Region-selective sharpening of magnetic resonance images

In many applications requiring a careful inspection of nuclear magnetic resonance images, preprocessing techniques can be used to enhance fine structures and expand the contrast. Nevertheless, due to the large nonstationarity of the bidimensional signal it is unlikely that a single method will perform well on the complete image. Since it is frequently observed that elementary image processing can highlight specific features, the authors propose a data reduction technique corresponding to the optimal combination of different processing results into a composite image which benefits from all elementary information. An adaptive implementation of this approach, based on the subdivision of the original images into rectangular surfaces, is described. The method is used to generate images with enhanced contrast and less noise.<>