Digital image watermarking is widely used for copyright protection of digital information. The effectiveness of a digital watermarking technique is indicated by the robustness of embedded watermarks against various attacks. A new method for watermarking is suggested is feature based watermarking. For getting robust watermark, the watermark should be embedded in silent part of the data and for these significant features of data is used. In this paper few methods of feature extraction as Harris Laplacian, Laplacian-of-Gaussian, Susan, Gilles are applied for feature extraction. Robust Non overlapping regions against different attacks are selected for watermarking. Comparison for robust feature selection is done against different feature extraction methods. In next stage those regions are pruned to get minimal primary feature region set using pruning algorithm and watermark is embedded in selected regions and then again results of extracted watermark is compared against different feature selection methods for robustness.