There are many feature extraction technique to extract feature from an image like PCA, LDA, HOG, Gabor etc. Which technique is better?
Firoz Mahmud Sir, among the mentioned ones, HOG is preferred over others. You can use Gabor Jet or Gabor like features using Convolution filters if you choose to use Convolutional network. HOG tends to perform better when used with conventional ML algorithms like SVM and still, people use it a lot. Usually, LDA is used as classifier (linear) or as dimension reduction technique. Moreover, you can think of SIFT, SURF, ORB, FAST, BRIEF, and so on. These features are scale invariant. But, you can't directly pass these features to your classifiers/ regressor. You need to use Bag-of-words to match the number of feature points and use the features later with your ML algorithm. Moreover, the options will change depending on your input data type. Say, if you use points cloud as input, then you may need to use different type of feature, may be curvature! Thus, you have to choose features carefully depending on objective and data, as the result depends a lot on feature engineering.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts