A novel hybrid fuzzy-SVM image steganographic model

This paper reviews the current soft computing (SC) techniques employed in image steganography as well as proposes a new hybrid approach of these SC techniques to exploit their complementary strengths. Four main SC techniques in image steganography - neural network (NN), genetic algorithm (GA), support vector machines (SVM) and fuzzy logic (FL) are assessed based on the three main measurements of steganography - imperceptibility, payload capacity and robustness. Most NN usage focuses on robustness, as well as the imperceptibility of the coverimage by exploiting its learning capability to produce a higher quality stego-image. GA is mostly employed to increase the payload capacity to be embedded as well as to find the best bit positions for embedding position in image steganography. SVM is normally used to increase the imperceptibility of the stego image via its strength to classify the image blocks by learning the relationship between the secret-message and cover-image to be used in the embedding and extracting procedures. A few works have been done in FL especially in preserving the imperceptibility and the number is increasing. Based on this review and leveraging the complementary strengths of FCM in clustering and SVM in classification, we propose a novel hybrid fuzzy c-means (FCM) and SVM (F-SVM) model in image steganography. The model uses the F-SVM as its main engine that is capable of embedding the secret-message imperceptible to human eyes while increasing the payload capacity. Currently, work is being done to develop and test this model.