Height and Width in the Image Plane Sample Clauses

Height and Width in the Image Plane. Beyond the very basic statistical measurements described in Section 5.1, the measurement of the height and width of objects in the image plane provides meaningful information, which is less affected by lighting conditions and more accurately represents class types. Unfortunately, as the height and width are measured in the image plane, affine distortion is more prevalent. This distortion, due to perspective, makes the height and width measurements more prone to inaccuracy. This technique is included because it allows a simple postulation to be made as to the likely degree of grouping of object classes, and therefore the prospective usefulness of more invariant statistical descriptors, such as those detailed in Section 5.1.2. In order to ascertain the degree of grouping this type of size or shape based technique may hold, the height and width in the image plane can be plotted as a graph and the clustering observed. From this, a set of group centroids are created, which represent an idealised height to width ratio for a set object type. Centroids are used to reduce the semi­relevant information content of the system, so that only one point is required for matching. This point retains most of the information content of the original clustering, as it is the mean point of the group. This technique also allows for the easy creation of decision boundaries. The clustering information is not completely disregarded, however, as some distance measures use this information to realise better classification results. The Mahalanobis distance measure is a good example of this (see Section 6.1.2).