Information Technology Reference
In-Depth Information
between the micro-images produced by the recording microlens array, is present.
This is due to the small angular disparity between adjacent microlenses. In order
to maximise the efficiency of a compression scheme for use with the 3D holo-
scopic image intensity distribution, both inter and intra micro-image correlation
should be evaluated.
In the last decade, a lossy compression scheme for use with 3D holoscopic im-
ages, making use of a three dimensional discrete cosine transform (3D-DCT) has
been developed [45, 46]. It was shown that the performance with respect to com-
pression ratio and image quality is vastly improved compared with that achieved
using baseline JPEG for compression of 3D holoscopic image data. More recently
a wavelet-based lossy compression technique for 3D holoscopic images was re-
ported [47]. The method requires the extraction of different viewpoint images
from the 3D holoscopic image. A single viewpoint image is constructed by ex-
tracting one pixel from each micro-image, then each viewpoint image is decom-
posed using a two dimensional discrete wavelet transform (2D-DWT). The lower
frequency bands of the viewpoint images are assembled and compressed using a
3D-DCT followed by Huffman coding. It was found that the algorithm achieves
better rate-distortion performance, with respect to compression ratio and image
quality at very low bit rates when compared to the 3D DCT based algorithms [47].
The 3D wavelet decomposition is computed by applying three separate 1D
transforms viewpoint images. The spatial wavelet decomposition on a single
viewpoint is performed using the biorthogonal Daubechies 9/7 filter bank while
the inter-viewpoint image decomposition on the sequence is performed using the
lifting scheme by means of the 5/3 filter bank [48]. All the resulting wavelet
coefficients from the application of the 3D wavelet decomposition are arithmetic
encoded.
5.1 Preprocessing of 3D Holoscopic Images
Prior to computation of the forward DWT, different viewpoint images are ex-
tracted from the original 3D Holoscopic image. The viewpoint image comprises
pixels of the recorded object scene corresponding to a unique recording direction
as discussed in section 4. The post-processing stage at the decoder essentially un-
does the effects of pre-processing in the encoder. The original nominal dynamic
range is restored and each pixel from each reconstructed viewpoint image is put
back into its original position within the microlens to reconstruct the whole 3D
holoscopic image. The intensity distribution of an omnidirectional 3D holoscopic
image consists of an array of micro-images as shown in figure 4. The intensity dis-
tribution is sampled so that each micro-image comprises (8×7) pixels. Since a
viewpoint image is obtained by extracting one pixel from each micro-image pro-
vided by the 2D array arrangement, a total of 56 different viewpoint images are
constructed. It is important to point out that the viewpoint image is different from
the traditional 2D image. It is a parallel projection recording of the 3D space
rather than a perspective projection as in the common 2D recording.
