Learning and Design of Principal Curves
IEEE Transactions on Pattern Analysis and Machine Intelligence
A Skeletonizing Reconfigurable Self-Organizing Model: Validation Through Text Recognition
Neural Processing Letters
| Hi-index | 0.00 |
We present a method for determining the skeletalshape description for letters in texts faded due to ageingand/or poor ink quality.The proposed algorithmis interesting in that it neither involves assumptionsabout demarcation of object regions from the background,nor does it require pixel connectivity in the textregions.Consequently, it may be applied for obtainingthe shape descriptions of "sparse" regions, which arecharacteristic of letters in faded documents.Given thepixel distribution for a letter or a word from a fadeddocument, the method involves an iterative evolutionof a piecewise-linear approximation of the principalcurve of this pixel distribution.By constraining theprincipal curve to lie on the edges of the Delaunaytraingulation of the shape distribution, the adjacencyrelationships between regions in the shape can be detectedand used in evolving the skeleton.The approximationof the principal curve, on convergence, givesthe final skeletal shape.The skeletonization is invariantto Euclidean transformations and is adaptive interms of the topology of the underlying shape distributionas well as in the number of units needed for thepiece-wise approximation of the principal curve.