Multi-Scale Blur Estimation and Edge Type Classification for Scene Analysis

• Authors: Wei Zhang;Fredrik Bergholm
• Venue: International Journal of Computer Vision
• Year: 1997

Signatures, in this work, are multi-scalerepresentations of local gray-level information, tied to places ingray scale images where regional differences are locally maximal. Theinformation may involve the regional differences themselves (calledGaussian differences or signed normalized gradient magnitudes, (Korn, 1988)), or, distancerelations between edges (apparent width measurements), or, absence of edges in pulseedge pairs, at coarser scales. Using signatures involves theclassical problem mentioned by Marr and others of relatinginformation across scales. A novel result is that a fruitful way ofdoing this is to build scale paths fromcoarse-to-fine exploiting edge focusing andassociate with pixel positions, along these paths, the threequantities Gaussian differences, apparent width and the binaryinformation absence/presence of edges (in edge-pairs). Such astructure, if used together with proper conditional tests, serves thepurpose of classifying edges with respect to profile-type, and can also be used for measuring global contrast, degree of diffuseness, deblurred line width, and qualitative labels such as diffuse versus sharp. The structure isused simultaneously for labelling tasks and quantitativemeasurements. Theory on apparent widths, absence/presence of edges inpulse edge pairs is developed. For measuring diffuseness and globalcontrast from Gaussian difference signatures a linear least squares approach is suggested. Extensiveexperimental results are presented. Possible applications are inimage segementation, junction analysis, and depth-from-defocus. Forthe purpose of distinguishing between objects and illuminationphenomena, such as diffuse shadow edges, classification of contourswith respect to diffuseness seems useful.