Page 79 -

Chapter 2 ■ Edge-Detection Techniques 53

Table 2.6: Evaluation of Canny VS ISEF: E1
IMAGE EVALUATOR NO NOISE SNR = 6 SNR = 2 SNR = 1
ET1 Canny 0.9651 0.9498 0.5968 0.1708
ISEF 0.9689 0.9285 0.7929 0.7036
ET2 Canny ISEF 0.9650 0.9155 0.6991 0.2530
Canny ISEF 0.9650 0.9338 0.8269 0.7170

ET3 Canny ISEF 0.8776 0.9015 0.7347 0.5238
Canny ISEF 0.8776 0.9015 0.7347 0.5238
ET4 Canny ISEF 0.5157 0.5092 0.3201 0.1103
Canny ISEF 0.4686 0.4787 0.4599 0.4227
ET5 Canny ISEF 0.5024 0.4738 0.3008 0.0955
Canny ISEF 0.4957 0.4831 0.4671 0.4074

Table 2.7: Evaluation Canny VS ISEF: E2
IMAGE EVALUATOR NO NOISE SNR = 6 SNR = 2 SNR = 1

ET1 Canny 1.0000 0.5152 0.5402 0.5687
ISEF 1.0000 0.9182 0.5756 0.5147
ET2 Canny ISEF 1.0000 0.6039 0.5518 0.5726
Canny ISEF 1.0000 0.9462 0.6018 0.5209
ET3 Canny ISEF 0.9291 0.7541 0.6032 0.5899
Canny ISEF 0.9965 0.9424 0.5204 0.4829
ET4 Canny 1.0000 0.7967 0.5396 0.5681
ISEF 1.0000 0.5382 0.5193 0.5096
ET5 Canny ISEF 1.0000 0.5319 0.5269 0.5706
Canny ISEF 0.9900 0.6162 0.5243 0.5123

2.7 Color Edges

So far, only edges created by a change in brightness, as indicated by grey level
value, have been examined. This involves probably 90% of edges of interest
in real problems, but not all of them. It turns out that changes in color, or
hue, are not always detected by the edge detectors described so far. If edges
are the boundaries of objects, then boundaries that are marked by color alone
should be detectable, and because most images involve color it is important to
consider it when looking for edges.
There are two main ways that color edges are located. One method is to apply
one of the edge detectors already discussed to each of the color channels —
red, green, and blue — and then to merge the three results into a single result.
The other method involves multi-dimensional gradients, or partial derivatives.
The former scheme is pretty obvious and easy to implement. The code already
exists to apply any of the methods already discussed to any grey level image,
and a red, green, or blue component image is effectively grey, consisting as it
does of 8-bit pixels.