Implementation

The system takes a native image of any file format as input, and if required converts it to gray scale for processing.

A global threshold value is computed using Otsu’s method, which chooses the threshold to minimize the interclass variance of the thresholded black and white pixels. Also, minute impurities which may affect processing are removed using morphological operator. After labeling the connected components, their region properties need to be measured so as to accurately define the objects and extract them to individual image files. These extracted chromosome files may also contain certain images with large unwanted artifacts or conjoined chromosomes. These may be either deleted or re-segmented using manual intervention, depending on user choice. Next, boundary thresholding is performed which takes into consideration the use of morphological operation, by setting a pixel to zero if its 4-connected neighbors are all ones, thus leaving only boundary pixels. Skeletonization is yet another morphological operation which removes pixels on the boundaries of objects without allowing objects to break apart.  These steps are required in order to facilitate proper chromosome rotation. As mentioned before, Hough Transform computes the edge linkages of the images. The angle of the edge determines the amount of rotation required for vertical chromosome alignment. As a result of bi-cubic interpolation due to rotation, pixel values are produced which lie outside the original range. Removal of such pixels is obligatory which may otherwise interfere with further image processing techniques.  Lengthwise arrangement of chromosomes is also done in the meanwhile, which may ease the process of ordered karyotyping.

There may be disputes where the chromosome after being vertically aligned needs to be flipped upside down. This facility is provided to the users for pointing out the particular chromosome. After complete pre-processing of individual chromosomes, it is essential to merge them together in a single karyotype image. Image cleaning is performed to eliminate background pixels due to presence of residual artifacts or interpolated pixels. For attaining a clear output, the image is traced with green boundary pixels around the chromosome outline. It has been identified that presence of green pixel gives a truth value, 255, in the g-component of the image already decomposed into r-, g- and b-components. Hence, pixels lying outside green boundaries of the chromosome outline can be excluded. Band pattern matching can be accomplished using Laplace filtering by matching the band components of chromosomes which are represented in the form of hills and valleys in gray values. The user interface also offers a facility to delete any residual artifacts, retain the deleted objects as well as flip the chromosomes upside down. The total number of objects at any point of time is always indicated at the top of the screen.