Information Technology Reference
In-Depth Information
Object depth was calculated from the viewpoint image displacement using a
depth equation, which gave the mathematical relationship between object depth
and correspondence viewpoint image displacement. To improve the performance
of the disparity analysis, an adaptation of the multi-baseline technique taking ad-
vantage of the information redundancy contained in multiple viewpoint images of
the same scene was used [37]. The idea of viewpoint image extraction on 3D
object reconstruction was also reported by Arimoto and Javidi [38].
The 3D holoscopic imaging requires only one recording in obtaining 3D infor-
mation and therefore no calibration is necessary to acquire depth values. The
compactness of using 3D holoscopic imaging in depth measurement was soon at-
tracting attention as a novel depth extraction technique [39, 40]. In the conven-
tional stereo matching system, the quantization error is increased with the object
depth and a considerable quantization error will be caused when the object depth
is large. While different to the conventional stereo vision method, the quantization
error obtained from the extracted viewpoint images is maintained at a constant
value and irrelevant with the depth [37, 40]. To take the advantage of both, Park,
et. al. proposed a method for extracting depth information using a specially de-
signed lens arrangement [40]. A drawback of the work reported in [37, 38, 39, 40]
is that the window size for matching has to be chosen experimentally. In general, a
smaller matching window gives a poor result within the object/background region
while a larger window size gives a poorer contour of the object.
More recently, a method was reported which addresses the problem of choosing
an appropriate window size, where a neighbourhood constraint and relaxation
technique is adapted by considering the spatial constraints in the image [41]. The
hybrid algorithm combining both multi-baseline and neighborhood constraint and
relaxation techniques with feature block pre-selection in disparity analysis has
been shown to improve the performance of the depth estimation [41].
Another method which uses a blur metric-based depth extraction technique was
proposed [42]. It requires the estimation of plane objects images using the compu-
tational 3D holoscopic imaging reconstruction algorithm. The algorithm was
shown to extract the position of a small number of objects is well defined situa-
tions. However, the accuracy of the depth map depends on the estimation of the
blur metric which is prone to large errors, as these metrics are sensitive not only to
the threshold used to classify the edges, but also to the presence of noise. The
scope of the 3D holoscopic imaging application also has been further extended to
3D object recognition [43, 44].
5 3D Holoscopic Image Compression
Due to the large amount of data required to represent the captured 3D holoscopic
image with adequate resolution, it is necessary to develop compression algorithms
tailor to take advantage of the characteristics of the recorded 3D holoscopic im-
age. The planar intensity distribution representing 3D holoscopic image is com-
prised of 2D array of micro-images due to the structure of the microlens array
used in the capture and replay. The structure of the recorded 3D holoscopic image
intensity distribution is such that a high cross correlation in a third domain, i.e.
Search WWH ::




Custom Search