Light fields enable to capture both the intensity and angular information of light rays emanating from objects in a given visual scene. This combined information allows the estimation of dense depth maps, using different methods currently available in related literature. One of such methods is based on the structure tensor, which globally yields acceptable results but fails to find coherent depth values in occlusion areas. In this paper we propose a method to obtain an accurate depth representation of object boundaries that are located in such occlusion areas. This is based on the re- construction of an edge map comprised of depth transitions while avoiding texture gradients without depth variation. This reconstructed edge map representation is then used to highly enhance the quality of disparity maps obtained from the structure tensor. The results show a significant reduction of the mean square error (MSE) and Badpix for a wide variety of light fields. Informal subjective analysis of the obtained disparity maps also reveals a significant reduction in artifacts near the object boundary regions.