A novel approach to incorporate local information in Possibilistic c-Means algorithm for an optical remote sensing imagery

Abstract

In this research article novel Possibilistic c-Means with Possibilitic Local Information c-Means (PCM with PLICM) and Possibilistic c-Means with Adaptive Possibilitic Local Information c-Means (PCM with ADPLICM) algorithms have been presented in a supervised way. Using the local convolution these algorithms are efficient to resolve the limitations of PCM. These algorithms are tested on a multispectral remote sensing data with the added the salt and pepper noise. The Mean Membership Difference (MMD), Root Mean Square Error (RMSE) and Fuzzy Error Matrix (FERM) techniques were used to examine the performance of proposed algorithms. MMD has been calculated on original classified image for independent accuracy in which PCM with ADPLICM shows higher MMD by 7% and 17% in comparison to PCM with PLICM and PCM respectively. The original and noisy images have been used to compute the RMSE. The combination of PCM with ADPLICM produces least RMSE. FERM overall accuracy has also been increased up to 15% after incorporating spatial local information.