Singular Value Decomposition for Compressive Decomposition of 3-Channel Color Images

This program employs singular value decomposition to compress and decompose 3-channel color images through the following implementation steps:

Compression Process:

1. Select sub-image size parameter K to decompose the image into M×M sub-images (IMG(s), where s=1,2,...,M², M=N/K, with original image dimensions N×N). The algorithm partitions each color channel (R,G,B) separately using block processing techniques.

2. Compute the average value of all M² sub-images, then subtract this mean image from each sub-image to obtain zero-mean normalized images. This preprocessing step enhances numerical stability during matrix operations.

3. Calculate the correlation matrix R, where elements are defined by covariance relationships between normalized sub-images. The implementation constructs R using vectorized sub-image data for computational efficiency.

4. Compute eigenvalues and eigenvectors of matrix R through spectral decomposition. The compression encodes each sub-image by computing dot products with the principal eigenvector (corresponding to largest eigenvalue), achieving dimension reduction based on maximum variance preservation.

In this implementation, we apply SVD-based methodology for color image compression. The algorithm begins by selecting sub-image block size K and decomposing the original image into M×M sub-images through systematic grid partitioning. Following mean normalization, we construct the correlation matrix and perform eigenanalysis to identify dominant data variations. The final encoding step projects each sub-image onto the principal component space, effectively compressing image data while preserving critical visual information.

The program's objective is to reduce data volume in color images through compressive decomposition. By optimizing sub-image size selection and leveraging eigenvalue analysis of correlation matrices, we achieve efficient image encoding that facilitates both high compression ratios and quality reconstruction. The implementation handles three color channels independently while maintaining spatial relationships through consistent block processing.

This explanation clarifies the program's operational principles and algorithmic workflow for technical implementation.