Bag-of-words (BoW) image description has shown good performance for a large variety of image recognition scenarios. We investigate approaches to alleviating a standard BoW image description pipeline representations for the specific task of recognizing pork loins. Specifically, we extend the BoW description to include depth maps, perform non-rigid image registration to align the images, and apply PCA dimensionality reduction on the BoW descriptors. Our results show that the combination of image registration and PCA yields a more distinctive recognition.
Workshop on Farm Animal and Food Quality Imaging 2013: Espoo, Finland, June 17, 2013, Proceedings, 2013, p. 49-54
Main Research Area:
Dtu Compute-technical Report-2013
Workshop on Farm Animal and Food Quality Imaging 2013