Concatenate features for image classification

Hello all,

If I want to concatenate features extracted from fully connected layer of a CNN with features extracted from a handcrafted feature extractor (ex : HOG), do I need to normalize these features befores concatenation?

For example: dimensions of vectores : FeaturesHOG (1*108900) and FeaturesCNN (1*4096)

Can I directly use Features = [FeaturesCNN, FeaturesHOG] to train a multiclass classifier (SVM) or should I follow other steps befor giving these features to the classifier ?

Thank you!!