Performance bound of Walsh-Hadamard transform for feature selection and compression and some related fast algorithms

A fast algorithm of finding the diagonal elements of the covariance matrix of the two-dimensional Walsh-Hadamard (WH) transform of data is described. Its usefulness and other interesting properties of the WH transform are discussed. The performance of the WH transform is compared with the Karhunen-Loeve transform for a first-order stationary Markov process.