Lung Fields Segmentation Algorithm in Chest Radiography - Advances in Image and Graphics Technologies - page 140

Six feature images were used in this paper: (1-2) x, y direction of the first order
partial derivatives, which means x, y direction gray variation; (3-4) x, y direction of
the second order partial derivatives, represents x, y direction of the gradation variation
rate; (5) x, y direction of the mixed partial derivative; (6) x, y direction with the
second order partial derivatives. The larger value indicates that the gray variation is
fast, and the possibility of being the lung outline is large.

3.2 Similar Cost

Gray Cost. For each point of prior lung outline in test images, the degree of similarity
between the gray vector of all pixels in search area of this point with six feature
images and gray vector of the corresponding point in training images was
computed, then the m points with maximum similarity were selected as the candidates
of boundary. Degree of similarity of the point i can be evaluated by the cosine of
the angle:

∑

(2)

Where is the gray vector of the point i in feature image j, the gray vector can be
described by the set of the gray of the points that located on a circle which centered at
the point i with radius , is the average of the gray vector of the corresponding
point in feature image j, N is the number of feature images, which the value is 6. The
larger value of boundary candidate points in test images indicated that the higher
similarity in gray distribution between this point and the corresponding point in train-
ing images.

Shape Cost. Shape cost of each boundary point was defined as the degree of similari-
ty between shape feature of this point and average shape feature of the corresponding
point in training images. The degree of similarity of the point i can be defined as:

(3)

Where equals represents the shape feature of the point i , and was

the coordinate of boundary point i , represents the average shape feature of the
corresponding point in training images.