next >

Three Step Search



Two variations of the Three Step Search (TSS) were implemented and tested. The first (TSS-D) decremented the step size at each iteration whereas the second (TSS-H) halved the step size after each iteration.

The number of matching criteria evaluated by both variations is illustrated in the graph immediately below. The TSS-H variation of the Three Step Search exhibited behaviour very similar to the Two Dimensional Logarithmic (TDL) search. This is not surprising since their methods of operation are very similar. The number of matching criteria evaluated was dependent on the displacement and not the sequence. The quality of the image sequences coded by the TSS at various displacements is illustrated in Figure 5.11. Beyond the operative range for each sequence the MSE began to increase, with the most significant rises in the Garden sequence.


graphic

Number of criteria evaluations by Three Step Search as the maximum displacement is increased.



The number of matching criteria required by the TSS-D variation increased linearly, which was as expected since each pixel increase in displacement usually required an additional 8 positions to be tested. The TSS-D produced images of better quality than the TSS-H variation, but required increasingly more criteria evaluations as the displacement increased.
The behaviour of the TSS on the Garden sequences can be attributed to the same causes as the behaviour of the TDL. The operative range appears to be ±5 pixels. The MSE was at a minimum at ±9 pixels for the Tennis and Football sequences.


graphic

MSE of reconstructed sequences coded using Three Step Search as the maximum displacement increased


[return to Experiments and Conclusions]


© Colin E. Manning 1996