A light-trapping image sensor includes a pixel array and a lens array. The pixel array is formed in and on a semiconductor substrate and including photosensitive pixels each including a reflective material forming a cavity around a portion of semiconductor material to at least partly trap light that has entered the cavity. The cavity has a ceiling at a light-receiving surface of the semiconductor substrate, and the ceiling forms an aperture for receiving the light into the cavity. The lens array is disposed on the pixel array. Each lens of the lens array is aligned to the aperture of a respective cavity to focus the light into the cavity through the aperture.