Efficient entropy-based spatial fuzzy c-means method for spectral unmixing of hyperspectral image

This study proposes efficient entropy based spatial fuzzy c-means technique for spectral unmixing of remotely sensed hyperspectral image by adapting local spatial information. The fuzzy membership function has been redefined by incorporating the advantages of both the global and local spatial membership functions. The local spatial membership is estimated to decide the correlation among neighbors, which is playing a responsible role for dramatic reduction of noise and misclassification in the resultant unmixed spectra and segments as well. The intensity-based distances in estimating the membership value has been concisely replaced by Gaussian based distance to reduce the effect of noise. The spatial attributes, entropy and weighted function have been considered as important correlation parameters to optimize the objective function by minimizing the degree of randomness. The estimated results of the proposed method and considered recent fuzzy based methods as Spectral Angle Distance, Partition Coefficient (Vpc), Partition Entropy (V-pe) and qualitative outcomes have been thoroughly compared for hyperspectral and brain MRI datasets. The estimated outcomes have been analyzed and observed that the proposed method have improved in many terms for both the datasets with lesser computational time.