A digital hologram is typically a two-dimensional (2D) complex image that is used to record a three-dimensional (3D) object scene. The development of digital recording has allowed the 3D content of a hologram, especially fluorescent microscopic 3D objects that are captured in real-time, to be processed or even analyzed. The first step of most applications is to segment the hologram into the isolated 3D objects of interest. However, it is challenge to directly separate these objects from the scene because there are full of the diffracted waves of the 3D objects in the measured hologram that is contaminated with noise. This thesis reports a novel digital-hologram segmentation method that can directly divide and extract multiple 3D objects of interest from the hologram through a projected plane during only one segmentation process, while other existing slice-based methods perform multiple segmentation processes on more sectional images for the objects to preserve their volume. Technically, the proposed method is based on the VDP (Virtual-Diffraction-Plane) framework, and an intelligent image segmentation system using two data-mining techniques that are composed of VQ (Vector Quantization) and DBC (Density-Based Clustering). The three elements of the VDP framework, VQ and DBC can benefit each other during the entire segmentation process. To begin with, the VDP framework is applied to project a given hologram onto the VDP that is near the object scene. This lowers the degree of diffraction in the hologram and then the intelligent image segmentation system is applied to the VDP as follows. An improved VQ method is employed to separate different foreground regions from the background region. This VQ segmentation results can resist the noise contaminated in the measured hologram and can be projected onto a small binary image entitled “indices-image”, that helps naturally accelerate the following labeling process for further extracting the 3D objects of interest. A novel DBC-based labeling method is proposed to efficiently label the unique foreground objects in the indices-image in one pass. This labeling method can also obtain rich information of the objects, such as the object’s ID, boundary and size simultaneously. The rich information is very useful to allow users to select the objects of interest in a user-defined manner, and to accurately preserve the depth information of only the selected 3D objects. Finally, an adaptive segmentation mask is obtained for extracting the selected objects, and the extracted 3D objects on the VDP are subsequently reverted to an output hologram. Thorough experimental tests have demonstrated that the proposed scheme is able to effectively divide and extract the 3D microscopic objects of interest from the measured hologram. Good visual qualities of the experimental segmentation results have been achieved. Furthermore, the entire hologram segmentation process can be accelerated up to a video rate with the help of graphics processing unit and a novel method of “Skip-Dimension”. In practical applications, the proposed method is particularly useful for dividing and extracting multiple microscopic 3D objects of interest, potentially like biological specimens, from the measured hologram in real-time