With the fast development of internet innocent over communication in the network environment has become an important research direction. Steganography means that secret information is embedded into cover data imperceptibly for transmission. Linguistic Steganography covers all the techniques that deal with using written natural language to hide the secret message. This paper, presents a linguistic steganography for Arabic language texts, using Kashida and Fast Fourier Transform on the basis of using a new technique entitled Random Singular Value Decomposition Image as a location to hide a secret message. The proposed approach is an attempt to present a transform linguistic steganography using levels for hiding to improve implementation of kashida, and improve the security of the secret message by using Random Singular Value Decomposition Image. The proposed algorithm achieves typical steganography properties such as capacity, security, transparency, and robustness.