Future Research
I summarized a list of possible future research topics below. I estimate the importance (application significance and possible improvement) on 1-5 scale and success (probability of achieving good experimental results on currently used databases) on 100% scale.
| Research Topic | Importance | Success |
|---|---|---|
| 1. Approximating optimal low complexity combination function in identification systems by iterative methods (current work) |
5
|
80%
|
| 2. Modifying 1. to use identification models; thus try to achieve the approximation to optimal medium II complexity combination in identification systems |
4
|
70%
|
| 3. By using duality of medium I and medium II type combinations, investigate the user specific combinations (medium I) type corresponding to 'second best score' identification models |
5
|
95%
|
| 4. Investigate usage of backround models for combinations, especially if class relationships can be defined (words in the lexicon) | 3 | 70% |
| 5. Use both identification models and background (user-specific) models to construct efficient high complexity combinations |
3
|
90%
|
| 6. Utilize independence assumption to construct non-density based combinations of independent biometric matchers |
2
|
90%
|
Notes
- I am note completely sure if it would be possible to prove that approximations found in 1. will be truly approximations of optimal combination functions.
- The experimental improvements of 2. will be probably quite small, but still statistically significant for NIST BSSR1 database.
- The expected improvements in 6 will be most probably very small.