Feature extraction is a key component of a Monocular Simultaneous Localization and Mapping (Monocular SLAM) system which permits to extract features and can also reliably track them over frames. In this paper, a novel approach for Monocular SLAM is proposed. This approach uses the information on the camera displacement and image saliency to adequately extract stable and suitable features, ones that will sufficiently produce parallax over a limited number of frames and improve the precision as well as robustness of the localization and mapping. According to the results obtained from real data, the proposed method outclasses the state of the art method both in the precision of the estimation and the number of mapped landmarks.