For imagery captured under different time, illumination and zenith and azimuth angles, What is the best algorithm to perform feature extraction?
Feature extraction is loosing its focus. Todays state-of-the-art technique is feature learning with convolutional neural networks.
In fact, more attention has been paid to learning features automatically, e.g., by means of deep neural network. With the development of high performance computing, I believe machines can complete this task more effectively and more efficiently.
More details can be found in the following Wiki page:
https://en.wikipedia.org/wiki/Deep_learning
Deep learning is efficient and offers insights into features. Certain versions, e.g., stacked sparse autoencoders, directly support feature extraction, but due to the cost of the computations they are not favored. Speedup was explored by Gregor and LeCun (ICML 2010), but memory requirements were too high. Light sparsification can be still advantageous. A related paper is this: https://www.cs.nyu.edu/~gwtaylor/.../zeilertaylorfergus_iccv2011.pdf
Sorry I think I misinterpreted your question. I supposed you would like to use the extracted features for image classification task. If so, then CNN (Convolutional Neural Net) is a good choice: feature learning (extraction) and classification is done in the same network. CNNs will probably learn the features that are 'best' for the training database in multiple representation levels. Eg. on some layers CNNs often learn a kind of edge detection. Furthermore, one and two dimensional CNNs are successfully applied in speech recognition and other signal processing tasks (eg. EEG classification), and 3D CNNs can be used for video streams. Computationally CNNs are considered efficient as they are using shared network weights for the filters and there are a number of tricks to decrease test error rate.
The paper András Lőrincz mentioned is a very good read. Besides I suggest the following paper about how CNNs are used for image recognition:
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems (pp. 1097-1105).
http://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf
Am very thankful from you Balint, My question regarding to extract the features from different viewing point and illumination not regard to classification, any way thank you so much for your reply.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Algorithms