This paper proposes an adaptive compressive sensing reconstruction method which provides a higher recovered image quality. Based on an initial compressive sampling reconstruction at a given sampling rate, the visually salient regions of the image that are more conspicuous to the human visual system are extracted using a classical graph-based method. The target acquisition subrate is further adaptively allocated among these regions, such that the new acquisition will favor the interest areas. The measurements produced by this adaptive method are fully compatible with the existing sparse reconstruction algorithms, which favors the utilization of the proposed scheme. Simulation results show that the saliency-based compressive sensing recovery method outperforms the conventional sparse reconstruction algorithms in terms of image quality at the same target sampling ratio with a smaller increment in the complexity.