Compressing image-based relighting data using eigenanalysis and wavelets

In image-based relighting (IBL) a tremendous number of reference images are needed to synthesise a high-quality novel image. This collection of reference images is referred as an IBL data set. An effective compression method for IBL data makes the IBL technique more practical. Within an IBL data set, there is a strong correlation among different reference images. In conventional eigen-based image compression methods, the principal component analysis (PCA) process is used for exploiting the correlation within a single image. Such an approach is not suitable for handling IBL data. The authors present an eigenimage-based method for compressing IBL data. The method exploits the correlation among reference images. Since there is a huge number of images and pixel values, the cascade recursive least square (CRLS) network based PCA is used to extract eigenimages. Afterwards, the wavelet approach is used for compressing those eigenimages. Simulation results demonstrate that this approach is much superior to that of compressing each reference image with JPEG and JPEG2000.