Hands-On Machine Learning with C++

Data compression with PCA

We can use dimensionality reduction algorithms for a slightly different task – data compression with information loss. This can be easily demonstrated when applying the PCA algorithm to images. Let's implement PCA from scratch with the Dlib library using SVD decomposition. We can't use an existing implementation because it performs normalization in a way we can't fully control.

First, we need to load an image and transform it into matrix form:

 void PCACompression(const std::string& image_file, long target_dim) {     array2d<Dlib::rgb_pixel> img;     load_image(img, image_file);      array2d<unsigned char> img_gray;     assign_image(img_gray, img);     save_png(img_gray, "original.png");      array2d<DataType> tmp; assign_image(tmp, img_gray); ...