Blind image deconvolution aims to estimate both a blur kernel and a sharp image from a blurry observation. It is not only a classical problem in image processing, but also serves as preprocessing in many advanced tasks including affective image content analysis. In terms of statistical inference, this problem can be viewed as maximizing the probability of latent image and kernel, given the observed blurry image. Proper formulation of latent image prior is crucial to the success of blind deconvolution methods. A novel latent image prior is proposed to penalize low contrast and dense gradients, thus playing the role of enhancing significant segments. Our latent image prior is based on a one-dimensional regularizer, which involves normalizing reciprocals of absolute differences between two neighbouring unequal components. To solve the resulting optimization problem, a dynamic programming based method is derived to approximately evaluate the proximal operator associated with the proposed regularizer. Both quantitative and qualitative experiments illustrate that our method is comparable to the top-performing algorithms.