Introduction and Discussion on Logistic Discriminant Metric Learning
13:15:00 - 13:45:00
308 , Mathematics Research Center Building (ori. New Math. Bldg.)
In visual identification, variations of one person result from different poses or movements may be larger than variations between different people. This in turn makes classification problem very difficult. Recent studies tried to use Mahalanobis distance as a measure of distance between images. One method called Logistic Discriminant Metric Learning (LDML) trains Mahalanobis distance using image data and showed good performance. In this talk, I will introduce the idea of LDML and limitations of it. Finally, I will give an alternative design concept, which is still inchoate.