About Project
Additional dimension in two-dimensional images to perceive them as three-dimensional is 'DEPTH' dimension. The process of producing illusion of depth in two-dimensional image by means of stereopsis for binocular vision is called Stereoscopy. Stereoscopy creates stereo-pair of a 2D image. A Stereo-pair of a 2D image contains two views of that image side by side. One of the views is intended for the left eye and the other for the right eye.
A technique to create illusion of depth is to create depth maps. The most important and difficult problem in converting 2D to 3D is how to produce or estimate the depth map using a single view of an image. In a traditional method of creating depth maps, it was a manual process which completely relying on "depth artists". A depth map was created by measuring distance of each pixel in an image from the camera. In an automatic approach of creating depth maps, the process consists of two steps:
To achieve the automatic approach using various computer algorithms, various methods have been developed that estimate shape from shading and structure from motion or depth from defocus. There is limitation to this approach that these methods do not work well for arbitrary scenes. In recent approaches, machine-learning based techniques have been used and proposed to predict the right view from left view of a single monocular image.
Here, we developed a machine-learning method using Deep Convolutional Neural Network to predict right frames from left frame to generate stereo pairs.
Additional dimension in two-dimensional images to perceive them as three-dimensional is 'DEPTH' dimension. The process of producing illusion of depth in two-dimensional image by means of stereopsis for binocular vision is called Stereoscopy. Stereoscopy creates stereo-pair of a 2D image. A Stereo-pair of a 2D image contains two views of that image side by side. One of the views is intended for the left eye and the other for the right eye.
A technique to create illusion of depth is to create depth maps. The most important and difficult problem in converting 2D to 3D is how to produce or estimate the depth map using a single view of an image. In a traditional method of creating depth maps, it was a manual process which completely relying on "depth artists". A depth map was created by measuring distance of each pixel in an image from the camera. In an automatic approach of creating depth maps, the process consists of two steps:
- Estimate depth map from input view
- Combine depth and input view to generate missing view of
To achieve the automatic approach using various computer algorithms, various methods have been developed that estimate shape from shading and structure from motion or depth from defocus. There is limitation to this approach that these methods do not work well for arbitrary scenes. In recent approaches, machine-learning based techniques have been used and proposed to predict the right view from left view of a single monocular image.
Here, we developed a machine-learning method using Deep Convolutional Neural Network to predict right frames from left frame to generate stereo pairs.