Evolution: The Simple Polygon
Normalized Moment Comparison:
Distortion Type: |
Correct Classifications |
% Correct |
Non-uniform I |
7 |
23% |
Non-uniform II |
9 |
30% |
Dilation |
14 |
47% |
Light Uniform Noise |
28 |
93% |
Heavy Uniform Noise |
26 |
87% |
The table above shows the results of normalized moment comparison for the 30 images used.
The results of the recognition experiment are summarized on the following graph. It shows that the overall raw performance of using tangent space comparison at the 6th evolution stage was almost equivalent to the performance of moments.
Discrete Contour Evolution, Using Tangent Space Comparison as a Similarity Measure
Distortion Type: |
Correct Classifications/evolution stage |
% Correct |
Non-uniform I |
16 (on 9th) |
53% |
Non-uniform II |
17 (on 4th) |
56% |
Dilation |
22 (on 6th) |
73% |
Light Uniform Noise |
23 (on 5th) |
76% |
Heavy Uniform Noise |
13 (on 4) |
53% |
The table above shows the results of performance of the algorithm on different types of noise. The best rates of classification achieved on different stages of evolution are shown. Note that evolution produced significantly better results than moments did on three of the subsets, in spite of overall lower performance.