Identifying the Interesting Points in Geometrical Figures of Certain Class

Identifying the Interesting Points in Geometrical Figures of Certain Class

Stať ve sborníku mimo WoS a Scopus

In this paper we present a novel approach for countinggeometrical objects and finding interesting points of geometrical objects withinan image. Primary attention is given to a certain class of images that containfeasibly visible objects at a considerable scale. The geometric shapes discussedhere are initially limited to squares, rectangles, circles, triangles. Inputs are also takenfrom photographs where geometrical objects are present. Since the segmentationalgorithms by means of edge detection [3] or any similar method provide highlyaccurate results only for gray level images, all the inputs are converted intogray images before subsequent processing. The results are very encouraging andshow the potential usage of this approach in various applications of robotics,geography, statistics etc, where we actually do not require recognizing theobject by its content but by the percentage of area of the objects in the wholeimage. The future scope could be potentially extended to polygons, ovals andcurves. Furthermore, the execution speed of our approach could be improved by adoptionof convenient parallel execution architecture and programming framework, i.e.CUDA (ComputeUnified Device Architecture) on GPU chip from NVIDIA.

In this paper we present a novel approach for countinggeometrical objects and finding interesting points of geometrical objects withinan image. Primary attention is given to a certain class of images that containfeasibly visible objects at a considerable scale. The geometric shapes discussedhere are initially limited to squares, rectangles, circles, triangles. Inputs are also takenfrom photographs where geometrical objects are present. Since the segmentationalgorithms by means of edge detection [3] or any similar method provide highlyaccurate results only for gray level images, all the inputs are converted intogray images before subsequent processing. The results are very encouraging andshow the potential usage of this approach in various applications of robotics,geography, statistics etc, where we actually do not require recognizing theobject by its content but by the percentage of area of the objects in the wholeimage. The future scope could be potentially extended to polygons, ovals andcurves. Furthermore, the execution speed of our approach could be improved by adoptionof convenient parallel execution architecture and programming framework, i.e.CUDA (ComputeUnified Device Architecture) on GPU chip from NVIDIA.

Hierarchal Partitionon Space Representation, Image Segmentation, Prune Length Threshold, GPU, CUDA

Hierarchal Partitionon Space Representation, Image Segmentation, Prune Length Threshold, GPU, CUDA

ŠIMEK, V.

2010

24.09.2008

Faculty of Electrical Engineering and Informatics, University of Technology Košice

Košice

978-80-8086-092-9

Proceedings 8th International Scientific Conference on Computers Science and Engineering

373

382

10