Performance of a modified gray level morphological gradient with low sensitivity to threshold values and noise

The performance of a new gray level morphological gradient which uses gradient projections is evaluated on image data sets, and this performance is compared to theoretical expectations. The image data includes well defined edges, less distinct edges, and digitized images which have both distinct edges and regions of texture. Many first level edge detectors rely on threshold values for gradient magnitudes. Sensitivity to these threshold values can have a significant impact on performance, since the selected threshold level is usually a compromise between accepting true edge segments and rejecting edge segments created by texture or noise. This morphological gradient method uses threshold values based on the relative structure of the image content rather than exclusively using gradient magnitudes. The performance is thus less sensitive to the level settings than it would be if gradient magnitude alone was used to accept or reject edge segments for an edge map. In addition, since no smoothing of the image data is used to reduce noise, edge information is preserved. Analysis of the performance on the image data set shows that the expected performance improvements are realized and indicates that this method has broad potential application.