Good descriptions of this tool can be found in the users guide in the section describing the pairs tool.
Execute this program using tinaTool -r to invoke the stored replay. This will initialise the various parts of the tool for graphic display and load in an example set of models and scene data. The following sequence of button presses will attempt to identify the match outline in the cluttered scene.
This will automatically load the polygonal models as?.poly stored in the local dinosaur directory and will build a data base of corresponding geometric histograms with the parameters specified in the pairs tool.
You can examine the typical appearance of a histogram by selecting the pick (geom) option in the pairs tool and picking a line from the scene displayed in the mono tool using the mouse. Executing histograms will generate the corresponding PGH on the top of the stack in the image calculator (you may need to init' the Tv periodically to refresh the imcalc Tv display with the new data).
You can match individual features of the scene to the object fragments stored in the PGH data base using the match line routine. Select a few lines to see how well the default options appear to work.
You can run an entire scene analysis using the match scene routine. The segment button will now allow you to identified the fragments of the scene associated with the Model Name parameter. Quantify the errors for each model in the data base as1-as5 in a table. The table should contain the number of line fragments in each object which have been correctly and incorrectly labelled.
Now clear the data base and reload the models with a different selection for the histogram type (directed). This form generates a larger histogram representation than previously such that the set of histohrams for each object comprise a complete representation of the shape.
Try to relate what you see in the histogram now to what you saw with the rotate parameter. Re-run the scene matching and evaluate the recognition performance once again.
What is the fractional improvement in labelling reliability? Is this improvement statistically significant?