Texture Analysis Using Gaussian Weighted Grey Level Co-Occurrence Probabilities

• Venue: CRV '04 Proceedings of the 1st Canadian Conference on Computer and Robot Vision
• Year: 2004

The discrimination of textures is a significant aspectin segmenting SAR sea ice imagery. Texture features calculatedfrom grey level co-occurring probabilities (GLCP) are wellaccepted and applied in the analysis of many images. Whencalculating GLCPs, each co-occurring pixel pair within the imagewindow is given a uniform weighting. Although a novel technique,co-occurring texture features have a tendency to misclassify anderode texture boundaries due to the large window sizes neededto capture meaningful statistics.A method is proposed whereby co-occurring pixel pairs closerto the center of the image window are assigned larger cooccurringprobabilities according to a Gaussian distribution. Byusing a Gaussian weighting scheme to calculate the GLCPs, lesssignificance is given to pixel pairs that are on the outlying regionsof the window, which have a tendency to produce erroneousstatistics as the image window overlaps a texture boundary. Thismethod proves to preserve the edge strength between texturesand provides better segmentation at the expense of computationalcomplexity.