HOUGH-SPACE ASSOCIATIVE PROCESSOR FOR PATTERN RECOGNITION.

摘要

The use of the Hough transform as a feature space is investigated. The Hough space is used directly, along with some new efficient transformations, for the determination of translation and rotation parameters of two-dimensional objects composed of straight-line segments. New methods to implement the required transformations using associative memory architectures are proposed. The Hough space is also used as a feature space for discriminating among various objects and these techniques are extended to curved objects and objects of arbitrary shapes. The proposed technique for detecting curves and target trajectories involves thresholding the Hough space and performing transformations and an inverse Hough transform. Peaks in the inverse space provide curve identification and determination of the parameters of the curve. Finally, an extension of the straight-line Hough transform to three-dimensional spaces is defined. This transform is capable of detecting planes in an input range image. This new 3-D Hough space can also be used as a feature space for discriminating among 3-D objects and for determining location and orientation of 3-D objects. The technique is very robust and efficient, since it uses range images directly, with no preprocessing such as edge detection and segmentation. It is shown that, in both the 2-D and 3-D Hough spaces, the effects due to the translation and rotation of the input object can be easily separated and estimated, so that an efficient hierarchical search for these parameters can be performed.