Due to the low accuracy of the existing image dehazing methods with prior, an image dehazing method based on a learning framework with sparse coefficient matching is proposed. Firstly, the solution of the hazy degradation model is transformed to sparse coefficient matching with the database from the view of image restoration. Then, to improve the visual effect of the result, a feedback iteration is quantified by the enhancement of the contrast in highlighted areas from the view of image enhancement. Experiments demonstrate that the proposed method can remove effectively haze as well as provide a good local detail, and it has good generality.