SUBSECTION VARIATION Sample Clauses
SUBSECTION VARIATION. As can be seen from comparing Figures 5.1 and 5.2, the advantages of using discriminate patches for object representation are quite clear. Discriminate patches allow important points in the image (in this case corner points) to be isolated, and the role played by background pixels is reduced. These points are more likely to contain usable appearance information than nondiscriminant patches. Figure 5.2– Example of Discriminate Image Patches (right) Found From ▇▇▇▇▇▇ Corner Points (Section 5.2.4) (left) It is in fact possible, using neural network classifiers, to apply the ▇▇▇▇▇▇ corner point image patches in a reasonably invariant manner as they are. This invariance can be trained into the neural network weights file by resizing and rotating images, then presenting them to the network as correct samples (see Section 6.3). Other techniques for creating invariant image segments without neural network training are given in the following (see Section 5.2.5).