How to combine two features HOG and LBP to get better accuracy?
15 views (last 30 days)
I am working on people detection with training datasets. I used two different different feature extraction, HOG and LBP. Each of them will extract 324 feature vector (HOG) and 59 feature vector (LBP) for each training image. I am currently using simple concatenation to combine bothe features and then the results will be feed into SVM classifier to be classified. However, seems like this method gives a domination to HOG.
I have tried to find out another method, which is using fusion features, yet I have no knowledge how to do the algorithm. I really need help from the expert who know about the fusion feature to combine two features.
Really thank you
Nikolay S. on 16 Mar 2015
Hi there. I've worked on a project where we have combined multiple features to achieve best detection. HOG and LBP were some of the used features BTW :). What you've been doing so far is a good point to start. You can try PCA on top fo that to reduce irrelevant features, preserving most dominant ones. Voting can also be nice- you apply an SVM to each feature vector, and the you have to combine their decisions- you can have "and"/"or"/"majority vote" schemes, to decide true (detection) or false (no detection). All above are relatively easy methods, with good potential to succeed. Another thing is use Cascade & Boosting- as Viola & Jones did. This is a powerful tool that will reject features in ascending order. But you will need to invest effort to get into it. Best regards, Nikolay