A method for image processing includes determining features of multiple stored images from a pre-trained deep convolutional network. The method also includes clustering each image of the multiple stored images based on the determined features.