Linear feature extraction in dot patterns

The problem of object detection is very important in the field of computer vision, as it represents one of the basic operations needed for the eventual interpretation of a given scene. Object detection involves matching of the features derived from images with those known for the object. Intrinsic features of objects, such as spectral, spatial, and topological features are typically extracted from the images. In addition, features associated with the inter object relationship can also be useful for detection.

The main objective of this study is to extend the capabilities of the object detection approach to incorporate and analyze "'inter-object" relationships for better detection accuracy, minimizing false alarm, and a more robust performance. This is approached by considering objects as patterns of dots in an image. Some approaches in dot patterns group dots irrespective of the cluster shape, such as MST, and Voronoi tesselation. Also other approaches such as Fuzzy c-Lines, and Hough Transform are used to detect linear features in the dot patterns. Implementation procedures for the various algorithms are discussed. Experimental comparisons between the performance of different algorithm and their results are also included. A quantitative measure for evaluating the performance of the algorithms that generate prototypes is developed. Fuzzy c-lines and the Hough Transform produced better results in detecting the linear features of the chain like dot patterns. Periodicity of the points along the lines, their density and their distance from the line are also used to describe the lines.