EMF- MHCRF: Enhanced Median Filter (EMF) Based Noise Removal and Multilayer Hidden Conditional Random Field (MHCRF) Model for Dense Depth Map Reconstruction

Depth map reconstruction also named as disparity estimation and it becomes very difficult task in
computer vision three dimensional (3D) tasks which has been studied used for more than three decades.
Present high quality depth sensors proficient of creating dense depth maps are costly, noise and bulky, at
the same time as dense low-cost sensors be able to simply consistently generate sparse depth
measurements. In this work, propose an Enhanced Median Filter (EMF) for noise removal of images.
Since the preprocessing stage is an important and initial step in the depth map construction step, it
retaining the important information of patches or frames. And the major issue in processing
stereophotogrammetry images is the presence of noise which presents as bright dots or dust particles more
than the image. So EMF is proposed to remove impulse noise from stereo images. EMF is proposed to
remove the noisy pixel from the original pixel; here the noise is removed depending on the threshold value
computed from genetic operations. Secondly propose a novel Multilayer Hidden Conditional Random
Field (MHCRF) model to restructure a dense depth map of a target scene known the sparse depth
measurements and related to photographic measurements computed from stereo photogrammetric systems.
This MHCRF model assured global optimum in the modeling of the temporal action dependencies
following the Hidden Markov Model (HMM) stage. In MHCRF model, dense depth map is estimated by
formulating the as a Maximum A Posteriori (MAP) inference problem wherever efficiency previous is
assumed. The any middle representation, depth is computed directly from the Middlebury stereo vision
data for ground truth, it has been also deal with any number of cameras. Shows potential experimental
results demonstrate the ability of this EMF- MHCRF model and compared to MCRF, CRF models.