Geometric Model Reconstruction - Computational Hemodynamics: Theory, Modelling and Applications - page 47

$UHD$

$UHD%

&2521$5<
$57(5<

+LVWRJUDPRIRYHUDOOLPDJH

+LVWRJUDPRI$UHD%

+LVWRJUDPRI$UHD$

JUH\VFDOHYDOXH

Fig. 3.3 Histogram analysis for a scanned image of the coronary artery. Local histograms can also
be defined by selecting the local area of interest. The horizontal axis of the histogram represents
the greyscale values that range from 0 ( black ) to 255 ( white ). The vertical axis represents the
number of pixels that contain the greyscale value

A useful method to determine the appropriate θ is to analyze the histograms of
the set of scanned images that are to be segmented. A histogram plots the occurrence
(or number of pixels) of all greyscale intensities (0-255). By selecting sub-areas of
the image for histogram analysis, the correct θ can be found. Figure 3.3 shows a
scanned image of the coronary artery with histogram analysis. An overall histogram
shows a range of peaks as the image contains a collection of different anatomical
structures. By selecting a sub-area namely Area A and Area B, the local histogram
reveals the dominant threshold pixel values that represents the structures within the
selected sub-area. By looking at the histogram for a specific image a viewer can
determine the greyscale distribution at a glance.
When a greyscale range is defined, boundaries are traced for all pixels within
the range in the image. Greyscale threshold works well when an image that has uni-
form regions and contrasting background. Typically selected pixels from a given
threshold criteria will not usually constitute a single connected component. This
would lead to the initial selection being subdivided into smaller sub-selections.
Each separate threshold selection can be treated as a 'mask' which can be used in
Boolean operations. Figure 3.4 shows the result of a single threshold criterion ap-
plied to a scanned coronary artery image. The criterion sets all pixel values great-
er than θ converted to white; while all pixels less than θ are converted to black.
A number of thresholding algorithms exist and the following review papers that