In this work all we have centred all efforts on the segmentation of mammographic masses. However, a not easy initial step is breast localization. We can distinguish between works locating the boundary between the background and the breast and works looking for the boundary of the pectoral muscle and the breast (in MLO mammograms).
Nowadays, with the introduction of digital imaging, the background is homogeneous and is really easy to find the breast skin-line. However, for digitized mammograms, this is not an easy task because there is an important amount of non-homogeneous noise in the background (and usually some annotations). Different approaches to this task can be found in literature [27,49,69,100,,131,159,192,].
On the other hand, a still more complicated problem is to correctly localize the boundary between the breast and the pectoral muscle. In fact, there are some images where is easy to find it because the grey-level intensity of the pectoral muscle is highly greater compared to the grey-level of the histogram. However, this is not always the case, and in some images only expert radiologists are able to find in. Not too many automatic approaches are found in the literature [49,50,84,97]. See the work of Raba et al. [150] for a more extensive survey of both segmentation types.
In what follows we present two new approaches for breast localization. The first one is a rough approach to segment the breast: only those pixels visually perceptible are segmented, ignoring the pixels of the skin-line boundary. Subsequently, using a region-growing approach, the pectoral muscle is also removed. On the other hand, the second approach performs a better estimation of the skin-line boundary of the breast. It uses scale-space concepts in order to enhance the real boundary.