Showing papers on "Bounding overwatch published in 1976"

Non-linear image processing

Abstract: Processing of nuclear medicine images is generally performed by essentially linear methods with the non-negativity condition being applied as the only non-linear process. The various methods used: matrix methods in signal space and Fourier or Hadamard transforms in frequency or sequency space are essentially equivalent. Further improvement in images can be obtained by the use of inherently non-linear methods. The recent development of an approximation to a least-difference method (as opposed to a least-square method) has led to an appreciation of the effects of data bounding and to the development of a more powerful process. Data bounding (modification of statistically improbable data values) is an inherently non-linear method with considerable promise. Strong bounding depending on two-dimensional least-squares fitting yields a reduction of mottling (buttermilk effect) not attainable with linear processes. A pre- bounding process removing very bad points is used to protect the strong bounding process from incorrectly modifying data points due to the weight of an extreme but yet unbounded point as the fitting area approaches it. (auth)