The total variation (TV) regularization has phenomenally boosted various variational models for image processing tasks. We propose combining the backward diffusion process in the earlier literature of image enhancement with the TV regularization and show that the resulting enhanced TV minimization model is particularly effective for reducing the loss of contrast, which is often encountered by models using the TV regularization. We establish stable reconstruction guarantees for the enhanced TV model from noisy subsampled measurements; non-adaptive linear measurements and variable-density sampled Fourier measurements are considered. In particular, under some weaker restricted isometry property conditions, the enhanced TV minimization model is shown to have tighter reconstruction error bounds than various TV-based models for the scenario where the level of noise is significant and the amount of measurements is limited. The advantages of the enhanced TV model are also numerically validated by preliminary experiments on the reconstruction of some synthetic, natural, and medical images.