How to add hand-crafted features as an input to a CNN model in MATLAB

I have extracted hand-crafted features for a face image dataset and now I want to feed these feature vectors to a CNN model, so as to improve my image retrieval accuracy. Please guide me in this regard. Thank you.

[Abhijit Bhattacharjee]

Are you trying to use the hand-crafted features instead of images or in addition to images as inputs to the CNN?

[Aasim]

Yes, I want to use them in addition to images as input to the CNN, to improve my retrieval rate.

[Aasim]

Currently I'm try to do so through autoencoders, but struggling with that. Can you assist?

[Hiro]

Do you have them as image data or in another form?

[Aasim]

I have in the form of 1-dimensional Array of type double.

[Abolfazl Chaman Motlagh]

do you have specific network architecture for your problem? because generally adding this extra feature to network has a lot of ambiguities. for example in which level of information (layer) you want to input these data to network?

[Aasim]

I am using a pre-trained model Vgg-face

Answer 1

[David Willingham]

Hi,

This example shows how to train a model using image and feature data, Train Network on Image and Feature Data.

Is this what you were looking for?

[Aasim]

This is helpful though, but I am using a pre-trained network vgg-face and I want feature vectors only to input the network.

[David Willingham]

Are the features you're inputting the same as those the network was trained with? Or you want to add them in?

[Aasim]

Not same. I have extracted those features from a hand-crafted technique and now I want to feed them as an input to the neural network so that I can use the flatenned output of the fully connected layer to retreive images.

[Aasim]

Can you suggest me best tool to reconstruct images from the feature vectors?

[David Willingham]

Has the network already been trained? If it requires re-training, why does the example above not suffice?