Ambient Diagnostics - Ambient Intelligence for Scientific Discovery - page 234

Fig. 5. Cracks detected by the algorithm.

The previous result image is then masked with the inverse of the tongue-edge
mask to retain only data that does not appear on the edge of the tongue.
After all objects in the image are shrunken with erosion, the pixel groupings
are separated out into particles so that they can be analyzed individually. This
is done by scanning the image and grouping pixels that are connected to each
other. There are two typical definitions of pixels being “connected” to another.
According to the 4-way definition, a pixel is connected to a group if the group
appears to the right, left, bottom or top of the pixel. Under the 8-way definition,
a pixel is connected to a group if any of the bordering pixels are in the group.
Once all pixel groupings are determined, then only the largest pixel groupings
are chosen. This is done simply by comparing the area of the group to a threshold
value. The area of a group is the number of pixels that form it. The threshold
value is calculated by taking the mean plus one standard deviation of particle
area values. This is done to keep only “large” particles and exclude smaller
particles that result from noise in the image. Figure 5 shows the finished crack-
detected image.
In order to summarize the crack information of an image, we have developed
a simple numerical descriptor to describe cracks called “Crack Index”. Crack
Index (CI), is a number between 0 and 100 that describes the content of cracks
in the image.

A cracks
A total

CI = 100 ∗ (8)

A CI of 0 indicates no cracks were found, whereas increasing values indicate
an increasing density of cracks. Crack Index is calculated by taking the area of
cracks divided by the area of the total tongue and multiplying by 100.

5.4 Describers for Texture Homogeneous and Complexity

To give further descriptions of texture details such as homogeneous and com-
plexity, we use energy and entropy functions that are based on the co-occurrence
matrix. It describes the repeated occurrence of some gray-level configuration in
the texture [36]. P is a 2-D n x n co-occurrence matrix, where n is the num-
ber of gray-levels within an image. The matrix acts as an accumulator so that
P [ i, j ] counts the number of pixel pairs having the intensities i and j .Theidea