&lt;title&gt;Adaptive procedure for threshold selection in directional derivative edge detectors&lt;/title&gt;

Amodaj, Nenad; Popović, M.

doi:10.1117/12.23524

Cited by 4 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Kanijev algoritam takođe ima i mogućnost promene vrednosti parametra σ u cilju postizanja komprimisa između dobre detekcije i dobre lokalizacije. Određivanje optimalne vrednosti praga detekcije u Kanijevom algoritmu deteljno je analizirano u literaturi, a osnovna ideja je da se odredi vrednost parametra σ koja smanjuje devijaciju gradijenta usled šuma na vrednost koja odgovara verovatnoći lažne detekcije (Amodaj & Popović, 1990).…”

Section: Detektori Bazirani Na Gradijentnim Operatorimaunclassified

Poređenje performansi različitih metoda detekcije ivice u digitalnoj slici

Stefanovic¹,

Štrbac-Savić²,

Milić³

2015

Proceedings of the International Scientific Conference - Synthesis 2015

View full text Add to dashboard Cite

Apstrakt: U radu su obrađeni neki od standardno korišćenih algoritama za detekciju ivice u slici, što je jedan od najznačajnijih koraka u interpretaciji sadržaja digitalne slike. Analizirani su gradijentni operatori za izdvajanje ivice bazirani na digitalnim aproksimacijama prvog izvoda dvodimenzionalne funkcije gradijenta slike, a praktična implementacija urađena je u MATLAB programskom okruženju. Ilustrovana je primena Roberts-ovog, Prewitt-ovog, Sobel-ovog i Canny-jevog detektora, a nakon analize i poređenja dobijenih rezultata, predloženi su neki algoritmi povezivanja ivica u cilju formiranja neprekidnih granica regiona, uz upotrebu različitih metoda fitovanja krivih. Negativan uticaj šuma pri izdvajanju ivica delimično je eliminisan primenom prostornog usrednjavanja prilikom formiranja razlike osvetljenosti po određenoj koordinati, pri čemu se prostorno usrednjavanje vrši po ortogonalnoj koordinati. Uticaj parametra koji kontroliše proces usrednjavanja posebno je razmotren u slučaju Canny-jevog detektora. Konkretni primeri koji ilustruju upotrebu različitih vrsta detektora, uz poređenje sa originalnom slikom, rađeni su u MATLAB programskom okruženju, dok su, osim subjektivne vizuelne ocene kvaliteta, određeni i numerički pokazatelji kvaliteta kao što su srednja kvadratna greška (MSE-Mean-Square Error) i vršni odnos signal-šum (PSNR-Peak Signal-to-Noise Ratio).

show abstract

Section: Detektori Bazirani Na Gradijentnim Operatorimaunclassified

Poređenje performansi različitih metoda detekcije ivice u digitalnoj slici

Stefanovic¹,

Štrbac-Savić²,

Milić³

2015

Proceedings of the International Scientific Conference - Synthesis 2015

View full text Add to dashboard Cite

show abstract

“…Unfortunately this requires various heuristically set parameters which diminishes the robustness of the method. Alternatively, Amodaj and Popovic (Amodaj and Popovic 1990) iteratively fit a Rayleigh function to the lower portion of the histogram. However, the fitting may not be robust since it depends on the portion of the histogram used as well as on the initial estimate.…”

Section: Gradient Histogram Analysismentioning

confidence: 99%

Edges: saliency measures and automatic thresholding

Rosin¹

1997

Machine Vision and Applications

View full text Add to dashboard Cite

Edges are useful features for structural image analysis, but the output of standard edge detectors must be thresholded to remove the many spurious edges. This paper describes experiments with both new and old techniques for: 1) automatically determining appropriate edge threshold values, and 2) determining edge saliency (as alternatives to gradient magnitude). keywords: edge threshold saliency assessment which was manually chosen. In contrast, we wish to assign all edge pixels at the finest scale a lifetime value, and their relative values should be insensitive to the coarsest scale. Therefore, when tracking coarse-to-fine we also track edges that are created at all intermediate scales. A slight modification can be made to these two approaches to combine edge lifetime and gradient magnitude. Rather than simply counting the number of scales an edge exists over, the gradient magnitudes at each scale are summed during the tracking process. An alternative method for combining lifetime and gradient magnitude suggested by Griffin et al. (Griffin, Colchester, Robinson, and Hawkes 1992) is the product of the gradient magnitude ∇I with the angle of the isoluminance curve through scale space arctan ∇I ∇ 2 I. However, apart from an initial sharpening effect we have found that this measure is essentially the same as the unmodified gradient after non-maximal suppression. This is not unexpected since ∇ 2 I equals zero close to edges (e.g. Marr and Hildreth's zero-crossing edge detector), and so arctan ∇I ∇ 2 I simply evaluates to Π 2. 2.2 Wiggliness The expectation is that noisy edges are less likely to be locally straight or smooth than significant edges (Rogowitz and Voss 1990). There are several possibilities for measuring edge wiggliness. We have experimented with kregularity (Vasselle and Giraudon 1994) which is defined at each point p i on the curve as: r s,k (i) = p i+ks − p i k j=1 p i+js − p i+(j−1)s and averaged over the curve, and the circular variance of orientation as defined in Gregson (Gregson 1993) or Mardia (Mardia 1972). Other possibilities include orientation coherence (Larré and Montseny 1994), complexity (Dubuc and Zucker 1994), and difference in curvature (Rosin and West 1995). A drawback with such approaches is that they generally require some parameters specifying the scale of analysis, e.g. s and k for k-regularity, and the window size for orientation dispersion.

show abstract

Section: Gradient Histogram Analysismentioning

confidence: 99%