A novel method of performing the task required by an intermediate-level vision (ILV) process was derived. This new technique is based on the Hough Transform, specifically, the generalized version. The novel method solves problems of rotation-invariance and scale-invariance plaguing the generalized Hough Transform while providing a technique that solves the ILV problem and meets the goals of this research.

The use of the Hough Transform, in the case of hypothesis testing, is valid as shown by prior disclosures. When testing the hypothesis of an object in an image, a considerable number of subhypotheses are generated in order to extract a correct region in an image. This makes the problem of region extraction similar to combinatorial optimization. Although region extraction could be considered as a global optimization problem (if there were only a single region in the image), the problem of region extraction itself is not congruent with that of global optimization.

There exist many combinatorial optimization techniques that have been used to solve this problem, including genetic algorithms, simulated annealing, and tabu search. None of these methods guarantee finding the correct or exact solution, only the best possible one (after a number of iterations of the given algorithm.) Therefore, prior investigations have shown the value of these combinatorial optimization techniques should be reexamined, in addition to stating that the Hough Transform is a better search space strategy and ultimately more efficient. Consequently, this reexamination should be done by considering the Hough Transform, its generalized version, its properties, and the overall usefulness as a hypothesis support mechanism.

Unlike the methods previously listed, the novel version of the generalized Hough Transform, called the Pose-Invariant Hough Transform (PIHT), does not require the generation of numerous subhypotheses locating the desired region for extraction. Instead, by using a new version of the R-Table, called the J-Table, rotation-invariance and scale-invariance are built into the table (i.e., hypothesis). The novel PIHT method with its new J-Table is invariant to rotation or scale differences of the desired object in the image. This alleviates the need of generating subhypotheses and eliminates the associated complexities.

Nonetheless, the PIHT (and the generalized Hough Transform) alone does not provide the desired region extraction or contour identification required by the ILV process. What is needed is a scheme using the results of the PIHT and performing the region extraction or indicator-of-match required. Hence, an entirely new technique was developed, called the Inverse-Pose-Invariant Hough Transform (IPIHT), which executes the indicator-of-match.

It should be palpable that the efficiency of the PIHT/IPIHT framework depends on several factors, including quality of segmentation, accuracy of gradient determination, and validity of Hough Transform capabilities. To determine if any or all of these factors affect performance of the framework, it becomes necessary to experiment on individual functions and analyze their performance.