A method includes fabricating an image sensing element in a substrate. A plurality of inter-metal dielectric (IMD) layers are formed over the substrate. Each IMD layer includes a metal layer and a dielectric layer. A planar top surface of a top IMD layer of the plurality of IMD layers is planarized. A portion of the top IMD layer is then removed to transform a region of the planar top surface to a curved recess. A lens is formed on the top IMD layer and in the curved recess. A color filter layer is disposed over the lens and the image sensing element.