I'm working on building a classifier that needs to find one particular object in the photo. I'm planning on using SIFT/SURF + kmeans for feature extraction and logistic regression for classification. I do have training set with images of the object but what should I use for "no_object_found" class. Training set with random images that do not include this object?
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Training set with random images that do not include this object?
I think that's a good start. In many previous works, like Cascade Classifier, the procedure is to prepare some negative and positive samples as the train set. The positive patches (or images) are the ones that contain the target object, and the negative patches are kinda any arbitrary scenes.
So, if you're going to dealing with your problem in the frame of classification, using arbitrary non-target images does make sense, as we see it has been exploited in previous works in computer vision community.
Hope it helps.
