A Comparative study in wavelets, curvelets and contourlets as feature sets for pattern recognition

Abstract EN

There have been a number of recent works in computer vision that had used new age multiresolution multidirectional transforms like curve lets and contour lets for face and character recognition.

Although these works produced high recognition accuracies they did not provide any comparative study against more well known techniques and hence could not justify the use of these new transforms as against more traditional methods.

In this work we will compare the recognition accuracies of the aforesaid two transforms against a very well known multiresolution transform viz.

the wavelet transform.

This study aims at showing the research community how good or how bad the aforesaid transforms are when compared against wavelets as a feature set for pattern recognition.