The basic principle of digital refocusing is analyzed and different depth of field corresponds to different refocusing results. The traditional gray gradient evaluation function introduces too much gradient information in light field photography systems, which affects the sensitivity of the evaluation function. In order to solve the defect, an improved image sharpness evaluation algorithm is proposed. A local area in the image is evaluated by the method. Firstly, image gray gradient energy function named Tenengrad function is used to extract high-frequency information to assess the sharpness of the image. And then, a threshold is added to distinguish the edge and non-edge points to reduce the influence on the evaluation function from the area with low image contrast. Comparing with other traditional gradient evaluation algorithms, the advantages of this algorithm is to obtain higher recognition accuracy. Experimental results show that this algorithm has better sensitivity comparing with traditional gradient algorithm through MATLAB platform simulation.