I have training sets of each of the indivdual objects shown in this following image. This one image is also the query image:
Four of these training sets have 18 images each, consisting of images cropped down and taken at different angles. I also have another training set of another object. That object should not be detected with this queried image.
Objective: Identify all of the objects in the query image. Could be any uniform background, hoping to go toward any background also (cluttered scene, random background). My research and questions:
Should I be using the cascade object detection which uses the Viola-Jone algoritm and discount SURF features? I would give negative examples in this case.
Use KNN (k=2) method? I tried using this and the accuracy isn't that great.
Use SVM? Don't know to much about this but I know that it has something to do with bag of features...
Just use matchFeatures function iterative and set up a ratio of number of features matched over the number of features in a training set.
Final question: I need some direction toward my objective. Any links and direction would be greatly appreciated. If you refer me to one of my four notes I made, please explain.