A method for the extraction of an object from a noisy background

The removal of the background from a noisy image to leave the object of interest, with well defined edges, is a difficult task. Images obtained using radioisotopes have low signal-to-noise ratios due to scattered photons, as well as emitted radiation from the background. Moreover, noise is, also, a consequence of the low number of detected photons. A scan of a row or column of such an image demonstrates that the pixels do not change value rapidly near the edge of the object. In addition, clusters of pixels, which can be found outside the edges of the object and in the background, can have higher values than those in the object. This effect is worse in abnormal images, where the abnormal clusters may have values as low as those of the noisy clusters. An image processing technique has been developed to determine the object edges. It is part of a research project in the field of the computer aided diagnosis, which involves artificial neural networks. It makes use of the so-called 'moving window operations' which are very common in digital image processing. The technique was tested using male lung images that were taken by a gamma camera.