RECOGNITION OF COMPLEX TWO-DIMENSIONAL SHAPES USING GEOMETRIC PRIMITIVES AND THE HOUGH TRANSFORM
MetadataShow full item record
The Hough Transform is a widely researched area of computer vision due to its unique promise for shape analysis in digital images. The Hough Transform is a robust technique for detection of geometric primitives in the presence of noise and partial occlusion. The shortcomings of the Hough Transform include its high computational complexity and excessive storage requirements. Every year there are many new techniques proposed as improvements to the Hough Transform or any existing Hough techniques. Complex objects can be decomposed into sub-objects, and sub-objects into further sub-objects. This decomposition can continue until geometric primitives are reached. For object recognition in digital images, lines are the lowest primitive constructed from the digital (pixel) information, and can then be used to construct other geometric primitives and sub-objects. This construction can be used to detect complex objects in digital images. This thesis reviews existing Hough Transforms and explores their strengths and weaknesses. The research work introduces a new Hough Transform that uses a hi- erarchical (pyramid) approach to first detect simple geometric primitives which are then used together to construct more complex shapes for the purpose of object recognition. This research provides a framework for complex object detection that is based on a model of human visual perception. Inside this framework is a discrete Hough transform that uses only discrete and local information for detection of line segments. The next level in the hierarchy uses the line segment information to detect circular and elliptical objects by directly computing all parameters directly from local information.