The goal of this processing stage is to extract six phalangeal regions of interest (ROI) from hand X-ray (distal and middle ROIs of II, III, and IV finger).
Hand Location
The basic step is to determine the location of the hand in the image. First, the image is transformed into a binary mask and simultaneously the image background is removed. Then, high intensity artifacts (labels, markers tubes, etc.) are separated from the hand region.
Due to the background nonuniformity, in most of hand images the background yields a rather slow varying slope making it difficult or even impossible to correctly separate the hand area using a single threshold value. Thus, in order to estimate the background plane, the image is divided into four quarters and a histogram of each area (called later a sub-histogram) is searched for a peak locating the background.
The second stage of the image analysis yields a binary hand mask. After implementing a thresholding function a binary image is obtained. Morphological operators remove isolated pixels . Afterwards a labeling procedure is performed. Its goal is to find the greatest object (i.e. hand area) in binary image.
Axes of II, II,IV fingers
The thresholded image is scanned in the horizontal direction and an array of horizontal profiles is defined.
The fingers area is delimited by two fidutial rows:
- top row containing only one high value level (finger III)
- bottom row only one high value level (metacarpal area)
The central point of each high value level marks an axial point of the corresponding phalanx. These points are subjected to an interpolation procedure in order to obtain axes of II, III, and IV phalanx. Since fingers of most people exhibit a natural curvature, the third-order polynomial fit is used.
Location of epiphyseal regions
A vector P (profile) representing the intensity changes along finger axis. Elements of P are defined as an average of neighbor pixels perpendicular to the local direction of the axis
The first-order derivative of the vector P is searched for three minima which mark roughly the gap between epiphyses and metaphyses. The lowest pixel values in P mark the final position of the gaps.
ROI boundaries: The phalangeal width is estimated by the width of the finger in binary image and distance between gaps is used for setting of the height.
