To remove ambiguity in the information presented in the image, it is desirable to reduce its complexity to just the dominant features such as edges and homogeneous regions. The conventional techniques of image processing are an obvious starting point for such an operation and the paper assesses their effectiveness when applied to images from sensors with apertures as small as 80 lambda , which have just a small number of pixels per feature. The authors discuss the spatial frequency spectrum of a typical section of the image and how this relates to the low resolution of the sensor. A gradient (Sobel) operator is then demonstrated as a means of trying to reduce the image to areas of change in pixel intensity and making a first step towards a parametric representation of features within the image. A comparison is then made of the frequency spectra of two filters and the image to account for the filters' performance. They present an approach to segmentation of the image based on regional statistics and illustrate how an adaptive control of statistical parameters can result in a satisfactory segmentation. This technique also has the advantage of automatically providing a parametric representation of segmented regions and the statistical representation is readily extendable to multidimensional image data.
展开▼
