Image denoising is a fundamental problem in image processing that removes noise while preserving important structural details. The Deep Image Prior (DIP) approach leverages the implicit regularization properties of convolutional neural networks (CNNs) by using a randomly initialized CNN to reconstruct an image. However, the method is prone to overfitting the noise when the iteration count is not properly managed. The proposed Split Bregman DIP (SB-DIP) technique enhances the denoising process by combining the strengths of DIP implicit regularization with the efficient optimization capabilities of Split Bregman. In this work, we investigate the Deep Image Prior (DIP) framework and propose to improve it by combining an automatic estimate of local regularization parameters with a space-variant Total Variation regularizer. Our experiments demonstrate that SB-DIP effectively suppresses noise while maintaining sharp edges and fine details, outperforming traditional DIP and other state-of-the-art denoising methods on benchmark datasets. The proposed method offers a promising direction for robust and efficient image denoising. Unlike previous methods, we use the flexible Split Bregman (SB) Technique to solve the ensuing minimization problem. The effectiveness of the proposed approach is demonstrated by its promising PSNR values and is validated through multiple experiments on both simulated and real-world natural and medical corrupted images.