Coverless steganographic techniques are considered as reliable solutions for avoiding steganalysis attacks. However, it demands massive shared databases and usually has low payload capacity, which makes it less appealing. An advanced DT-CWT based image steganographic approach has been presented to embed secret data over appropriate coefficient planes of the cover image. Payload capacity is boosted while reducing the embedding error using super-pixeling and intensity mapping in the preprocessing stage of the secret image. A template matching based embedding location detection is used to reduce the embedding error by making use of the similarity between secret data and DT-CWT planes. A machine learning classifier is employed for selecting the best cover-coefficient planes. Cover and stego-image difference is minimized and hence the data is retrieved from the cover image with the secret key generated during the embedding process using either the pre-shared cover image database or without it. When sharing the database, only the secret key is needed to be transmitted which is used for retrieving the concealed data from the original cover image. An automatic geometric correction stage is also proposed to defend against geometric attacks. Experimental results of the proposed approach show better performance among the state-of-the-art techniques.