Texture vector analysis using regular synthetic texture patterns

Image texture is described by different methods as e.g. variance analysis and 2^nd order statistics based on co-occurrence matrices. Statistics based on co-occurrence matrices are considered to provide the most precise descriptors but can be considered operational due to the extreme computing power needed for calculation. Further, statistics calculated from co-occurrence matrices are very sensitive to the size of the moving window describing the center pixel´s texture property. Thus, different texture descriptors can result for one and the same texture pattern by the use of different window sizes. Texture vectors seems to solve these problems. Texture vectors are defined by 5-dimensional vectors as the sum of all frequency dependency vectors in a given window. The first 4 dimensions are defined by normalized frequency of the co-occurrence of grey level combinations. The 5^th dimension is given by the distance considered by calculating the co-occurrence. Processing time is decreased by recoding grey values according to their value. Different regular texture patterns were used to analyze the texture vectors in comparison to common used texture descriptors