Image Processing Reference
images; (2) mammogram preprocessing, to remove the label and enhance the image; (3) mass
localization: for determining the regions of interest and mass location parameters; and (4)
mass accurate segmentation.
FIGURE 1 Flow chart of mass segmentation methodology.
2.1 Mammogram Database
The mammograms used in this work are taken from digital database for screening mammo-
bases are widely used in studies on mammogram analysis, because they are freely available
and consist of plenty cases. For DDSM database, there are as many as 2620 cases and each
image contains about 3000 × 5000 pixels with 16-bit or 12-bit gray level, we map the intens-
ity value into the range [0-255] by using brightness adjustment method. The full raw mam-
mograms provided in this database are with a format of LJPEG which is hard to read under
Windows, thus we first convert it to the usable PNG format by a DDSM-software proposed by
experts. The MIAS database contains 322 mammograms and each image is 1024 × 1024 pixels
with 8-bit gray level, the abnormality is given by a circle. These two databases also ofered
some other corresponding information such as type, severity, character, and so on.
2.2 Mammogram Preprocessing
2.2.1 Label removal
The mammogram image usually includes of breast region, pectoral muscle, background and
label. In order to reduce the processing time and further study, the label should be removed.
Here, we use the standard local threshold method which has been proved to be a convenient
method to remove the label.
2.2.2 Image enhancement
plied. Traditional gray level transformation is not working well since the gray values between
mass region, pectoral muscle, and gland are similar. RS theory that is used in reasoning from
imprecise data is applied in information processing and artificial intelligence widely [ 16 , 17 ] .
In this part, we choose image gradient atribute C 1 and noise atribute C 2 as the condition at-