Signal segmentation based on direct use of statistical moments and Renyi entropy

Signal segmentation based on statistical moments is considered in this paper. This method is called direct method, in opposition with methods based on transformation of the signal, which are referred as being indirect methods. The objective of the paper is to evaluate the potential of statistical moments for signal segmentation and to obtain a reference for other methods. The paper uses heuristic algorithms for the change detection and suggests a performance criterion based on the detection and false alarm probabilities. Results are satisfactory but could be used as references in the evaluations of other more complex methods, e.g. those based on the time-frequency transforms. All manuscripts must be in English, printed in black ink.