Information Technology Reference
In-Depth Information
5.2 2D WDT Based Compression Algorithm
The general structure of the 2D wavelet based compression algorithm is shown in
Figure 9. The input to the encoding process is a 3D holoscopic image. Prior to
computation of the 2D DWT, different viewpoint images are extracted from the
original 3D holoscopic image. The viewpoint image components are then decom-
posed into different decomposition levels using a 2D WDT. The 2-D transform is
performed by two separate 1-D transforms along the rows and the columns of the
viewpoint image data, resulting in four frequency subbands.
The lowest frequency subband is a coarse scale approximation of the original
viewpoint image and the rest of the frequency bands are detail signals. The 2D
transform can be applied recursively to the lowest frequency subband to obtain
decomposition at coarser scales. In [47] a two-level of decomposition was applied
by means of the Daubechies 9/7 filter.
Quantization
then Huffman
coding
Input Data
3D DCT on
lowest bands
2D WDT on
Viewpoint
Images
Preprocessing
Quantization and Arithmetic
coding of the remaining bands
(a) Coder
Compressed Data
Dequantization
then Huffman
decoding
Reconstructed Intensity
Distribution
3D IDCT on
lowest bands
2D Inverse
WDT on
Viewpoint
Images
Postprocessing
Dequantization and Arithmetic
decoding of the remaining
bands
(b) Decoder
Fig. 9 The general structure of the proposed scheme: (a) Encoder, (b) Decoder
After decomposition of the viewpoint images using the 2D-DWT (Figure 10(a)),
the resulting lowest frequency subbands are assembled as shown in Figure 10(b)
and compressed using a 3D-DCT. This will achieve de-correlation within and be-
tween the lowest frequency subbands from the different viewpoint images. The 3D
DCT is performed on an 8×8×8 volume. Hence, the 56 lowest frequency subbands
are assembled together, giving seven groups of eight viewpoint images each. The
Search WWH ::




Custom Search