We propose two computational techniques for extracting object cross-sectional shape information from digitized in-line Fresnel holograms that do not require prior knowledge of object depths but recover relative depth information along the way. The first algorithm is applicable to hologram segments involving a single particle only. It is based on estimating Fourier transform magnitude and phase of the particle from those of the hologram segment. The second algorithm conducts a joint inverse filtering and depth search procedure so as to minimize (or maximize) a binariness (or a concentration) criterion defined over the output object function. It is applicable to multiple-particle, multiexposure holograms as well. The proposed techniques are illustrated on synthesized holograms and their practical limitations are discussed.