Model Based Object Recognition be Geometric Hashing By

Model Based Object Recognition be Geometric Hashing By : H. J. Wolfson ניר סגיב : מציג

V = aex + bey + e 00 = a(e 10 – e 00) + b(e 01 – e 00) + e 00 e 01 e 00 e 10 דוגמא Geometric Hashing paradigm

אלגוריתם הלמידה a. Extract the objects interest points. b. For each ordered non-collonear triplet of model points (affine basis) do : 1. Compute the coordinates of all other m-3 model points in the affine coordinate frame defined by the basis triplet. 2. Use each such coordinate (after a proper quantization) as an address (index) to a hashtable, and record in the appropriate entry of the table the pair (model, basis). Algorithm Steps

הזיהוי / אלגוריתם ההשוואה a. Extract the interest points (of scene). b. Choose an arbitrary ordered triplet (affine basis) and compute the other point from the triplet. c. For each such coordinate check the appropriate entry in the hash-table & vote for Nevery pair which apears there. (the votes accumulator will have mi 3 entries – where N is the num of models and mi is the number of interes points on the I’th model) Algorithm Steps

הזיהוי / אלגוריתם ההשוואה d. If a certain pairscores a large number of votes, decide that this pair corresponds to the one chosen in the scene. e. Consider all the model-image point pairs which voted for the affine transformation of step (d) and find the transformation giving the best least-squares match between all these corresponding pairs. Algorithm Steps

הזיהוי / אלגוריתם ההשוואה f. Transform all the edges of the model according to the affine transformation of step (e) and verify them versus the scene edges. If the verification fails go back to step (b) and start over with three other points. Algorithm Steps

אלגוריתם הלמידה Object Acquisition Interest Feature Extraction Transformation Invariant Coordinate System HASH TABLE H(coordinate)=(Object, Basis) Algorithm Steps - Scheme

הזיהוי / אלגוריתם ההשוואה HASH TABLE H(coordinate)=(Object, Basis) Computation of interest feature coordinates in given basis Basis Choice Interesting feature Extraction SCENE Algorithm Steps - Scheme VOTE For (obj, basis) pair (obj, basis) with high vote NO YES BAD Find best least-squares match Verify obj edges against scene GOOD Eliminate obj PROCEED