Simply, PCA is the statistical procedures where new basis of vector space will be found to treat variation of data better. Each point (x,y) in the vector space (let say X-Y) which correspond to each observation will be transformed into another vector space (let say U-V) where point (u,v) are handled easily than in [X-Y] vector space.
In image processing, the principal direction will be identified in such a way that data variation is minimum. Statistically, mean of the given data set will be origin for the new vector space and straight line for which difference of sample data will be minimum is principal direction. Transformed data to new vector space is said to be de-correlated and the new data set is compact representation of the original sampled data.
In hyperspectral imaging, PCA is used for data dimension reduction resulting low bandwidth for data transfer and low memory space for storage. If the data variation is other than some natural process or caused by random experiment error, PCA is better way for data reduction. Thus PCA is another statistical procedures which can be used in image compression.
