Video Sampling involves high data transfer rate : very demanding on the data channel. After the sampling the video is compressed.
Abstract
Video Sampling involves high data transfer rate : very demanding on the data channel. After the sampling the video is compressed. By altering the sampling process data transfer rate may be lowered. This will result in energy saving and/or enhanced camera capabilities.
Video Compression Basics
- Image is sampled in Frequency Domain.
- The image is Quantized with different sensitivity for each frequency coefficient.
- Image prediction Based on Motion Vectors
Motion Vectors
- Usually motion vectors are calculated in pixel domain using correlation.
- There are known methods for motion estimation in Frequency Domain using correlation or phase difference, using the entire image/block
Phase Estimation
- Motion Estimation Without Ambiguity and movement in one direction
Block Correlation
- Uses Lena
- Two axis motion, with changing velocity
- Full sample for boundary blocks and Key Frames
- Sample only 5% of coefficients (6 coeffs) for all other Blocks
Conclusions
- Phased base Motion Estimation works for DFT but can not be easily used with DCT and Hadamard-Walsh
- Correlation Based method was sucessfully implemented using Hadamard-Walsh Transform
- When not in Key Frames, sampling savings of up to 95% were achieved for the analyzed videos
Further work
- Multi-resolution support
- Optimization of the ambiguity solving for the phase estimation method
- Calculating phase from DCT coefficients
- Realization of more complex motion scenarios in the block correlation method