Multi-Hypothesis Disparity-Compensated Light Field Compression Prashant Ramanathan EE368B Final Project December 6, 2000
Overview Light fields Light field compression Multi-hypothesis techniques Results Prashant Ramanathan
Light Fields I Represents the appearance of a scene or object from all viewing positions and directions Planar Light Field University of Erlangen Computer Graphics Group Prashant Ramanathan
Light Fields II Problems with light fields: Difficult to acquire Very large Buddha 200 MB uncompressed Night 95 GB uncompressed Stanford University Computer Graphics Laboratory Prashant Ramanathan
Light Field Compression Disparity-compensation (Magnor, Girod 2000) target image reference images reference images Prashant Ramanathan
Multiple Hypothesis Specify reference views to use and independent disparity values reference view 1 reference view 3 Prashant Ramanathan
Experimental Data Two sets of images from Flieger light field target image reference images reference images Prashant Ramanathan
Experiments Three schemes tested: STD: one disparity H1: reference index H2: two disparities + reference indices Prashant Ramanathan
Results I 4x4 blocks H2 H1 STD Prashant Ramanathan
Results II 32x32 blocks H2 H1 STD Prashant Ramanathan
Conclusions and Future Work Multi-hypothesis prediction reduces residual error Overall compression gain? Theoretical analysis? Prashant Ramanathan