During last few years for copyright protection, security and data authentication of digital media, digital watermarking has been raised as the one of the burning research topics due to the rapid expansion of Internet. This paper presents a new technique for image watermarking in the spatial domain where the concept of information theory is utilized with the popular LSB substation technique. Here, the cover image is segregated into a number of blocks and the watermark is embedded into the block(s) with the maximum entropy value. The extraction algorithm is also able to find the watermark correctly. The proposed algorithm was evaluated with the help of various standard performance measures like Mean Square Error (MSE), Peak Signal to Noise Ratio (PSNR) and Structural Similarity (SSIM) measure to verify the perceptibility and the robustness of the algorithm. Experimental resultsdemonstrate that the improved algorithm performs reasonably well over a large varied datasets of cover and watermark images.