Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Implementation and comparison of GPU accelerated (CUDA) algorithms for feature extraction and matching

: Iturbe, Mikel
: Kähm, Olga

Mondragon, 2012, 91 S.
Mondragon, Univ., Bachelor Thesis, 2012
Bachelor Thesis
Fraunhofer IGD ()
comparison; biometrics; feature extraction; object recognition; image matching; Business Field: Digital society; Research Area: Confluence of graphics and vision

This work presents a tattoo identification system based on GPU-accelerated feature extraction and matching algorithms. The system first extracts features from the tattoo image. Then it matches the features found in the input image with a variety of features previously saved in a database or a group of stored images. The most similar image in the database is defined as the image that has the highest number of matches with the input image.
The system implements a new feature descriptor vector, based on a previous similar implementation. To enable the remote storage of feature data, a database connection is also added to the system.
Final results show that the newly implemented descriptor performs better than the previous existing implementation and it is suitable to use in tattoo identification systems. Thanks to its GPU acceleration, the system is faster than other traditional CPU implementations.