In the dissertation, use of multivalued image coding is proposed for detecting geometrical features such as line slope of polygonal object and for skeletonizing go elongated objects such as fingerprints and handwritten characters.
Based on the fact that the multivalued image contains the local topological properties of the original binary image, a algorithm is proposed that extracts features by simple summation of weighted values of pixels in the multivalued image.
With the analysis of the relationships between multivalued images of adjacent pixels, a efficient and systematic parallel thinning algorithm is also presented which is found to be superior to the existing methods. The extraction of features of skeleton is also accomplished by analyzing the characteristics of multivalued image.
Owing to the fact that multivalued image represent local pattern of binary image as binary number, the proposed algorithms can be easily implemented by digital hardware, and its performance is experimentally shown to be very efficient in comparison with those of other thinning methods.
