Our motivation in this project was to achieve sample image frames with a minimal number of analog measurements. Thus achieving: compression before A/D
Abstract
Our motivation in this project was to achieve sample image frames with a minimal number of analog measurements. Thus achieving: compression before A/D, Energy saving and ability to Increase camera fps performance. Assuming typical Sparseness of natural Images, reconstruct Image from samples by minimizing L1/L2 error norm in the Sparse Space.
Reconstruction Results
Conclusions
- Although Most of the image energy is concentrated in few frequency coefficients, weak coefficients are still visually very important
- For small Block sizes (<=32×32) DCT Dictionary is better than Wavelet Dictionary
- Empiriclly for Lena we need a ratio of 5 between the image DCT coefficients and the number of random samples
- For compression ratio better than 2, Random samples aren’t compatible. it is better to sample low frequency DCT coefficients