Classical and neural PCA for image compression - Matlab files

Matlab scripts, functions and image files for the classical and neural network-based PCA aproaches for lossy image compression. Scripts (require Matlab Deep Learning Toolbox and the helper functions given below): - CPCA_compression.m: image compression/decompression procedure using the classical PCA approach, - APCA_compression.m: image compression/decompression procedure using the neural linear autoencoder, - GPCA_compression.m: image compression/decompression procedure using the GHA neural network. Functions: - prep_diff_image.m: preparing the vectorized difference image frame matrix based on the original image matrix, - prep_decompr_image.m: preparing the reconstructed (decompressed) image matrix based on the reconstructed (decompressed) vectorized difference image frame matrix, - calc_error_coeffs.m: calculating different indicators of the image compression error, - gha.m: training the GHA neural network (Matlab function shared by David Gleich in https://www.cs.purdue.edu/homes/dgleich/projects/pca_neural_nets_website/neural-pca-ica.zip). Images: - baboon.tif, - barbara.tif, - lighthouse.tif.

本数据集包含用于有损图像压缩的经典主成分分析（PCA）与基于神经网络的PCA方法的Matlab脚本、函数及图像文件。
脚本需搭配Matlab深度学习工具箱与下述辅助函数使用，具体包括：
- CPCA_compression.m：采用经典PCA方法的图像压缩/解压程序
- APCA_compression.m：采用线性自编码器神经网络的图像压缩/解压程序
- GPCA_compression.m：采用广义赫布算法（Generalized Hebbian Algorithm，GHA）神经网络的图像压缩/解压程序
辅助函数包括：
- prep_diff_image.m：基于原始图像矩阵构建向量化差分图像帧矩阵
- prep_decompr_image.m：基于重构（解压）后的向量化差分图像帧矩阵，生成重构（解压）后的图像矩阵
- calc_error_coeffs.m：计算图像压缩误差的各类评价指标
- gha.m：用于训练GHA神经网络的Matlab函数（由David Gleich共享于https://www.cs.purdue.edu/homes/dgleich/projects/pca_neural_nets_website/neural-pca-ica.zip）
附带测试图像包括：
- baboon.tif
- barbara.tif
- lighthouse.tif