doi:10.1016/S0167-8655(99)00138-5
Copyright © 2000 Elsevier Science B.V. All rights reserved.
A Hough transform technique for the detection of reflectional symmetry and skew-symmetry
Raymond K. K. Yip
, 
Department of Electronic Engineering, City University of Hong Kong, 83 Tat Chee Avenue, Kowloon Tong, Hong Kong
Received 27 August 1998;
Revised 9 July 1999.
Available online 14 February 2000.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
In this paper, a simple and yet robust Hough transform algorithm is proposed to detect and analyze reflectional symmetry and skew-symmetry (reflectional symmetry under parallel projection). It is applicable to shapes that contain global, local and slightly deformed reflectional/skew-symmetries under the presence of noise and occlusion.
Author Keywords: Author Keywords: Reflectional symmetry; Skew-symmetry; Skew-symmetry axis; Skew-transverse angle; Hough transformation
Fig. 1. (a) Properties of reflectional symmetry. (b) Under parallel projection, the reflectional symmetry of (a) becomes skew-symmetry.
Fig. 2. (a) A step-by-step example. (b) Detected result of (a).
Fig. 3. Input image and results of experiment 1: (a) Input image, (b) Edge detected image, (c) Midpoints of φ=130°, (d) Voted skew-symmetry axis (first pass) showing the four peak position with (ρs,θs)=(34, 126),(190, 60),(115, 100) and (101, 347), (e) Voted skew-transverse angle φ (second pass) for (ρs,θs)=(34, 126). Peak position indicates φt=150°, (f) Voted skew-transverse angle φ (second pass) for (ρs,θs)=(115, 100). Peak position indicates φt=77° and (g) Symmetry-axis found by the proposed algorithm.
Fig. 4. Input image and results of experiment 2: (a) Input image, (b) Edge detected image, (c) Midpoints of φ=30°, (d) Voted skew-symmetry axis (first pass) showing the peak position with (ρs,θs)= (164, 36), (e) Voted skew-transverse angle φ (second pass) for (ρs,θs)= (164, 36). Peak position indicates φt=60°, (f) Skew-symmetry axis (ρs,θs)= (21, 322) and φt=161° found by the proposed algorithm.
Fig. 5. Input image and results of experiment 3: (a) Input image, (b) Edge detected image, (c) Midpoints of φ=161°, (d) Voted skew-symmetry axis (first pass) showing the first three peak position with (ρs,θs)= (97, 107), (21, 322) and (41, 328), (e) Voted skew-transverse angle φ (second pass) for (ρs,θs)= (97,107). Peak position indicates φt=107°, (f) Recovered skew-symmetric points of (ρs,θs)= (97, 107) and φt=107°, T=1.0, (g) Recovered skew-symmetric points of (ρs,θs)= (21, 322) and φt=161°, T=1.0 and (h) Recovered skew-symmetric points of (ρs,θs)= (41, 338) and φt=164°, T=1.0.
Fig. 6. Input image and results of experiment 4: (a) Input image, (b) Skewed input image, (c) Edge detected image, (d) Voted skew-symmetry axis (first pass) showing the first two peak position with (ρs,θs)= (34, 125) and (25, 128), (e) Voted skew-transverse angle φ (second pass) for (ρs,θs)= (34,125). Peak position indicates φt=152°, (f) Detected results, (g) Recovered skew-symmetric points of (ρs,θs)= (34, 125) and φt=152°, T=1.0 and (h) Recovered skew-symmetric points of (ρs,θs)= (25, 128) and φt=152°, T=1.0.
Fig. 7. Input image and results of experiment 5: (a) Input image, (b) Edge detected image, (c) Midpoints of φ=137°, (d) Voted skew-symmetry axis (first pass) showing the first three peak position with (ρs,θs)= (68, 117), (83, 113) and (96, 108), (e) Detected results, (f) Recovered skew-symmetric points of (ρs,θs)= (68, 117) and φt=134°, T=1.0 and (g) Recovered skew-symmetric points of (ρs,θs)= (83, 113) and φt=152°, T=1.0 and (h) Recovered skew-symmetric points of (ρs,θs)= (96, 108) and φt=121°, T=1.0.
Table 1. Results of experiment 1

Table 2. Results of experiment 2

Table 3. Results of experiment 3

Table 4. Results of experiment 4

Table 5. Results of experiment 5
