/s at a clock frequency of 40MHz.

The LWT processor is generically applicable.

All wavelet-based image compression schemes where high-speed and/or memory cost is an issue, can benefit from applying IMEC's LWT.

For example, by using the LWT block together with a JPEG2000 compression stage, the tiling approach that introduces artifacts in the resulting image is avoided and the data-transfer and storage cost is kept under control.

IMEC's patented solutions for implementing the LWT functionality in any multimedia application are available for licensing.

Target companies are IDMs (integrated device manufacturers) or fabless companies that are looking for next-generation multimedia appliances or chip sets to facilitate these appliances.

Back to top