Define Preprocessing techniques for images
Closed this issue · 0 comments
DiegoHerrera262 commented
Define an algorithm for preprocessing training and test images for obtaining better correlation outputs. Some ideas include:
- Try zeroing u-v axis in image spectra to reduce edge effects.
- Implement normalization before filter construction and correlation computation to reduce intensity influence on the output.
- Try adaptative histogram equalization.
In order to check effectiveness of those techniques the following methodology is proposed:
-
Compare correlation metric with 1 test image and 1 training image before and after preprocessing, for both true and impostor classes.
-
Repeat process for different test - training pair. And asses qualitatively whether preprocessing increases performance.
-
Build several-image filter with preprocessed training set and compute correlation metric with preprocessed test image and raw image. Asses whether performance is better than one-image filter built with and without preprocessing.