Some Analog Linear Filtering Tasks Useful in Medical Image Processing

In this paper some analog linear filtering tasks and their possible applications in medical image processing are approached. Although the presented processing tasks can be also implemented digitally, we treated their realization using cellular neural networks (CNNs). We discuss here the implementation of a 2-D FIR Gaussian filter and a class of IIR maximally-flat low-pass filters. Finally, simulation results on some real medical images are shown. One of the main issues regarding medical image processing is image enhancement, employing linear or nonlinear filtering techniques. For applications requiring image enhancement in real-time, fast digital processors are used. When a fast pre-processing is necessary, a noteworthy alternative might be an analog, parallel array processing using for instance cellular neural networks (CNNs), which are large, regular arrays of dynamic processing elements (cells) connected together, and their behavior can be controlled by the template parameters, adjusted digitally [3]-[5]. This parallel processing leads to a large equivalent computing power. Several applications of CNNs in biomedical imaging have been proposed [8]. These systems can perform very efficiently linear image filtering with small-size kernels implemented on templates. For separable kernels, a higher order filtering can be implemented as a sequence of simple filtering tasks. For instance, a currently-used linear filtering is with a Gaussian kernel of different sizes. The original grayscale image and its blurred version obtained through low-pass filtering are subtracted, obtaining a weighted difference. In the output image the high-frequency components are amplified, which leads to an overall enhancement of the image, improving its perception.