Dual Feature and Decision Fusion in Texture Classification

In multi-texture images, we are using features such as edges, color and texture. Texture is the most important feature to achieve higher accuracy between these features. Inthis paper, a new algorithm is presented for texture classification based on adaptive power spectral density. Otherfeatures extracted by GLCM, GLRLM and purposed filter banks and to achieve a decision in a region each pixel classified by Bayes classifier. First, this algorithm created neural networkper each class, so train for each train sample. Next, Nets update by fisher discrimination popular. The net updates and makes thebest filter per each class. Feature selection processes apply, and the new sets of features use for texture classification. The maximum likelihood classifier (MLC) found similar class. Genetic Algorithms (GA) fused features and decisions. This method costume acceptable time when compared with previous methods. Experimental results show the accuracy and validity of the classification increased up to 88.14%.

Keywords:

Directional filter banks , feature fusion , Adaptive Power Spectral Density