Signal processing via COSHAD transform (Q1091101): Difference between revisions

The mth order \(COSHAD_ m(n)\) matrix of size \(2^ n\times 2^ n\) (0\(\leq m\leq n)\) is defined by \(COSHAD_ m(n)=DCT(m)\otimes WHT(n-m)\) where DCT(m) and WHT(n-m) are discrete cosine transform (DCT) and Walsh- Hadamard transform (WHT) matrices of sizes \(2^ m\times 2^ m\) and \(2^{n-m}\times 2^{n-m}\), respectively (\(\otimes\) \(=\) Kronecker product). Clearly \(COSHAD_ m(n)\) belongs to a family of discrete orthogonal transforms ranging from WHT \((m=0)\) to DCT \((m=n)\). It combines the nearly optimal performance of DCT with the computational advantage of WHT at the rate determined by m. Fast algorithms for computing the COSHAD transform and its inverse, based on sparse matrix factorization, are discussed through specific examples of \(m=2\). Its performances are studied in terms of quantitative measures such as variance distribution, Wiener filtering, decorrelation property and transform image coding.