Constraint optimum well-log signal segmentation

A special classification algorithm is proposed that can be applied to a preprocessed (or the original noisy) well-log signal for segmentation. Knowledge of the number of segments or any other constraint, if existent, along with a criterion function can be used to complete the algorithm. The preprocessing routine consists of a running window change-detection algorithm which detects all the potential candidates for the location of changes in the signal. This routing can be applied in way that significantly overestimates the number of changes. These points of change along with other estimated parameters are used by the classification algorithm to find the global best-segmentation that agrees with the a priori knowledge of the number of segments (or any other constraint) and satisfies a criterion function. The resultant optimum classification algorithm is recursive and computationally efficient. The performance of the overall algorithm is demonstrated by several examples