Anisotropic Multi-Scale Edge Detection Algorithm

Fig. 1. Comparison of two multi-scale fusion techniques. (a) Test image; (b) multi-scale fusion effect based on geometric average; (c) multi-scale fusion effect based on arithmetic average

Download full size

Fig. 2. Effect of multi-scale fusion

Download full size

Fig. 3. Flow chart of proposed anisotropic multi-scale edge detection algorithm

Download full size

Fig. 4. Sampling results of discrete anisotropic filters: (a) Anisotropic Gaussian filters; (b) anisotropic edge filters

Download full size

Fig. 5. Comparison of ROC curves of different algorithms

Download full size

Fig. 6. Comparison of FOM curves of different algorithms

Download full size

Fig. 7. Edge detection results of different algorithms: (a) Test images; (b) manually marked edge maps ; (c) Canny algorithm^[6];(d) SMED algorithm^[7]; (e) ANDD algorithm^[8]; (f) proposed algorithm

Download full size

Table 1. Comparison of different algorithm models

View table

Table 1. Comparison of different algorithm models

Parameter	Canny^［6］	SMED^［7］	ANDD^［8］	Proposed algorithm
Scale mechanism	Single scale	Multiple scales	Single scale	Multiple scales
Filter	Isotropic filter	Isotropic filter	Anisotropic filter	Anisotropic filter
Fusion strategy	Unused	Geometric average	Unused	Arithmetic average

Table 2. Objective evaluation results of different algorithms on BSDS500 dataset
View table
Table 2. Objective evaluation results of different algorithms on BSDS500 dataset
Algorithm Precision Recall F-score
Canny^［6］ 0.5521 0.6982 0.6166
SMED^［7］ 0.5644 0.7127 0.6299
ANDD^［8］
Proposed algorithm
0.5643
0.5710
0.7436
0.7558
0.6417
0.6505

Table 3. Performance comparison of different algorithms

View table

Table 3. Performance comparison of different algorithms

Algorithm	Precision			Recall			F-score			Noise robustness			Computation complexity
Algorithm	Best	Good	Bad	Best	Good	Bad	Best	Good	Bad	Best	Good	Bad	Best	Good	Bad
Canny^［6］			√			√			√			√	√
SMED^［7］		√			√			√			√			√
ANDD^［8］		√			√			√			√			√
Proposed algorithm	√			√			√			√				√